Choice of Host Cell Line Is Essential for the Functional Glycosylation of the Fc Region of Human IgG1 Inhibitors of Influenza B Viruses

Key Points Choice of cell line is critical for functional glycosylation of human IgG1-Fc. Sialylated IgG1-Fc monomers block influenza B only when made by CHO-K1 cells.

I n therapeutic approaches in which the Fc of human IgG1 is critically important, receptor binding and functional properties of the Fc are lost after deglycosylation or removal of the Asn-297 N-linked glycosylation attachment site located in the body of the Fc (1)(2)(3). More detailed studies into the types of sugars involved in this functionality have shown enhanced FcgRIIIA binding and Ab-dependent cellular cytotoxicity of IgG1 in the absence of fucose (4,5); enhanced FcgRIIIA binding but rapid clearance from the circulation of IgG1 enriched for oligomannose structures (6)(7)(8); and improved solubility, anti-inflammatory activity, thermal stability, and circulatory half-life of terminally sialylated glycans from IgG1 (9-13).
These findings have generated an incentive to modify the existing IgG1 glycans attached to Asn-297, either by glycoengineering/ chemoenzymatic means (12,14), by mutagenesis programs on the Fc protein backbone that disrupt the protein-Asn-297-carbohydrate interface (15), or by expression in glycosidase-adapted transgenic cell lines (reviewed in Ref. 16). For example, the Food and Drug Administration-approved humanized Ab mogamulizumab, which is used to treat lymphoma and is manufactured in Chinese hamster ovary (CHO) cell lines in which the a (1-6)-fucosyltransferase (FUT8) gene is removed, results in an afucosylated IgG1 with enhanced FcgRIIIA-dependent tumor cell killing by Ab-dependent cellular cytotoxicity (17). Although similar approaches have yielded enhanced sialylation of IgG, with zero to moderate improvements in binding to FcgRs (12,15,18,19), these have not led to significant enhancements in binding to inhibitory glycan receptors that are important in controlling unwanted inflammation (19,20), a finding we and others have attributed to the buried location of the Asn-297-attached glycan within the Fc (21,22).
Less is known about glycosylation of the receptors to which the IgG1-Fc binds. Although recent studies implicate an active role for Fcg receptor-associated carbohydrates in fine-tuning Ab-receptor interactions (23,24), aglycosylated FcgRs expressed in Escherichia coli still retain the ability to bind IgG (25)(26)(27), bringing into question the exact role of Fcg receptor glycosylation to IgG1-Fc binding. Similar arguments apply for most of the glycan receptors. For example, DC-SIGN and DC-SIGNR when expressed by E. coli still bind Man 9 GlcNAc 2 oligosaccharides with high affinity (28). The N-linked glycans on glycan receptors are mostly not located near the CRD binding sites but are involved in cismediated clustering of receptors within the plasma membrane. For example, refolded recombinant E. coli-expressed Siglec-5 is sufficient for binding sialylated carbohydrates (29).
We took an alternative approach to enhancing the sialic acid content of the Fc of IgG1 (30,31) by adding the 18-aa tailpiece from IgM to the C terminus of the IgG1 Fc, into which a cysteine-to-alanine substitution is made at Cys-575, and including an extra N-glycosylation site to the N terminus at position Asn-221. The tailpiece also contains an N-glycosylation site at Asn-563. This site is known to contain a high proportion of oligomannose glycans that may explain earlier observations of enhanced binding of IgM to mannose-dependent receptors (30)(31)(32). When expressed in CHO-K1 cells, these molecules displayed enhanced binding to the low-affinity Fcg receptors (FcgRIIIA and FcgRIIB), and to multiple glycan receptors that control excessive inflammation by IVIG (30,31,33). Two such hypersialylated molecules (D221N/C575A and D221N/C309L/N297A/C575A) also bound recombinant hemagglutinin (HA) from influenza A and B viruses and disrupted influenza A-mediated agglutination of human erythrocytes (31).
CHO cell-based systems remain, by far, the most common mammalian cell line used by the pharmaceutical industry; 84% of products are produced in this cell system, and the remaining approved Abs are produced in either NS0 or Sp2/0 cells (34). Although CHO cells account for the largest number of Food and Drug Administration-approved biotherapeutics (34), they do not express a1,2/3/4 fucosyltransferase and b-1-4-N-acetylglucosaminyltransferase III, which are enzymes expressed in human cells (35). Furthermore, humans have active a2,6-sialyltransferase. As such, CHO-derived IgG1 Fcs are only sialylated through a2,3 linkages, whereas both a2,3 and a2,6 linkages can be found on human IgG1 Fc (30,35). Previous detailed N-glycan structural analysis has confirmed that CHO cells produce only a2,3-sialylated N-glycans, whereas human endothelial kidney (HEK) cells produce both a2,3and a2,6-sialylated N-glycans (36,37). Most nonhuman mammalian cell lines can also attach N-glycolylneuraminic acid. Humans do not have an active CMP-N-acetylneuraminic acid hydroxylase and, so, do not attach N-glycolylneuraminic acid, which can elicit immunogenic responses (35), and consequently, nonhuman cell lines are stringently screened to identify clones that produce proteins with desirable glycan profiles (38).
Human cell lines are a promising alternative to nonhuman cell lines, as they possess fully human posttranslational modifications that reduce downstream processing costs and, more importantly, circumvent any risks associated with immunogenicity from nonhuman glycans. However, human cell lines also have significant limitations, including the capacity to produce sialyl-Lewis x , which binds to endothelial selectins in areas of inflammation (39). Although this may potentially be favorable for anti-inflammatory therapies (39), the attached sialyl-Lewis x may also adversely affect the biodistribution and pharmacokinetics of an Fc when used in other clinical contexts, for example, antitumor mAbs. Human cell lines also carry the risk of contamination and forward transmission of human pathogens, in particular viruses, that may explain why CHO-K1 cells are still the preferred cell line used by the pharmaceutical industry (34). These issues led us to compare the functional properties of a panel of Fc mutants generated in CHO-K1 cells with the same set of proteins manufactured by HEK 293-F cells (31).

Production of mutants
The generation of glycan mutants in all combinations has been described previously for the hexa-Fc that contains cysteines at both positions 309 and FIGURE 1. Schematic showing the various HEK 293-F glycan mutants in which Cys-575 is mutated to alanine to create the C575A panel of mutants. Stars indicate the hinge Asn-221, the Cg2 Asn-297, and the tailpiece Asn-563 glycan sites, respectively. 575 (30). To make the new mutants described in Fig. 1 in which Cys-575 was mutated to alanine, PCR overlap extension mutagenesis was used with a pair of internal mismatched primers 59-ACCCTGCTTGCTCAACTCT-39/39-GGCCAGCTAGCTCAGTAGGCGGTGCCAGC-59 for each plasmid vector coding for a designated glycan modification. The parental plasmids used for these new PCR reactions have been described previously (30). The resulting C575A mutants were then further modified to remove Cys-309 using primer pair 59-TCACCGTCTTGCACCAGGACT-39/39-AG-TCCTGGTGCAAGACGGTGA-59 to create the panel of double cysteine knockouts described in Fig. 2. To verify incorporation of the desired mutation and to check for PCR-induced errors, the open reading frames of the new mutants were sequenced on both strands using previously described flanking primers (30). CHO-K1 cells (European Collection of Cell Cultures) were stably transfected with plasmids using FuGENE (Promega), stable cell lines were created, and Fc-secreting clones were expanded and proteins purified as previously described (22,40). HEK 293-F cells were transiently transfected using the FreeStyle MAX 293 Expression System (Life Technologies) and proteins purified as for CHO-K1 cells.

Size analysis using size exclusion HPLC
An SEC 3000 (300 3 7.8 mm) Column (Beckman Coulter) was set up on a Dionex ICS 3000 HPLC System and pre-equilibrated with 0.2-mm-filtered PBS. Protein samples at concentrations ranging from 0.5 to 1 mg/ml were placed in a precooled autosampler at 4˚C, and 10 ml of each was sequentially injected onto the column. Each sample was run for 1.5 column volumes in PBS at a flow rate of 0.25 ml/min. Elution was monitored at 280 and 214 nm. The column was calibrated by running standard proteins (thyroglobulin, bovine IgG, OVA, myoglobin, and cyanocobalamin; Bio-Rad Laboratories) under the same conditions.
Binding of the secondary detecting F(ab9) 2 and levels of Ag coating were checked using either anti-human Fc (Fig. 3A) or anti-histidine ( Fig. 3B) Abs, respectively, by direct ELISAs to every mutant and/or receptor to ensure potential biases in the detection of binding of different mutants to different receptors could be considered. Because some of the receptors used are not histidine tagged (e.g., DC-SIGN and MBL) or were poorly recognized by the detecting anti-histidine mAb (e.g., dectin-1, dectin-2, clec-4a, and clec-4d), we also checked protein concentrations prior to coating to plates by SDS-PAGE (Fig. 3C).

Hemagglutination inhibition assay
Native influenza B Hong Kong 5/72 was obtained from Meridian Life Science. To determine the optimal virus-to-erythrocyte ratio, 2-fold virus stock dilutions were prepared in U-shaped 96-well plates (Thermo Fisher Scientific). The same volume of a 1% human O + RBC suspension (Innovative Research) was added to each well and incubated at room temperature for 1 h until erythrocyte pellets had formed. After quantifying the optimal virus-to-erythrocyte concentration (4 HA units), serial 2-fold dilutions of Fc, control IVIG (Gammagard; Baxter Healthcare), and polyclonal goat anti-influenza B (Bio-Rad Laboratories) were prepared, all starting at a concentration of 2 mM, and mixed with 50 ml of the optimal virus dilution. After 30 min incubation at room temperature, 50 ml of the human erythrocyte suspension was added to all wells, and plates were incubated at room temperature for 1 h, after which erythrocyte pellets could be observed in the positive controls and positive samples.

Binding to FcgRs by biacore
Binding to FcgRs was carried out using a Biacore T200 biosensor (GE Healthcare). Recombinantly expressed FcgRS (R&D Systems or Sino Biological) were captured via their histidine tags onto CM5 chips precoupled with ∼9000 reflective units anti-His Ab (GE Healthcare) using standard amine chemistry. Fc mutants were injected over captured receptors at a flow rate of 20 ml/min, and association and dissociation monitored over indicated time scales before regeneration with two injections of glycine (pH 1.5) and recalibration of the sensor surface with running buffer (10 mM HEPES, 150 mM NaCl [pH 7]). Assays were visualized with Biacore T200 evaluation software v2.0.1. N-glycomic analysis N-glycomic analysis was based on previous developed protocol with some modifications (41). Briefly, the N-glycans from 50 mg of each sample was released by incubation with New England Biolabs Rapid PNGase F and isolated from peptides using Sep-Pak C18 cartridges (Waters). The released N-glycans were permethylated prior to MALDI MS analysis. Data were acquired using a 4800 MALDI-TOF/TOF mass spectrometer (Applied Biosystems) in the positive ion mode. The data were analyzed using Data Explorer (Applied Biosystems) and GlycoWorkbench (42). The proposed assignments for the selected peaks were based on composition together with knowledge of biosynthetic pathways.

Results
Disulfide bonding and glycosylation influence the multimerization states of Fc mutants expressed by HEK 293-F cells Two panels of glycosylation-and cysteine-deficient mutants previously expressed by CHO-K1 cells were generated in HEK 293-F cells (Figs. 1-3). As observed with CHO-K1 cells, the HEK 293-F cells were capable of making all the mutants to high yields (∼30 mg/l from our limited tissue culture facilities) with the exception of the N297A/N563A/C575A mutant, for which we were unable to generate sufficient protein for further work. Generally, all the mutants migrated on SDS-PAGE with the expected molecular weights for their glycosylation or disulfide bonding states (Fig. 4) and as previously described for the same mutants expressed by CHO-K1 cells (30,31).
In an earlier study with CHO-K1 cells, we demonstrated that a proportion of molecules in which the tailpiece Asn-563 glycan was substituted for alanine ran as multimers in solution when examined by size exclusion HPLC (31). The loss of the bulky Asn-563 glycan exposes hydrophobic amino acid residues in the tailpiece that facilitate noncovalent interactions in solution. Such N563Adependent multimerization was also observed with mutants expressed by HEK 293-F cells, although the proportion of multimers to monomers (with the notable exception of the N563A/ C575A mutant) was generally lower when mutants were made by this cell line (Fig. 5). Clearly, the choice of cell line and, consequently, the types of posttranslational modifications dramatically impact on the biophysical properties of these molecules in solution.

Fc glycan mutants expressed by HEK 293-F cells show differences in binding to glycan receptors when compared with CHO-K1 proteins
To determine the impact of the cell line on receptor binding by the two panels of Fc mutants, we investigated their interaction with soluble recombinant glycan receptors by ELISA (Fig. 6, Supplemental Fig. 1). For most of the Fc mutants, including hexa-Fc, C575A, N297A/ C575A, D221N/N297A/N563A/C575A, C309L/C575A, D221N/ C309L/C575A, D221N/C309L/N297A/C575A, and D221N/C309L/ N297A/N563A/C575A, expression in HEK 293-F cells reduced the binding to glycan receptors when compared with equivalent molecules expressed in CHO-K1 cells (Fig. 6). However, two Fc mutants (D221N/N563A/C575A and D221N/C309L/N573A/ C575A) were notable for their enhanced and nonspecific binding to all the glycan receptors investigated when expressed in HEK 293-F cells. Given that both mutants multimerize, although not as efficiently as their equivalents made in CHO-K1 cells (Fig. 5), we attribute this enhanced nonspecific glycan receptor binding not only to increased avidity effects, but also to fine differences in the attached glycan structures (Supplemental Figs. 3, 4). Therefore, the choice of cell line can dramatically impact on the ability of individual Fc mutants to interact with glycan receptors.  tested the impact of cell line on binding to the classical human FcgRs (Fig. 7, Supplemental Fig. 2).
To examine the interaction with human FcgRIIA (Arg-167) and FcgRIIIB in more detail, we tested binding of two of these mutants (C309L/N563A/C575A and D221N/C309L/N563A/C575A) to FcgRIIIB and FcgRIIA (Arg-167), respectively, by surface plasmon resonance analysis (Fig. 8, Tables I, II). Both mutants displayed slower apparent off rates compared with the control IgG1-Fc monomer, consistent with avidity effects either through binding to multiple immobilized FcgR molecules or rebinding effects (Fig. 8). Therefore, the choice of cell line impacts on the ability of individual Fc mutants to interact with FcgRs and, in particular, FcgRIIA (Arg-167) and FcgRIIIB.

Fc glycan mutants expressed in HEK 293-F cells have more complex glycosylation profiles than the equivalent mutants expressed in CHO-K1 cells
The structure of the N-glycans on the Fc of IgG Abs has been shown to influence multiple receptor interactions (3,43,44). Unlike the relatively simple glycosylation of the Fc mutants previously described for CHO cells (30,31), HEK cells are capable of producing more complex N-glycan structures on their glycoproteins (45).
We investigated the nature of the N-glycans on the two panels of glycosylation-and cysteine-deficient mutants by MALDI-TOF mass spectrometry-based glycomic analysis (complete dataset for both panels of mutants provided in Supplemental Figs. 3, 4). A core-fucosylated biantennary structure without antennary galactosylation, m/z 1835 (GlcNAc 4 Man 3 Fuc 1 ), is the base peak of spectra from all IgG1-Fc mutants produced by HEK cells (Fig. 10,  Supplemental Figs. 3, 4). An indication of the types of glycans attached to either Asn-221, Asn-297, or Asn-563 could be determined using both the C575A or the C309L/C575A panels of mutants. For example, only N-glycans attached to Asn-297 are available for sampling in either the N563A/C575A or C309L/N563A/C575A, mutants that, therefore, also allow the contribution of disulfide bonding to glycosylation at Asn-297 to be elucidated.
In these mutants, N-glycosylation of Asn-297 is dominated by core-fucosylated biantennary glycans (m/z 1835 and 2040) with varied galactosylation levels (Gal 0-2 GlcNAc 4 Man 3 Fuc 1 ), and a Man 5 GlcNAc 2 (m/z 1579) oligomannose structure is also observed (Fig. 10B). The Asn-563 N-glycans are much more complex and heterogeneous. Abundant truncated structures at m/z 2081 and 2285 have potentially terminal N-acetylglucosamine (GlcNAc) or  N-acetylgalactosamine (GalNAc) (Fig. 10A). Antennal fucosylation and sialylation are also observed on structures that can assemble sialyl-N-acetyllactosamine, sialyl-Lewis x/a, fucosylated LacdiNAc, or sialylated LacdiNAc (GalNAc-GlcNAc), for example, peak m/z 4039 (NeuAc 2 Gal 4 GlcNAc 6 Man 3 Fuc 2 ). The presence of m/z 2674 (GalNAc 2 GlcNAc 4 Man 3 Fuc 3 ) in the N297A/C575A mutant confirms the presence of fucosylated LacdiNAc epitopes on the Asn-563 site. Thus, glycosylation at Asn-563 is different to that seen from CHO-K1 cells that assemble less-diverse structures without antennal fucosylation and, therefore, more terminal sialyl-N-acetyllactosamine (30,31). In agreement with earlier work (32) and irrespective of the cell line used for their manufacture, a high proportion of oligomannose structures can be found attached to Asn-563 (e.g., in the C309L/N297A/ C575A mutant) (Supplemental Fig. 4), although these were not observed in the corresponding N297A/C575A mutant (Fig. 10A). We do not have an obvious explanation for these Asn-563 differences, although the presence of redox-sensitive cysteines are known to modify protein glycosylation in other systems (46).
In summary, these data show that the types of glycans attached to either Asn-221, Asn-297, or Asn-563 are different between cell lines but are not grossly affected by disulfide bonding.
Fc glycan mutants expressed in HEK 293-F cells are less sialylated than the equivalent mutants expressed in CHO-K1 cells Site-specific levels of sialylation were semiquantitatively assessed for both panels of mutants and compared with levels seen in the equivalent mutants expressed in CHO-K1 cells (Fig. 11). Although levels of sialylated glycans attached at positions Asn-297 (the N563A/C575A mutant) and Asn-563 (the N297A/C575A mutant) are similar for both cell lines (Fig. 11), a marked reduction in levels of sialylated glycans at Asn-221 (the D221N/N297A/N563A/ C575A mutant) is observed when this mutant is expressed in HEK cells (2.8% against 81.8% in CHO; Fig. 11). Removal of Asn-297 generally enhanced levels of sialylation at both Asn-221 and Asn-563, irrespective of the cell line or the multimerization state of the proteins (e.g., compare N297A/C575A versus C575A and C309L/N297A/C575A versus C309L/C575A) (Fig. 11).    Table I. Binding was to receptors sourced from R&D Systems (Bio-Techne).
The choice of cell line, therefore, dramatically affects the overall levels of sialylation at individual N-linked attachment sites within the glycan-modified Fc variants.

Asn-221-containing mutants are poor inhibitors of hemagglutination by influenza virus when expressed in HEK 293-F cells
To test if the choice of cell line affected the functionality of the two panels of mutant Fcs, we used the World Health Organization hemagglutination inhibition assay (HIA) to quantify the inhibitory titers for each mutant against an influenza B virus (Fig. 12). As shown previously with an avian influenza A (H1N1) (31), mutants containing Asn-221 hinge-attached glycans, and, in particular, the D221N/C309L/N297A/C575A mutant, prevented hemagglutination by an influenza B virus at concentrations as low as 30 nM, an 8-fold improvement over equimolar IVIG or polyclonal anti-influenza B antisera ( Fig. 12). In stark contrast, the same mutants expressed by HEK 293-F cells were unable to inhibit hemagglutination by either influenza A (data not shown) or influenza B virus (Fig. 12). This shows that the functional potential of individual glycanmodified Fc mutants is dependent on the choice of cell line used for their manufacture.

Discussion
We have shown using CHO-K1 cells that the structure and effector function of human IgG1-Fc can be profoundly altered by the addition or removal of N-linked glycosylation (30,31). For example, we could show that Fc fragments containing complex biantennary glycans attached to both the N-and C-terminal ends of the Fc could inhibit influenza A-mediated agglutination of human erythrocytes (31). The aim of the current study was to reveal possible variation in functional glycosylation related to differences in two host cell lines, CHO-K1 and HEK 293-F, particularly as Abs and Fc fusions are the fastest growing therapeutic class in the pharmaceutical industry (34,47,48). Two intriguing aspects of N-linked glycosylation are relevant to this study. First, the differential binding seen to human glycan (Fig. 6) and Fcg (Fig. 7) receptors between the same mutants expressed by two different cell lines. These differentially manufactured mutants now need to be compared in relevant in vivo disease models in which the Fc is therapeutically useful, given that differential sialic acid linkages, a2,6 and a2,3, are known to impact on the anti-inflammatory properties of the Fc (49,50). Such nuanced glycosylation may also explain why the therapeutic efficacy of molecules generated by different expression systems, and subsequently tested in different animal models, do not always translate to efficacy in human studies (51).  Second, we have studied the exquisite impact of the host cell line on the efficacy of sialylated Fcs to inhibit influenza viruses (Fig. 12). One possible explanation is that overall sialylation levels for all the influenza blocking mutants, in particular the D221N/C309L/ N297A/C575A mutant, are ∼5-fold lower when expressed by HEK 293-F cells (Fig. 11). However, overall level of sialylation is not the only possible explanation for the relative efficacy of the CHO-K1 mutants in inhibiting influenza virus hemagglutination, as the CHO-K1-expressed D221N/C575A mutant also contained ∼5-fold-less sialyation than the D221N/C309L/N297A/C575A mutant made in the same cell line (Fig. 11). This indicates that the fine specificity (e.g., a2,3 versus a2,6 linkages) of these sialylated glycans may also be a contributing factor to their efficacy.
As demonstrated previously for influenza A (31), binding and inhibition of influenza B viruses is stronger with mutants containing Asn-221 and, in particular, by the monomeric mutant D221N/ C309L/N297A/C575A in which the N-and C-terminal sialylated sugars are spaced ∼60Å apart [ Fig. 12 and (31)]. Recent biophysical studies with alternative glycan-decorated scaffolds have shown that ∼1000-fold enhancements over monovalent binding to HA can be achieved with only two sialylated ligands, provided the sugars are arranged 50-100Å apart (52,53).
As observed with Fc-multimerizing mutants from panel 1, no additional benefit with respect to virus neutralization was gained with larger, more complex sialylated structures (Fig. 12, and as seen with the D221N/N297A/N563A/C575A and D221N/N563A/ C575A mutants). The lack of monomers observed for both these mutants when expressed by HEK cells (Fig. 5A) may offer a simpler explanation for why the HEK-expressed proteins did not inhibit influenza viruses in the HIA, yet their equivalents expressed by CHO-K1 cells and that contain a large proportion of monomers did partially inhibit hemagglutination by virus (Fig. 12). As multimers, neither of the sialylated Fc-capping glycans (Asn-221 or Asn-563) may be optimally available for interactions with HA as they are with mutants that uniquely exist as monomers (e.g., D221N/C309L/N297A/C575A) (30,31).  We do not yet know if sialylated Fcs are susceptible to cleavage by the viral neuraminidase (NA). Although a decoy for NA may be a therapeutically attractive strategy (54), we have not observed a direct decay in the HIA after prolonged incubation. This suggests that the high specific avidity of these molecules for HA may reduce their susceptibility to NA, a hypothesis that fits with the relatively low efficiency of NA (k cat = 30-155 s21 ), together with the asymmetric distribution of NA in relation to HA on the surface of filamentous influenza viruses (55)(56)(57).
To be useful compounds when administered intranasally, or as an aerosol, the sialylated Fc needs to outcompete sialylated mucins that viruses use through ratchet-like interactions with HA and NA to migrate to the underlying respiratory epithelium (55). Of the 15 known human mucins in the human lung, only MUC5 has been shown to protect from influenza (58,59). Most sialic acid found on human mucins are O-glycosylated, and where N-linked attachments do occur, these are mostly sialylated via a2,6-linkages (59). Thus, we were surprised that none of the Fc leads inhibited influenza A (H1N1 propagated in hen eggs) or influenza B (Hong Kong 5/72 propagated in Madin-Darby canine kidney cells) agglutination of human O + erythrocytes when manufactured by HEK 293-F cells that attach more human type a2,6-linked sialic acid (Fig. 10).
The apparent importance of a2,3-linked N-glycans to inhibition of both influenza A and B by the CHO-K1 Fc mutants indicates that viruses can evolve away from inhibition by mucus whose predominant O-linked glycans are mostly a2,6-linked. Our working hypothesis is that HEK-expressed compounds may therefore inhibit influenza viruses that circulate in human populations or that are propagated in cell lines that attach more human-like a2,6-linked sialic acid.
Consequently, by careful consideration of the cell line used in their manufacture, new glycan repertoires with desirable binding attributes and functionality can be imparted to the therapeutically attractive Fc molecule.