Expression of recombinant glycoproteins in mammalian cells: towards an integrative approach to structural biology

Abstract

Mammalian cells are rapidly becoming the system of choice for the production of recombinant glycoproteins for structural biology applications. Their use has enabled the structural investigation of a whole new set of targets including large, multi-domain and highly glycosylated eukaryotic cell surface receptors and their supra-molecular assemblies. We summarize the technical advances that have been made in mammalian expression technology and highlight some of the structural insights that have been obtained using these methods. Looking forward, it is clear that mammalian cell expression will provide exciting and unique opportunities for an integrative approach to the structural study of proteins, especially of human origin and medically relevant, by bridging the gap between the purified state and the cellular context.

Introduction

Mammalian cells in culture have long been used in the manufacture of biopharmaceuticals, in particular monoclonal antibodies engineered for human therapy, for example, Herceptin^® and Avastin^® [1]. Processes for protein production in mammalian cells for both pre-clinical and clinical studies are well-established, using both transient and stable cell line expression systems. However, due to perceived difficulties including the need for specialized culture facilities and higher costs compared to microbial alternatives, only relatively recently have mammalian expression systems become routinely used in laboratory-scale production of recombinant proteins for structural biology applications. Although only ~3% of the unique structures in the Protein Data Bank currently come from mammalian cell expression [2], their number has increased by 40% in the last two years (Figure 1a). Considering the recent technological progress, this trend is likely to continue.

Figure 1.

Mammalian expression technology applied to structural biology. (a) Plot of the cumulative total number of chains deposited in the PDB whose expression system was identified as either HEK-293 (Human Embryonic Kidney) or CHO (Chinese hamster ovary) cells by year of deposition. Expression data were parsed from the set of PDB files available from ftp://ftp.wwpdb.org/pub/pdb/data/structures/divided/pdb as of November 2012. Chains were counted rather than PDB entries as expression information is recorded by chain in the PDB. Note: entries marked ‘obsolete’ could not be included, which might have excluded a small number of early 1990s structures. (b) Workflow for the production of recombinant proteins in HEK-293 cells for structural biology applications.

Mammalian expression is particularly beneficial for the production of human and other vertebrate proteins, especially large, multi-domain, cell surface and/or secreted constructs, which require a complex folding machinery and post-translational modifications. For example, over 50% of the proteins encoded by the human genome undergo glycosylation of either serine/threonine (O-linked) or asparagine (N-linked) residues [3]. Such modifications are often essential during the folding process, but typically hamper crystallographic work [4]. Therefore, an important early development was the derivation of mutant Chinese Hamster Ovary (CHO) cell lines deficient in glycan processing which produce samples with defined glycoforms and minimal micro-heterogeneity [5]. One of these CHO glycosylation mutants (CHO LecR 3.2.8.1), in combination with a highly efficient selection system based on glutamine synthetase [6,7], was used to produce the extracellular region of rat CD2 [8] and opened the way for crystallographic analyses of glycoproteins in the early 1990s (Figure 1a). Other developments, such as selenomethionine (SeMet) labeling of proteins expressed in CHO cells for crystallographic phasing by multiple anomalous dispersion methods were soon reported [9,10]. Introduction of large-scale transient expression of Human Embryonic Kidney 293 cells (HEK-293) [11–13] in early/mid 2000s has been associated with an exponential increase in the number of structures of mammalian-expressed proteins (Figure 1a). In this article we review the current status of the use of mammalian cells for sample preparation in structural biology. Examples of how the application of these methods has contributed to obtaining new structural information, in particular novel principles of cell surface receptor organization and signaling, will be discussed.

Enabling technology

A contemporary workflow for mammalian cell production of glycoproteins is shown in Figure 1b. HEK-293 cells have become the host of choice due to their ease of culture in either attached or suspension formats, high transfection efficiency and capacity to express recombinant proteins in large amounts. Large-scale transient transfection of HEK-293 cells has become economically feasible since the discovery that the inexpensive linear cationic polymer, polyethylenimine, is an excellent DNA condensing agent [11,14]. Yields of purified proteins range between 1 and 80 mg/L of culture [13,15,16^•] (unpublished), though for a highly optimized process production of 1 g/L of a recombinant antibody has been reported [17]. In our experience yields of secreted proteins from HEK-293 cells exceed those for the same product expressed in baculovirus infected insect cells [13]. Cells are typically grown attached, in either roller bottles [13] or cell factories [18], or in suspension, in either Erlenmeyer or spinner flasks [19^•], but can be scaled up to multi-litre volumes in bioreactors such as the Wavebag^® system [12]. Automation of the culture processes for both attached and suspension-adapted HEK-293 cells has been successfully implemented in medium/large-sized academic laboratories, thereby reducing manual handling and greatly increasing batch reproducibility and cost-effectiveness [16^•,20].

Three main variants of HEK-293 cells are currently in use for large-scale protein production: first, 293T which expresses the SV40 large-T antigen [12,13,21]; second, 293E which expresses the Epstein-Barr virus nuclear antigen 1 [11,21] and third, 293S-GnTI⁻, deficient in N-acetylglucosaminyl transferase I activity and therefore lacking complex N-glycans [22]. For sustained production of a particular protein, or to increase yields compared to transient expression, stable cell lines may be constructed by co-selection or further streamlined by site-specific integration of the target gene into the host cell genome, at transcriptionally active sites, removing the need to clone out high producers [23^•]. For some targets, notably integral membrane proteins, it is important to control the timing of gene expression, typically by using an inducible system [24^•]; the most widely used one exploits the prokaryotic tetracycline repressor/operator to control expression [25].

Management of N-linked glycosylation

The structure of mammalian N-linked glycan chains, unlike insect equivalents, is highly variable (Figure 2a and b) and management of this heterogeneity is usually essential for structural biology studies [4]. Treatment of HEK-293T cells with small molecule inhibitors such as kifunensine, which targets α-mannosidase I [26]; swainsonine, which blocks α-mannosidase II activity [27]; or the use of cells deficient in N-acetylglucosaminyl tranferase I [22] results in relatively simple and chemically uniform glycans (Figure 2c–e), cleavable using endoglycosidase (endo) H or endo F1 to leave only one GlcNAc residue attached to the N-glycosylation site (Figure 2f) [4]. Ultimately, the strategy chosen for a particular recombinant protein depends on the glycan structure that needs to be preserved. The large number of glycosylation sites typically present on cell surface proteins brings a considerable extra volume and flexibility (Figure 2g–i), therefore enzymatic treatment is often a pre-requisite for crystallization [28]. Recent examples where de-glycosylation was beneficial include the EphA2-ephrin [29^••], netrinG-NGL [30^••] and semaphorin-plexin/neuropilin signaling complexes [31^••,32^••], the Wnt inhibitory factor WIF-1 [33^•], heteromeric kainate receptor amino-terminal domains [34^••] and the HIV GP120 [35^•]. Complete removal of N-linked glycans, achievable via peptide-N-glycosidase F (PNGAse F) treatment, should be avoided as it introduces an Asn to Asp mutation and the Asn-linked GlcNAc may establish contacts with protein surface residues and play an important structural role. It is also important to note that, in some circumstances, removal of glycans is not required, or indeed not desirable, for crystallization, as exemplified by the structures of FcER1α [36^•], human IgG1-Fc [37^•], cytokine receptor–ligand complexes [38^••,39^••] and neurexin 1α [40^•].

Figure 2.

The structure and management of canonical N-linked glycans. (a) Paucimannose glycans typically produced by insect cell lines (e.g. Sf9), difficult to remove enzymatically [4]. (b) Mammalian cell glycans, in contrast, have a larger size and chemical heterogeneity. (c)–(e) Glycosylation inhibitors or mutant cell lines are used to trap specific structures in the biosynthetic pathway. These include Man₉GlcNAc₂, upon kifunensine treatment (c), Man₅GlcNAc₂, in GnTI⁻ cells (d) and hybrid mannose, following treatment with swainsonine (e). (f) All these intermendiates (c)–(e) can be trimmed down to a single GlcNAc by Endo H/F1 treatment. (g) Crystal structure of the human RPTPμ trans-adhesive interaction [28]. (h) Each monomer carries 12 N-glycosylation sites (around one per 100 residues, a typical amount for cell surface receptors). Man₉GlcNAc₂ glycans were modeled at nine sites per monomer, to illustrate their considerable volume and coverage of the protein surface. (i) Successful crystallization could only be achieved upon EndoH glycan trimming [28]. Colour/shape coding of glycan units: GlcNAc, blue square; Man, green circle; Gal, yellow circle; NeuNAc, pink diamond; Fuc, red triangle.

Furthermore, while N-linked glycans may pose a problem for crystallization, their ability to mask relatively extended protein surface patches has been used to restrict protein conformation (for example in the integrin head-piece [41]) or to probe protein–protein interactions. Engineering of ‘glycan wedges’ at structure-guided locations is increasingly employed to functionally validate and manipulate the architecture of oligomeric proteins (for example the interaction interface between GABA_B receptor subunits [42]), receptor–ligand assemblies (e.g. EphA2-ephrin [29^••], netrinG-NGL [30^••]) or predict interaction sites in protein complexes (e.g. WIF1-Wnt3a) [33^•].

Specific amino acid labeling in mammalian cells

Biosynthetic labeling of glycoproteins expressed in mammalian cells with SeMet for crystallographic phasing has been successfully adapted to the HEK-293 transient expression format [13,29^••,43–47]. Furthermore, increased availability of isotope-labeled media has also opened-up mammalian expression to NMR studies (reviewed in [48^•]). Relatively high levels of incorporation (approximately 50% sequence coverage) have been achieved using a medium containing an ¹⁵N amino acid mix (GKLQSTVW) [49]. More recently, over 90% isotope enrichment has been reported [50]. Such developments, combined with the rapid refinement of mammalian in-cell NMR approaches (reviewed in [51]), hold great promise for the structural characterization of suitable mammalian proteins within their cellular milieu. In this context, an atomic resolution NMR description of the complete post-translational maturation process of human superoxide dismutase 1 (SOD1), endogeneously expressed in HEK-293T cells, has been achieved [52^••].

Recent efforts aimed at ‘expanding’ the genetic code led to successful incorporation of non-natural amino acids at specific sites in mammalian cell proteins [53,54]. This involves co-transfection with vectors encoding paired bacterial aminoacyl-tRNA synthetases, engineered to recognize specific non-natural amino acids, and an amber-supressor tRNA that recognizes UAG codons introduced into the target sequence. The technology has already opened up a multitude of exciting applications, including novel possibilities for experimental phasing or isotope-labeling [53,54], probing protein–protein interactions in situ using residues with light-induced cross-linking activity [53,55] and site-specific protein labeling with fluorophores for live cell imaging experiments [56^•–58^•].

Novel structural insights into cell signaling mechanisms

The last three years have witnessed a sharp increase in structural studies that have made use of mammalian cell expression technology. In this short review we cannot do justice to all these investigations, therefore we will highlight examples in which a combination of structural, or structural and cellular, techniques has been employed to yield new concepts in cellular signaling.

Receptor–ligand complexes: specificity determinants and viral interference

A combined approach involving mammalian cell expression and live cell fluorescence microscopy has recently shed light on two key aspects of brain wiring, one of the most puzzling and fascinating processes in developmental biology. Firstly, the comprehensive crystallographic investigation of all possible netrinG-NGL trans-synaptic molecular pairs [30^••] explained how migrating axons connect to specific subdomains on the dendrites of target neurons. This is a remarkable example of molecular patterning on an individual cell surface, very likely of broad significance in biology. Secondly, three groups have independently described crystal structures of semaphorin-plexin complexes [31^••,59^••,60^••], providing unprecedented insights into this prototypical neuronal guidance system. On the extracellular side, binding of dimeric Sema ligands triggers Plexin receptor dimerization (Figure 3a and b). This is facilitated by a co-receptor, neuropilin-1, in the case of Sema3-PlexinA pairs [32^••]. However, a mature signaling assembly appears to be much larger because the intracellular region of Plexin-B1, in complex with a Rho GTPase (Rac1), was found to form a non-crystallographic 3:3 arrangement (Figure 3c and d) [61^•]. This points towards an intriguing honeycomb-like cluster of receptor–ligand complexes that may facilitate the transduction of bi-directional (inside-out and outside-in) signals (Figure 3e). The functional impact of the interactions that stabilize this assembly has been validated in cell collapse assays (Figure 3f and g) using structure-guided mutant constructs [61^•].

Figure 3.

Emerging concepts in cell surface signaling: examples of combined structure/function approaches that have revealed supramolecular receptor organization. Such assemblies are driven by either protein–protein interactions (e.g. the semaphoring/plexin system) or by non-protein ligands (e.g. type IIa RPTPs interactions with HSPGs and CSPGs). (a) Crystal structure of a Sema4D-PlexinB1 ectodomain complex [31^••]. (b) Binding of dimeric Sema4D ligands (pink/red) to the PlexinB1 extracellular region (dark blue) triggers receptor dimerization. Note that non-ligand-dependent PlexinB1 ectodomain oligomerization has also been reported, but its extent or architecture are still unclear [31^••]. (c) Crystal structure of the PlexinB1 intracellular region in complex with Rac1 [61^•]. (d) This interaction leads to the formation of a 3:3 receptor–ligand assembly. (e) Schematic representation of the putative impact of simultaneous Sema4D and Rac1 binding to PlexinB1. The combined 2-fold and 3-fold interactions may lead to a hexagonal ‘honeycomb’ arrangement (shown in part here) that facilitates bi-directional signaling. (f)–(g) COS7 cell collapse assays were used to validate crystallographic interfaces using structure-guided mutagenesis. Scalebar: 40 μm. (h) Electrostatic potential representation (±5kT/e) of the glycosaminoglycan (GAG)-binding region of human RPTPσ [45^••]. (i) A basic residue cluster, conserved in all family members, interacts with sulphated sugars (shown here is human LAR in complex with sucrose-octasulphate). (j)–(k) The polymeric nature of heparan sulphate (HS) triggers receptor clustering. (l)–(m) Importantly, the distribution of sulphate groups along the HS chains is not even: 12–14 sulphate-rich units (red), flanked by intermediate sulphation regions (yellow) are separated by long low-sulphation portions. This imposes an uneven distribution of receptors on the cell surface (~four receptors per cluster), which is essential to promote neuronal motility [45^••]. (n) In contrast, the distribution of sulphate groups on chondroitin sulphate molecules prevents formation of RPTPσ clusters and inhibits cell motility [45^••].

Remarkable progress has also been made in the study of signaling pathways that drive the development of the mammalian immune system, in particular those centered around the colony stimulating factor-1 receptor (CSF-1R or FMS), a class III receptor protein tyrosine kinase. A combination of structural methods, as well as mammalian expression techniques, has revealed surprisingly conserved principles of organization in the architecture of CSF-1R extracellular complexes with their structurally and functionally distinct cytokine ligands, CSF-1 and IL-34 [62^••,63^••].

Two of the pathways discussed above are also targeted during viral infections, and recent structural data provided important new insights into the molecular mechanisms of viral mimicry. BARF1, a secreted Epstein-Barr virus-encoded decoy receptor, acts as a molecular ‘trap’ for CSF-1 ligands that are stabilized in an inactive conformation upon attachment to its toroid-shaped hexamer [39^••]. By contrast, Vaccinia (smallpox) virus and Alcelaphine herperviruses secrete semaphorin homologues (A39R and AHVsema) that bind to and activate the PlexinC1 receptor. Importantly, while the overall arrangement of a dimeric A39R in complex to PlexinC1 is highly similar to that observed in the Sema7A-PlexinC1 complex, the A39R has evolved a smaller yet more efficient receptor interface, leading to a 20-fold tighter interaction compared to the endogenous Sema7A ligand [59^••].

Higher order organization of cell surface receptors

The rapid expansion of structural information has recently set the foundations for a fundamentally novel concept of receptor organization at the cell surface. Increasingly, instead of appearing ‘isolated’ or simply dimerized upon ligand binding, it is now clear that many such proteins have the potential to form supra-molecular clusters or arrays, driven or stabilized by cis (same cell) and trans (across the inter-cellular space) interactions. The initial observation of ‘molecular zippers’ in the crystal packing of C-cadherin [64], a homophilic cell adhesion molecule, gained further support from cryo-electron tomographic analysis of desmosomal cadherins in human skin samples [65]. Importantly, the C-cadherin arrangement has been confirmed by two additional crystal structures (for E-cadherin and N-cadherin ectodomains), and also visualized outside the constraints of crystalline packing by cryo-electron microscopy of cadherin-coated liposomes [66^••]. The physiological relevance of these assemblies was demonstrated by mutational analysis of protein–protein interfaces in cellular assays [66^••]. Furthermore, the wealth of experimental data that emerged from the extensive, multi-disciplinary, analysis of cadherin interactions has now been rationalized within a robust theoretical framework [67^••].

While the concept of an array-like arrangement might have been expected at molecularly homogeneous cell contacts, a similar scenario is currently emerging from the study of heterophilic inter-cellular interactions. For example, a complex between neurexin-1β and its adhesive partner, neuroligin-1, has also revealed crystallographic packing consistent with a trans 2D-array [68^•]. Importantly, this architecture was supported by correlative fluorescence-electron microscopy observations at points of contacts between cells expressing recombinant neurexin-1β and neuroligin-1, respectively, on their surface [68^•]. The relevance of this molecular architecture, even over a short range, within the physiological environment of a neuronal synaptic cleft remains to be demonstrated. If genuine, it will likely have a key impact on synapse formation, function and remodelling.

The importance of supra-molecular interactions is better understood in the case of receptor enzymes. For example, the multi-domain, conformationally flexible, ectodomain of erythropoietin-producing hepatoma A2 (EphA2) kinase tends to cluster in linear arrays. Certain relative arrangements within such arrays are signaling-compatible and appear to be stabilized by interactions with trans ligands, the ephrins [29^••,69^••]. This architecture facilitates signal amplification and its discovery solved a long-standing puzzle in the field: while Eph signal triggering is strictly dependent on direct contact with a small area of pre-clustered ephrins, the regions of receptor activation vastly exceed these points of contact [70].

Receptor clustering may also be modulated by non-protein ligands, as recently demonstrated by crystallographic and functional analysis of type IIa receptor protein tyrosine phosphatases, such as RPTPσ (Figure 3h–n) [45^••]. The linear glycosaminoglycan chains of extracellular or cell surface attached proteoglycans, a ubiquitous presence around eukaryotic cells, play fundamental roles in signaling regulation. In the case of RPTPσ, the ratio between heparan and chondroitin sulphate proteoglycans it encounters determines whether receptors cluster or not (Figure 3m and n), which in turn represents a key control of cellular motility as demonstrated in the case of neuronal extension [45^••]. This mechanism is likely to be widespread in cell signaling systems, including the semaphorin-plexin discussed above, to which it adds an additional level of complexity [71].

Novel insights into antibody complexes

The interaction between IgG and Fc receptors is modulated by the structure of the IgG N-linked glycans. Antibodies lacking core fucosylation show a significant increase in FcRγIIIa binding, which results in improved receptor-mediated effector activity [72^••]. This is important for the function of therapeutic antibodies, whose efficacy has been improved by glyco-engineering. Structural studies of the complex between glycosylated IgG-Fc and FcRγIIIa revealed a unique type of carbohydrate interface, involving the N-glycans of both proteins, which is only formed in the absence of core fucosylation (Figure 4) [72^••]. The use of mammalian cells to produce defined glycoforms of both proteins was essential for this discovery.

Figure 4.

The impact of N-linked glycans in modulating protein complexes. (a) Crystal structure of the glycosylated complex between the IgG1-Fc region and its FcγRIIIa receptor, essential for antibody-mediated cellular cytotoxicity [72^••]. (b) Antibodies lacking core fucosylation show a large increase in affinity for FcγRIIIa, due to additional glycan–glycan and glycan–protein interactions (putative hydrogen bonds are shown as dashed lines). This leads to an improved receptor-mediated effector function and forms the basis for a next generation of therapeutics, glycoengineered antibodies. (c) In the complex structure containing fucosylated IgG1-Fc, such contacts are very limited, explaining the decreased affinity for the FcγRIIIa receptor. Fuc: fucose attached to the Asn297-linked GlcNAc in IgG1-Fc. Colour coding of carbon atoms in the glycan units (sphere and stick representation) in panels (b) and (c) is as described in Figure 2.

More generally, mammalian cells are increasingly being used for the production of antibodies and Fab fragments for structural studies. Novel antigen-binding specificities identified by phage-display technology have been converted into recombinant IgGs from which Fab fragments have been derived by papain cleavage for structural analysis of antibody–antigen complexes [73,74]. Alternatively, recombinant Fab fragments have been produced directly in mammalian cells for co-crystallization studies [75,76,77^••]. Mapping the epitopes recognized by broadly neutralizing antibodies using X-ray crystallography became a powerful guide for the development of ‘universal’ vaccines against viruses including HIV-1, influenza and hepatitis C [73,77^••–80^••]. Similarly, the crystallographic characterization of a human EphrinA2 receptor construct in complex with an agonistic antibody has implications for the development of novel anti-cancer therapies [81^•].

Conclusion and future directions

Expression of proteins in mammalian cells for structural biology applications is set to increase over the next few years. Technical developments have lowered the barriers of cost and time, and increased the throughput of protein production. Strategies for incorporation of selenomethionine and amino acid isotopes, extending the genetic code with non-natural amino acids and controlling glycosylation heterogeneity have all been reliably established. The combined structural and cellular investigation of the same protein constructs, afforded by the use of mammalian cells, is likely to expand and reveal novel mechanistic insights into molecular pathways relevant to human health. The power of such an interdisciplinary approach will likely allow a direct observation of complex processes, such as signal transduction across cellular membranes, in genuine molecular detail: current efforts directed at understanding epidermal growth factor receptor signaling represent a spectacular example [82^••–84^••]. Further technological developments in super-resolution microscopy [85,86^••], correlative fluorescence-electron microscopy [87,88^••] and X-ray microscopy [89^•,90^•] will offer more exciting opportunities for a modern, integrative approach to the structural study of recombinant proteins across the resolution scale, bridging the gap between the purified state and the quasi-physiological, cellular context. The successful achievement of such goals will vitally depend on a broad and seamless accessibility to cutting-edge technologies. Therefore, efforts aimed and coordinating and integrating infrastructure, exemplified by the European INSTRUCT (http://www.structuralbiology.eu/), BioStruct-X (http://www.biostruct-x.eu/) or Bio-NMR (http://www.bionmr.net/) projects, are likely to expand and play a crucial role.