An engineered eukaryotic protein glycosylation pathway in Escherichia coli

Abstract

We performed bottom-up engineering of a synthetic pathway in E. coli for the production of eukaryotic trimannosyl chitobiose glycans and the transfer of these glycans to specific asparagine residues in target proteins. Glycan biosynthesis was enabled by four eukaryotic glycosyltransferases, including the yeast uridine diphosphate-N-acetylglucosamine transferases Alg13 and Alg14 and the mannosyltransferases Alg1 and Alg2. By including the bacterial oligosaccharyltransferase PglB from C. jejuni, glycans were successfully transferred to eukaryotic proteins.

N-linked protein glycosylation is the most common post-translational modification in eukaryotes, affecting many important protein properties¹. N-linked glycosylation is not limited to eukaryotes, however, as bona fide N-linked glycosylation pathways are found in proteobacteria² and can be transferred to E. coli³. There are several notable differences between bacterial and eukaryotic N-glycosylation systems. First, bacteria assemble oligosaccharides on undecaprenyl pyrophosphate (Und-PP) in the cytoplasmic membrane whereas eukaryotes use dolichyl pyrophosphate (Dol-PP) in the ER membrane. Second, the N-X-S/T consensus sequence for N-glycosylation in eukaryotes appears to be extended to D/E-X₋₁-N-X₊₁-S/T (X₋₁, X₊₁ ≠ P) in bacteria⁴ with few exceptions^5,6. Third, bacterial N-glycans are completely distinct from any known eukaryotic glycan⁷. As a result, glycoproteins derived from existing bacterial expression systems are restricted to bioconjugate vaccines^8,9 or glycoproteins that require extensive in vitro modification¹⁰. The construction of a eukaryotic glycosylation pathway in E. coli that generates human-like N-glycans remains an elusive challenge despite much speculation^7,11.

To address this challenge, we focused on engineering E. coli to produce mannose₃-N-acetylglucosamine₂ (Man₃GlcNAc₂) glycans. We chose Man₃GlcNAc₂ because it is: (i) the core structure common to all human N-glycans; (ii) the predominant N-glycan produced by baculovirus-insect cells, carrot root plant cells, and Tetrahymena thermophila, all of which yield glycans that are fit for pre-clinical and clinical products; and (iii) the minimal glycan required for a therapeutic glycoprotein currently on the market¹². To generate Man₃GlcNAc₂ on the cytoplasmic membrane of E. coli, a synthetic pathway was designed (Fig. 1).

Figure 1. Engineering eukaryotic glycan biosynthesis in E. coli.

Schematic of the synthetic pathway for synthesis of a trimannosyl core glycan and transfer to acceptor sites in target proteins. Enzyme names in black are native to E. coli; enzyme names in red are heterologous. See text for details. Glycan in brackets to the right of open arrows depicts terminally sialylated structure common in human glycoproteins, but outside the scope of this study.

The first step in this pathway involved WecA, an endogenous glycosyltransferase (GTase) that transfers GlcNAc-1-phosphate to undecaprenyl phosphate (Und-P). To extend the glycan, several heterologous GTases from Saccharomyces cerevisiae were selected because these can be solubly expressed in E. coli^13–15 and in some cases the expressed enzymes are active in vitro^13,14. Specifically, for addition of the second GlcNAc residue to GlcNAc-PP-Und, we chose the S. cerevisiae β1,4-GlcNAc transferase that is comprised of the Alg13 and Alg14 subunits. In yeast, Alg14 is an integral membrane protein that functions as a membrane anchor to recruit soluble Alg13 to the cytosolic face of the ER membrane¹⁵, where synthesis of GlcNAc₂-PP-Dol occurs. Consistent with their localization in yeast, both Alg13 and Alg14 localized in the membrane fraction of E. coli while Alg13 was also detected in the soluble fraction (Supplementary Fig. 1). For the subsequent steps, we employed S. cerevisiae β1,4-mannosyltransferase Alg1, which specifies the addition of the first mannose to the glycan¹⁴, and the bifunctional mannosyltransferase Alg2, which carries out the addition of both an α1,3- and α1,6-mannose in a branched configuration¹³. Like Alg13/14, both Alg1 and Alg2 localized in the membrane fraction of E. coli (Supplementary Fig. 1).

To determine if enzyme co-expression was capable of producing a functional Man₃GlcNAc₂ biosynthesis pathway, we constructed plasmid pYCG that encoded a synthetic gene cluster comprised of ALG13, ALG14, ALG1 and ALG2 (Supplementary Fig. 2). To increase the availability of the GDP-mannose substrate for Alg1 and Alg2, GDP-mannose dehydratase (GMD) that converts GDP-mannose to GDP-4-keto-6-deoxymannose in the first step of GDP-L-fucose synthesis was deleted from E. coli strain MC4100. To assay glycan synthesis, we exploited the fact that bacterial cell surfaces can display engineered oligosaccharides in their lipopolysaccharide layer^16,17. This approach depends upon the O-antigen ligase WaaL, which catalyzes the transfer of Und-PP-linked oligosaccharides to lipid A. These oligosaccharides are shuttled to the cell surface where they can be conveniently labeled^16,17. Upon labeling with fluorescent concanavalin A (ConA), a lectin that binds terminal α-mannose, MC4100 gmd::kan cells expressing the synthetic pathway but not empty-vector control cells became highly fluorescent (Fig. 2a). The fluorescence was clearly localized on the cell surface (Supplementary Fig. 3a). In the absence of ALG1 or ALG2, cell fluorescence was significantly diminished (Supplementary Fig. 3b) confirming that these enzymes were required for producing surface-associated α-mannose residues. Likewise, when the synthetic pathway was expressed in MC4100 gmd::kan that also lacked waaL, cells were minimally fluorescent (Fig. 2a) confirming WaaL-dependent transfer of α-mannose-containing oligosaccharides to lipid A. Importantly, a native E. coli flippase (e.g., Wzx) must be involved since WaaL uses Und-PP-linked oligosaccharides that are present on the periplasmic face of the cytoplasmic membrane¹⁸.

Figure 2. Characterization of LLOs produced by glycoengineered E. coli.

(a) Flow cytometric analysis of E. coli MC4100 gmd::kan or MC4100 gmd::kan ΔwaaL cells carrying plasmids as indicated. Cells were labeled with ConA-AlexaFluor prior to flow cytometry. Median cell fluorescence (M) values are given for each histogram. (b) MALDI-MS profile of permethylated glycans released from LLOs by acid hydrolysis. LLOs were extracted from E. coli MC4100 gmd::kan carrying plasmid pYCG. The major signal at m/z 1171 corresponds to [M+Na]⁺ of Hex₃HexNAc₂.

To verify the glyan structure, lipid-linked oligosaccharides (LLOs) were extracted and characterized by matrix-assisted laser desorption/ionization tandem time-of-flight (MALDI-TOF/TOF) analysis. The MALDI-MS spectrum revealed Hex₃HexNAc₂ as the primary oligosaccharide, consistent with the expected Man₃GlcNAc₂ glycan. In addition, Hex₂HexNAc₂ and Hex₄HexNAc₂ were detected (Fig. 2b). The MALDI-MS spectrum of LLOs isolated from MC4100 gmd::kan ΔwaaL cells also revealed Hex₃HexNAc₂ as the primary oligosaccharide (Supplementary Fig. 4). This confirmed that the lack of cell surface labeling observed for these cells was a result of the waaL deletion and not the inability to synthesize oligosaccharides. Finally, released glycans analyzed by ¹H NMR spectroscopy were consistent with the eukaryotic core glycan Manα1–3(Manα1–6)-Manβ1–4-GlcNAcβ1–4-GlcNAc (Supplementary Figs. 5–7). NMR analysis also revealed a residue with H-1 (5.080 ppm) and H-2 (4.065 ppm) chemical shifts indicating that the fourth hexose residue was likely Man linked to one of the branching Man residues (Supplementary Fig. 5). The presence of a putative Man₄GlcNAc₂ was surprising because elongation of Man₃GlcNAc₂ is attributed to the bifunctional Alg11¹³. It should be noted, however, that both Man₃GlcNAc₂-PP-Dol and Man₄GlcNAc₂-PP-Dol accumulated in a S. cerevisiae ALG11 mutant¹⁹, suggesting that Alg1 or Alg2 may catalyze Man₄GlcNAc₂-PP-Dol production in vivo.

To transfer Man₃GlcNAc₂ glycans to secretory glycoproteins in vivo, we focused our attention on PglB from C. jejuni (PglB_Cj) because it is the best characterized bacterial OTase²⁰ and can utilize diverse Und-PP-linked oligosaccharides as substrates^2,3,8,9. For glycoprotein targets, we chose (i) E. coli maltose binding protein (MBP) which is a native periplasmic protein and (ii) anti-β-galactosidase single-chain antibody fragment called scFv13-R4 that was modified with an N-terminal co-translational export signal from E. coli DsbA¹⁷. These proteins were each modified C-terminally with four tandem repeats of the bacterial glycan acceptor motif DQNAT¹⁷. MC4100 gmd::kan ΔwaaL cells were transformed with plasmids encoding one of these target proteins and the Man₃GlcNAc₂ pathway with PglB_Cj (Supplementary Fig. 2). The MBP^4x-DQNAT and scFv13-R4^4x-DQNAT produced in these cells, but not in cells carrying an inactive PglB_Cj mutant³, was bound by ConA (Fig. 3a and Supplementary Fig. 8). When these target proteins were first treated with peptide:N-glycosidase F (PNGase F), an amidase that specifically cleaves between a reducing-end GlcNAc and asparagine, ConA binding was eliminated (Fig. 3a). To further confirm that glycans were linked specifically to asparagines in target proteins, a version of scFv13-R4 with a single C-terminal DQNAT sequon was digested with Pronase E and the resulting glycopeptides were identified using MS²¹. The major ion seen at m/z 1282 was consistent with Man₃GlcNAc₂-Asn, wherein the asparagine residue underwent β-elimination during the permethylation procedure (Fig. 3b)²¹. MS analysis of the PNGase F-released glycans from glycosylated scFv13-R4^4x-DQNAT revealed Hex₃HexNAc₂ as the predominant glycoform along with a lesser amount of Hex₄HexNAc₂ (Fig. 3c). MS² sequencing of the glycan at m/z 1171 confirmed the biantennary trihexosyl structure (Supplementary Fig. 9a). When PNGase F-released glycans were treated with α-exomannosidase to specifically hydrolyze terminal α-mannose residues, HexHexNAc₂ emerged as the major glycoform at the expense of both Hex₃HexNAc₂ and Hex₄HexNAc₂ (Supplementary Fig. 9b). Finally, ¹H NMR analysis on PNGase F-released glycans was consistent with Manα1–3(Manα1–6)-Manβ1–4-GlcNAcβ1–4-GlcNAc (Supplementary Figs. 10 and 11).

Figure 3. Transfer of eukaryotic glycans to target proteins in E. coli.

(a) Western blot analysis of scFv13-R4^4x-DQNAT affinity purified from E. coli MC4100 gmd::kan ΔwaaL cells carrying pYCG-PglB_Cj or pYCG-PglB_Cjmut as indicated. Proteins isolated from cells expressing wild-type PglB_Cj were further treated with PNGase F for removal of N-linked glycans. Polyhistidine tags on the proteins were detected using anti-His antibodies while mannose glycans on the proteins were detected using ConA. See Supplementary Fig. 8 for full, uncut gel image. (b) MALDI-MS profile of permethylated glycopeptides generated by digestion of scFv13-R4^1x-DQNAT with Pronase E. The major signal at m/z 1282 corresponds to the permethylation product of Hex₃HexNAc₂-N, where the asparagine residue underwent β-elimination during the permethylation procedure (see inset). (c) MALDI-MS profile of permethylated glycans released from scFv13-R4^4x-DQNAT by PNGase F treatment. The major signal at m/z 1171 corresponds to [M+Na]⁺ of Hex₃HexNAc₂.

We next attempted to transfer Man₃GlcNAc₂ to eukaryotic glycoproteins including: (i) the Fc domain of human IgG1 at its conserved N297 glycosylation site, (ii) bovine ribonuclease A (RNaseA) at its N34 acceptor site, and (iii) the placental variant of human growth hormone (hGHv) at its N140 glycosylation site. The genes encoding these proteins were cloned downstream of an N-terminal DsbA export signal or full-length MBP in the case of hGHv. Since the N-X-S/T consensus motif in eukaryotes is extended to D/E-X₋₁-N-X₊₁-S/T in bacteria⁴, we mutated the native glycosylation motifs in the Fc (QYNST, residues 295–299) and hGHv (IFNQS, residues 138–142) to DQNAT. Likewise, we used an RNaseA variant with an S32D substitution²². Expression of these target proteins in cells carrying the pYCG-PglB_Cj plasmid yielded clearly glycosylated proteins (Supplementary Fig. 12a and b). It should be noted that RNaseA glycosylation was unexpected because the acceptor site is located in a structured domain that is not glycosylated by PglB_Cj in vitro²² Hence, our data indicate that PglB_Cj can glycosylate residues in both unstructured and structured regions of eukaryotic acceptor proteins in vivo.

Since it does not have native glycosylation pathways, our engineered E. coli strain is the only platform for glycoprotein expression that offers bottom-up synthesis of precise glycan structures by expression of diverse GTases and OTases. Despite our success, however, there remain some important challenges that need to be overcome for the practical application of this technology. For example, an acidic group at the -2 position to the asparagine seems to be a common prerequisite of PglB homologs for efficient glycosylation⁴. Relaxed acceptor site specificity has been reported for C. lari and Desulfovibrio desulfuricans PglB homologs^5,6. However, this has only been shown for one very unique site (₂₇₁DNNNST₂₇₆) in the C. jejuni AcrA acceptor protein. PglB_Cl did not glycosylate the wild-type C_H2 domain of a human IgG1⁵. In our hands, PglB_Cj and PglB_Cl were able to transfer Man₃GlcNAc₂ to extended sites (Supplementary Fig. 12c) but not to minimal glycosylation sites in engineered or eukaryotic target proteins (data not shown). Another issue is that only a small fraction (<1%) of each expressed protein was glycosylated under the conditions tested here. With that said, the yield of glycosylated proteins has reached up to ~50 μg/L in our hands and might be further improved by increasing expression in the periplasm, relieving enzymatic and metabolic bottlenecks, and/or optimizing the glycosylation enzymes. Along these lines, simple optimization strategies have previously been used to generate nearly 25 mg/L of bacterial glycoproteins in E. coli⁹. We anticipate further improvements will be achieved by applying new glyco-display technologies including cell surface and phage display systems^17,23,24. Such methods will be needed to create bacterial OTase variants that efficiently glycosylate minimal N-X-S/T acceptor sites. Alternatively, novel bacterial OTases with distinct properties⁶ or single-subunit eukaryotic OTases²⁵ could prove useful. Overall, the engineering of defined glycosylation pathways in E. coli sets the stage for further engineering of this host for the production of vaccines and therapeutics with even more structurally complex human-like glycans. Moreover, glycoengineered E. coli has the potential to serve as a model genetic system for deciphering the “glycosylation code” which governs the non-template driven synthesis of diverse glycans and their specific attachment to proteins.