Engineering Pathways in Central Carbon Metabolism Help to Increase Glycan Production and Improve N-Type Glycosylation of Recombinant Proteins in E. coli

Escherichia coli strains have been modified in a variety of ways to enhance the production of different recombinant proteins, targeting membrane protein expression, proteins with disulphide bonds, and more recently, proteins which require N-linked glycosylation. The addition of glycans to proteins remains a relatively inefficient process and here we aimed to combine genetic modifications within central carbon metabolic pathways in order to increase glycan precursor pools, prior to transfer onto polypeptide backbones. Using a lectin screen that detects cell surface representation of glycans, together with Western blot analyses using an O-antigen ligase mutant strain, the enhanced uptake and phosphorylation of sugars (ptsA) from the media combined with conservation of carbon through the glyoxylate shunt (icl) improved glycosylation efficiency of a bacterial protein AcrA by 69% and over 100% in an engineered human protein IFN-α2b. Unexpectedly, overexpression of a gene involved in the production of DXP from pyruvate (dxs), which was previously seen to have a positive impact on glycosylation, was detrimental to process efficiency and the possible reasons for this are discussed.


Introduction
Glycosylation is regarded as one of the most common post-translational modifications of proteins, occurring on almost half of all proteins in nature [1]. The addition of a sugar group impacts on protein biophysical properties, which can influence native functions, for example, activity, localisation and half-life. Although present across all domains of life, eukaryotic systems remain the most characterised and therefore the best understood. In humans, it is thought that nearly half of all proteins are modified by the addition of sugars, with 90% of the molecular mass of some glycoproteins made up by glycans. More recently, a growing interest in prokaryotic glycosylation has led to improved understanding of the molecular mechanisms involved and the functional traits associated with this relatively energy-expensive protein modification process. However, overall knowledge is relatively poor due to the diversity of glycan structures and variability of the processes in prokaryotic systems.
Escherichia coli is a well-established protein production cell factory, which has been exploited industrially for several decades. From high value therapeutics to high bulk industrial enzymes, E. coli glycosylation process. In the same study, overexpression of dxs gave a 1.3-fold increase in glycosylation efficiency. The dxs gene encodes for 1-deoxyxylulose-5-phosphate (DXP) synthase, which catalyses the production of DXP from pyruvate and D-glyceraldehyde-3-phosphate (D-GAP) in a thiamine diphosphate-dependent manner [23,24]. The production of DXP is deemed to be the rate-limiting step in the mevalonate pathway of isoprenoid synthesis [25], but downstream of DXP synthesis, it is involved in the production of a number of molecules including undecaprenyl phosphate (Und-P), which acts as the bacterial lipid anchor on which the glycans are built. These modifications target metabolic processes where carbon flux can be diverted from central carbon metabolism and re-directed towards the recombinant glycosylation pathway, or enable efficient conversion of carbon without compromising cellular growth rates.
In this study, the effect of the overexpression of these three genes (ptsA, icL and dxS), in various combinations, were analysed using a lectin screen that detects cell surface representation of glycans through the exploitation of the native E. coli O-antigen ligase, WaaL, in MC4100 E. coli cells. Western blot analysis was conducted to determine the protein glycosylation efficiency and relative glycoprotein production in the bacterial strain, CLM24, where waaL has been knocked out to prevent cell surface representation of recombinant glycans [26]. The aim was to firstly characterise the effect of the overexpression of these previously identified genes on glycan production, followed by impacts on glycosylation efficiency. The bacterial exemplar N-glycoprotein, AcrA, was initially used as it is the most understood target glycoprotein, together with the model bacterial N-glycosylation machinery termed protein glycosylation pathway 2 (pgl2) [27]. To explore whether the engineered cells could improve the glycosylation of an industrially relevant protein, IFN-α2b was also tested as a target glycoprotein. IFN-α2b is a frequently used cytokine for therapeutic purposes and commercial variants are produced in E. coli e.g., Grippferon ® (Firn-M, Moscow, Russia) [28]. It contains a known O-glycosylation site in a relatively unstructured region, and this was modified to enable bacterial N-glycosylation.

Materials and Methods
All materials were purchased from Sigma-Aldrich (Dorset, UK), with working antibiotic concentrations of 100 µg/mL ampicillin, 35 µg/mL chloramphenicol, and 35 µg/mL kanamycin used where appropriate, unless stated otherwise.

PCR, DNA Cloning and Vectors
The strategy employed to overexpress the genes of interest was to place them under a constitutive promoter on the same plasmid as the target protein. To achieve this, a new multiple cloning site (MCS) was inserted into pEC(acrA). The MCS contained the necessary restriction sites to insert the three genes of interest and was designed with the constitutive promoter, J23119, from the iGEM catalogue (Table  S1). This 35 bp promoter was placed upstream of the restriction sites required for insertion. Forward primers used to amplify the genes of interest contained the Shine Dalgarno sequence, AGGAGG, and a six base pair gap to ensure the ribosome binding site (RBS) was situated upstream of the start codon (Table S2). Upon creation of the various constructs, plasmids were sent off for sequencing to confirm insertion and to ensure that no PCR errors were incorporated.
PCR reactions: 98 • C for 3 min, 30 cycles of 30 s at 98 • C, 30 s at the annealing temperature of the primer, and an extension step at 72 • C run at 1 kb per 30 s, extended depending on the size of the fragment being amplified. A final extension step was also run at 72 • C for 10 min before cooling the reaction to 4 • C.
Post amplification of the MCS, the reaction was subjected to agarose gel analysis, with the DNA being extracted using a Qiagen ® kit (Hilden, Germany) prior to digestion with EcoRI and XmaI at 37 • C for 1 h. The target plasmid pEC (acrA) was also digested before insertion of the MCS using T4 ligase (NEB) following the manufacturer's protocol to create the pEC (acrA_MCS) plasmid. The three genes to be inserted were amplified from the plasmids outlined in the aforementioned studies [10,20] and cloned into pEC (acrA_MCS) with the specific restriction enzymes that were required for the particular inserts. The seven constructs were all validated with restriction digestion and sequence analysis prior to transformations in the appropriate strains with pACYC (pgl2).
The amino acid sequence for an IFN-α2b with a C. jejuni glycosylation consensus sequence was designed by locating a suitable region based on the crystal structure available in the PDB (1ITF). The T corresponds to residue T106, an O-linked glycosylation site and due to being in a relatively unstructured region on the protein, it was chosen as the site for modification. A glutamic acid (E) and asparagine (N) were inserted at site 102 and 104, respectively, into the amino acid sequence using the nucleotide sequence for IFN-α2b (NCBI sequence ID CAA23809.1), together with the pelB signal sequence and a C-terminal hexa-his tag sequence (Supplementary Files). The subsequent sequence was codon optimised, synthesised by DNA 2.0 and used to replace acrA in the pEC (acrA_MCS) plasmid. A Western blot using a lectin peroxidase blot was used to test for glycosylation as described previously [10].

Cell Surface Representation of Glycans
One mL of LB in a sterile 1.5 mL centrifuge tube was inoculated with the bacterial strain of interest, and the appropriate antibiotics were added. Oxygen-limited growth took place overnight at 37 • C shaking at 180 rpm to limit cell density. The O.D at 600 nm was measured and cultures were normalised down to 0.6 using sterile deionised water. The cells were then diluted by a factor of 1 in 75,000 using sterile deionised water to a final volume of 1 mL. One-hundred µL of the diluted cells were plated out on LB agar with the appropriate antibiotics and left to incubate and grow aerobically at 37 • C overnight. A piece of Protran™ nitrocellulose paper was cut to fit a petri dish and soaked in the appropriate antibiotics for 5 min before being left to dry in a flow hood. The paper was placed over the petri dish, making contact with the colonies and left to incubate for 3 h at 37 • C. The nitrocellulose membrane was blocked in PBS containing 2% Tween 20 for 2 min at room temperature, before being washed twice with PBS for 10 min. The membrane was incubated in PBS with 0.05% TWEEN ® 20, 1 mM CaCl 2 , 1 mM MnCl 2 , 1 mM MgCl 2 , and 3 µg soybean agglutinin lectin peroxidase (specific for GalNAc) for 16 h at 20 • C. The membrane was then washed twice in PBS for 10 min before detecting and analysing the colonies using Immobilon™ chemiluminescent HRP substrate (Millipore) with ImageQuant™ RT ECL (GE Healthcare, Chicago, IL, USA), fitted with a 16-bit CCD camera.

Colony Analysis
Images were analysed using version 7 of the ImageQuant software (GE Healthcare, Chicago, IL, USA). Single colonies were selected and those in close proximity of another colony were disregarded to ensure minimal contamination resulting from two colony selection. Pixel intensity and area were calculated and corrected to remove background. Density measurements were tallied, the means plotted and compared to the control.

Bacterial Growth and Protein Expression
The starter inoculum of the bacterial strains were cultured overnight at 37 • C, 180 rpm. One-hundred-mL cultures of Luria Broth media were inoculated with 1 mL of the starter culture and incubated at 37 • C, 180 rpm. At an optical density (O.D) of 0.5 at 600 nm, 0.2% (v/v) L-arabinose was added to induce target protein expression through the araBAD promoter. No induction of the pgl pACYC (pgl2) plasmid was necessary due to constitutive expression. Cells were then left to incubate and express the protein for 4 h at the lower temperature of 30 • C 180 rpm in order to aid in the expression of correctly folded protein and minimise growth. The final OD of the cultures was measured and 40 OD units worth was harvested through centrifugation at 4 • C, 4500× g for 10 min. The supernatant was discarded and the pellet stored at −20 • C.

Periplasmic Protein Extraction
Protein pellets were thawed on ice and resuspended in 1 mL of periplasmic lysis buffer (20% sucrose, 1 g/L lysozyme, 30 mM Tris-HCl pH 8.5, 1× Halt protease inhibitor complex (Thermo Fisher Scientific, Massachusetts, MA, USA)) and extracted on ice for 2 h. Samples were centrifuged at 4500× g at 4 • C for 10 min. The supernatant, representing the soluble periplasmic fraction, was collected and the total protein quantity was measured using a Bradford protein assay.

SDS-PAGE and Western Blot Analysis
Five micrograms of periplasmic protein extract was added to 6 µL NuPAGE™ LDS sample loading buffer (4×) (Thermo Fisher Scientific, Massachusetts, MA, USA), 2.4 µL of sample reducing agent, with sufficient water added to make a final volume of 24 µL. Samples were heated to 80 • C for 10 min and left to cool to room temperature. Precast NuPAGE™ Novex 4-12% Bis-Tris gels (Thermo Fisher Scientific, Massachusetts, MA, USA) were loaded into the gel apparatus and the two chambers filled with the required MOPS SDS buffer. Samples were loaded into the wells and the gel run for 1.5 h at 180 V. The gel was extracted from the cast and washed in deionised water. Proteins were transferred to a nitrocellulose membrane using an iBlot ® (Thermo Fisher Scientific, Massachusetts, MA, USA). The membrane was subsequently blocked in blocking buffer (5% milk powder in Tris-Buffered Saline (TBS) with 0.1% v/v Tween20) for 1 h at room temperature. Three × 10 min washes of the membrane in TBS 0.1% Tween20 was followed by overnight incubation at 4 • C in blocking buffer with a His-tag antibody (Abcam ® Anti-6X His tag ® (HRP), Cambridge, UK) (1:10,000). The excess unbound antibody was washed off with 3 × 10 min washes in TBS 0.1% Tween20. The HRP linked antibody was detected on the blot using 10 mL of TMB-Ultra blotting solution (Thermo Fisher Scientific, Massachusetts, MA, USA) for 20 min. For the lectin peroxidase blot, the same procedure was followed except the antibody was replaced with soybean agglutinin (SBA) lectin (specific for GalNAc) peroxidase as described previously [10]. Pictures were captured using a 16-bit CCD camera. Images were analysed using Image Studio Lite v5.2. Bands were highlighted and the pixel intensity and area calculated to obtain the densitometry measurements with background correction factored into the analysis. Where applicable, a quantified standard of His-tag purified aglycosylated AcrA was used to measure the amounts of the target produced, as described previously [19]. Samples were run in triplicate and an unpaired t-test with Welch's correction (assuming both the populations do not have the same standard error) was undertaken to look for statistically significant differences (P < 0.05). The quantification of total glycoprotein per litre of pgl2 IFN generated in flask culture was calculated through the analysis of the final OD of the culture and the amount of glycosylated target protein extracted from the periplasm of 30 OD units of culture as a percentage of the total protein present in the same 30 OD of periplasm. In brief, the final OD 600 measurement of the culture was measured with 1 mL of OD 600 , 1.0 being equal to 1 OD unit. Thirty OD units were harvested, the periplasmic protein extracted, quantified and 5 µg loaded onto a gel for western alongside 0.5 µg of aglycosylated, His tagged pure AcrA standard. The target protein was quantified against the standard, and the amount of target protein present per 30 OD of periplasm was determined. This was then used to determine the total amount of target glycoprotein present in 1 L of culture.

MC4100 Cell Surface Representation of Glycans
Constructs were transformed into E. coli MC4100 along with the protein glycosylation machinery Pacyc (pgl2). As a positive control, the cell line MC4100 with pEC (acrA_MCS) pACYC (pgl2) was created. For the negative control, just the pEC (acrA_MCS) plasmid was transformed into the cell line MC4100. MC4100 was chosen because of the presence of the waaL gene. This gene expresses an O-antigen ligase that can recognise the glycan as a potential substrate and attach it to a lipid A core, which subsequently gets exported to the cell surface, placing the glycan on the extremity of the cell. By having this pathway intact and not inducing the expression of the target protein, it allows the presentation of the glycans on the cell surface. This enabled semi-quantitative analysis of glycan production through immunoblotting the surface of the cells with a lectin specific to the sugar residues in the pgl2 glycan.
Cell surface representation of glycans produced by the various cell lines are shown in Supplementary Files ( Figure S1). Fifty-five single isolated colonies were analysed at random on each plate and their densitometry compared to that of the positive control. From this, the glycan production of each strain was measured and compared based on relative intensity (Figure 1). Although five of the engineered strains expressed statistically significantly more glycans on their cell surface, changes were very small. Of the combinations tested, the cells expressing both ptsA and icl had the biggest influence on glycan production, but they only had a 6.4% increase ( Figure S1B). However, cells expressing dxs alone or in combination with icl had an inverse effect ( Figure 1). The 3.2% increase observed with pEC (acrA_ICL) was reduced to a negative 2.1% change when icl was over expressed in combination with dxs in pEC (acrA_dxs_ICL). In subsequent experiments, a bacterial strain with a waaL deletion was used so that glycans were transferred to the expressed target protein. This enabled us to analyse whether the findings of the surface glycan screen were actually translated to overall glycoprotein production and glycosylation efficiency. By having this pathway intact and not inducing the expression of the target protein, it allows the presentation of the glycans on the cell surface. This enabled semi-quantitative analysis of glycan production through immunoblotting the surface of the cells with a lectin specific to the sugar residues in the pgl2 glycan. Cell surface representation of glycans produced by the various cell lines are shown in Supplementary Files ( Figure S1). Fifty-five single isolated colonies were analysed at random on each plate and their densitometry compared to that of the positive control. From this, the glycan production of each strain was measured and compared based on relative intensity (Figure 1). Although five of the engineered strains expressed statistically significantly more glycans on their cell surface, changes were very small. Of the combinations tested, the cells expressing both ptsA and icl had the biggest influence on glycan production, but they only had a 6.4% increase ( Figure S1B). However, cells expressing dxs alone or in combination with icl had an inverse effect ( Figure 1). The 3.2% increase observed with pEC (acrA_ICL) was reduced to a negative 2.1% change when icl was over expressed in combination with dxs in pEC (acrA_dxs_ICL). In subsequent experiments, a bacterial strain with a waaL deletion was used so that glycans were transferred to the expressed target protein. This enabled us to analyse whether the findings of the surface glycan screen were actually translated to overall glycoprotein production and glycosylation efficiency.

Expression of Glycosylated AcrA in E. coli Cells, CLM24
Constructs were transformed into E. coli CLM24, a strain that is waaL deficient, along with the protein glycosylation machinery pACYC (pgl2). As a positive control, the cell line CLM24 with pEC

Expression of Glycosylated AcrA in E. coli Cells, CLM24
Constructs were transformed into E. coli CLM24, a strain that is waaL deficient, along with the protein glycosylation machinery pACYC (pgl2). As a positive control, the cell line CLM24 with pEC (acrA_MCS) pACYC (pgl2) was used. For the negative control, the pEC (acrA_MCS) plasmid was transformed into cells. The bacteria were grown in culture and induced with L-arabinose to express the target protein. SDS-PAGE and subsequent Western blotting was used to analyse production (using an anti-His antibody), with the densitometry of the bands used to calculate the relative glycoprotein production and glycosylation efficiency ( Figure S2). Figure 2 shows the relative glycosylation efficiency of the engineered strains. Some are represented with an absence of data due to the lack of glycoprotein producing samples indicating a detrimental effect on glycan transfer to the protein target. Of the seven combinations tested, three were shown to have statistically different glycosylation efficiency when compared to the control. pEC (acrA_ptsA_ICL) gave the highest glycosylation efficiency, with a 1.69-fold increase. Two of the other significant strains, pEC (acrA_ptsA_dxs_ICL) and pEC (acrA_ptsA), caused a 1.48-fold and 1.27-fold increase in efficiency, respectively.
Bioengineering 2019, 6, x FOR PEER REVIEW 7 of 13 (acrA_MCS) pACYC (pgl2) was used. For the negative control, the pEC (acrA_MCS) plasmid was transformed into cells. The bacteria were grown in culture and induced with L-arabinose to express the target protein. SDS-PAGE and subsequent Western blotting was used to analyse production (using an anti-His antibody), with the densitometry of the bands used to calculate the relative glycoprotein production and glycosylation efficiency ( Figure S2). Figure 2 shows the relative glycosylation efficiency of the engineered strains. Some are represented with an absence of data due to the lack of glycoprotein producing samples indicating a detrimental effect on glycan transfer to the protein target. Of the seven combinations tested, three were shown to have statistically different glycosylation efficiency when compared to the control. pEC (acrA_ptsA_ICL) gave the highest glycosylation efficiency, with a 1.69-fold increase. Two of the other significant strains, pEC (acrA_ptsA_dxs_ICL) and pEC (acrA_ptsA), caused a 1.48-fold and 1.27-fold increase in efficiency, respectively. Glycosylation efficiency of the strains expressing acrA, the metabolic engineering genes and the pgl2 machinery. Cells without metabolic engineering of central carbon metabolism, CLM24 pEC(acrA) pACYC(pgl2), was included for a comparative control. Efficiency calculated through measuring the density of aglycosylated and glycosylated bands of AcrA. "A" denotes acrA expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired ttest with Welch's correction, n = 3; P < 0.05). AcrA is the control strain comprising of the target protein gene (acrA) and glycosylation machinery contained in the pgl2 plasmid without any additional metabolic engineering elements.
Two of the constructs were detrimental towards the glycosylation process within the cell, as no or few glycoprotein bands were observed on the Western blots. Both of the affected strains over expressed dxs (either alone or with icl). Overexpression of dxs reduces glycosylation efficiency when combined with the increased expression of both ptsA and icl (expressing all three ptsA, dxs and icl genes), compared to strains engineered with ptsA and icl alone. Of the seven combinations tested, the Glycosylation efficiency of the strains expressing acrA, the metabolic engineering genes and the pgl2 machinery. Cells without metabolic engineering of central carbon metabolism, CLM24 pEC(acrA) pACYC(pgl2), was included for a comparative control. Efficiency calculated through measuring the density of aglycosylated and glycosylated bands of AcrA. "A" denotes acrA expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05). AcrA is the control strain comprising of the target protein gene (acrA) and glycosylation machinery contained in the pgl2 plasmid without any additional metabolic engineering elements.
Two of the constructs were detrimental towards the glycosylation process within the cell, as no or few glycoprotein bands were observed on the Western blots. Both of the affected strains over expressed dxs (either alone or with icl). Overexpression of dxs reduces glycosylation efficiency when combined with the increased expression of both ptsA and icl (expressing all three ptsA, dxs and icl genes), compared to strains engineered with ptsA and icl alone. Of the seven combinations tested, the most efficient combination in terms of glycosylation efficiency was ptsA and icL, which raised efficiency from 24.6% to 41.6%. This 1.69-fold increase from the control was reduced to a 1.47-fold increase with the further overexpression of dxs and highlights its negative impact. The reduction of efficiency from 31.2% in pEC (acrA_ptsA) to 24.8% in pEC (acrA_ptsA_dxs) and the negative effect from dxs addition in pEC (acrA_dxs_ICL), whereby its inclusion inhibited the cells ability to produce detectable glycoproteins in two of the three biological replicates, adds to this argument.
The cell response to overexpression of dxs was unexpected, as this was previously shown to increase glycosylation efficiency 1.3-fold in a similar background strain, although with a different glycan structure [10]. Moreover, the growth rates of the strains tested did not differ significantly (data not shown). In our previous study, although overexpression of dxs showed a slight negative impact on the growth rate, this was not statistically significant. Interestingly, it did show a 1.6-fold reduction in recombinant protein yields [10]. DXP is used in isoprenoid biosynthesis, which is known to be involved in a wide variety of biological functions. It is therefore expected that fluxes within the isoprenoid metabolic network are tightly regulated, temporally and spatially. This makes it very difficult to dissect the mechanistic impact of DXP overexpression in terms of growth and protein production. When dxs was over expressed with a stronger promoter system (synthetic trc promoter), as in a previous study [29], growth yield (this was based on final OD) was severely inhibited, and it is hypothesised that this may be due to a drain on glycolytic intermediates necessary for balanced growth. It could be that if induced more moderately, growth is not impacted significantly, but when switched to protein production phase, the intermediates for protein production are affected.
In another study, it was elegantly demonstrated how promoter strength can affect the impact of dxs on downstream metabolic pathways in E. coli [30]. In the study, dxs was identified as the first enzymatic step in the production of a carotenoid, and in a wild type E. coli strain, increasing promoter strengths led to elevated carotenoid expression only to a certain level. It was hypothesised that other enzymes in the pathway become rate-limiting as well as cause toxic accumulation of the dxs product, DXP. In contrast, a strain engineered to overexpress these downstream enzymes showed a linear increase in carotenoid expression as promoter strengths for dxs were increased [30]. This implies that an E. coli glycoprotein platform strain would require enzymes downstream of dxs to be optimised to see the benefit of dxs overexpression. A map illustrating metabolic pathways downstream of dxs as well as the role of icl in central carbon metabolism is shown in Supplementary Files (Figure S3).

Expression of Glycosylated IFNα2b in E. coli Cells, CLM24
Three of the engineered cell lines are capable of improving the glycosylation efficiency of AcrA, and this is based on the hypothesis that the quantity of glycan precursors are limited during recombinant glycoprotein induction. Therefore, we expected that other glycoprotein targets should also have higher sugar occupancy levels in the engineered strains. We selected IFNα2b as this is an industrially-relevant, therapeutic protein. IFNα2b is a relatively small protein at 23.7 kDa. A single bacterial N-glycosylation consensus sequence of EVNVT was engineered from amino acid 102 to 106 (Figure 3). Human IFN fragments have been expressed in E. coli previously, including transport to the periplasmic space [31,32]. A Western blot using soyabean agglutinin lectin and lectin peroxidase was used to detect a glycosylated protein band to ensure the site was accessible for the C. jejuni oligosaccharyltransferase PglB (Figure 3). Bioengineering 2019, 6, x FOR PEER REVIEW 9 of 13 9 Figure 3. N-glycosylation of IFN-α2b in E. coli (left: crystal structure. GVGVT amino acid sequence was changed to EVNVT. In PDB, 1ITF GVGVT is at position 102-106. The T corresponds to the Olinked glycosylation site as residue T106 (The 102GVGV105 is marked in red and T106 in green); right: lectin peroxidase blot with AcrA control.
As Figure 4 shows, the glycosylation efficiency of IFNα2b in control cells pEC (IFN_MCS) pACYC (pgl2) was approximately 10.9% (Western blots shown in Figure S4). However, all engineered cells showed statistically significant improvements in the ratio of protein which carried the hexasaccharide glycan structure. pEC (IFN_ptsA) was the highest with 24.5% glycosylated. pEC (IFN_ptsA_ICL) and pEC (IFN_ptsA_ICL_dxs) glycosylated with 22.6% and 22.8% efficiency, respectively.  IFN-α2b, the metabolic engineering genes and the pgl2 machinery. CLM24 pEC (IFN-α2b) pACYC (pgl2) is included for a comparative control. Efficiency was calculated through measuring the density of glycosylated and glycosylated bands of IFN-α2b. "I" denotes IFN-α2b expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05). As Figure 4 shows, the glycosylation efficiency of IFNα2b in control cells pEC (IFN_MCS) pACYC (pgl2) was approximately 10.9% (Western blots shown in Figure S4). However, all engineered cells showed statistically significant improvements in the ratio of protein which carried the hexasaccharide glycan structure. pEC (IFN_ptsA) was the highest with 24.5% glycosylated. pEC (IFN_ptsA_ICL) and pEC (IFN_ptsA_ICL_dxs) glycosylated with 22.6% and 22.8% efficiency, respectively. As Figure 4 shows, the glycosylation efficiency of IFNα2b in control cells pEC (IFN_MCS) pACYC (pgl2) was approximately 10.9% (Western blots shown in Figure S4). However, all engineered cells showed statistically significant improvements in the ratio of protein which carried the hexasaccharide glycan structure. pEC (IFN_ptsA) was the highest with 24.5% glycosylated. pEC (IFN_ptsA_ICL) and pEC (IFN_ptsA_ICL_dxs) glycosylated with 22.6% and 22.8% efficiency, respectively.  IFN-α2b, the metabolic engineering genes and the pgl2 machinery. CLM24 pEC (IFN-α2b) pACYC (pgl2) is included for a comparative control. Efficiency was calculated through measuring the density of glycosylated and glycosylated bands of IFN-α2b. "I" denotes IFN-α2b expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05). . Glycosylation efficiency of the strains expressing IFN-α2b, the metabolic engineering genes and the pgl2 machinery. CLM24 pEC (IFN-α2b) pACYC (pgl2) is included for a comparative control. Efficiency was calculated through measuring the density of glycosylated and glycosylated bands of IFN-α2b. "I" denotes IFN-α2b expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05).
We also quantified total IFN-α2b glycoprotein due to its industrial relevance, and pEC (IFN_ptsA_ICL) pACYC (pgl2) cells were the largest producer, with 882.9 µg L −1 compared to 351 µg L −1 in the control ( Figure 5). The need to optimise the expression of dxs was again evident in these strains, but the successful improvement in N-glycosylation capability through overexpression of ptsA and icl was evident.
Bioengineering 2019, 6, x FOR PEER REVIEW 10 of 13 10 We also quantified total IFN-α2b glycoprotein due to its industrial relevance, and pEC (IFN_ptsA_ICL) pACYC (pgl2) cells were the largest producer, with 882.9 μg L −1 compared to 351 μg L −1 in the control ( Figure 5). The need to optimise the expression of dxs was again evident in these strains, but the successful improvement in N-glycosylation capability through overexpression of ptsA and icl was evident. Figure 5. Average µg glycoprotein per litre of cell culture expressing IFN-α2b along with the metabolic engineering genes and the pgl2 machinery. CLM24 pEC (IFN-α2b) pACYC (pgl2) is included for a comparative control. Quantity was calculated through measuring the density of glycosylated bands of IFN-α2b compared to a loading control of 0.5 µg of purified AcrA. "I" denotes IFN-α2b expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired ttest with Welch's correction, n = 3; P < 0.05).
Variations in glycosylation efficiency in different proteins, for example, AcrA and IFN-α2b, are expected. These proteins are expressed to different levels and have very different structures, potentially affecting the accessibility of the PglB oligosaccharyltransferase. Moreover, AcrA has two bacterial N-glycosylation sites, N123 and N273, where the nitrogen on asparagine acts as the site for covalent attachment to the glycan, compared to just one site engineered into IFN-α2b at N104. Both the construction of glycans through glycosidic bonds and attachment of the hexasaccharide construct to polypeptide backbones requires energy in the form of ATP [33] and, therefore, less sites would reduce energy demands on the cell. However, both target proteins showed higher glycosylation efficiency when both ptsA and icl were over expressed. As suggested previously, the enhanced uptake and phosphorylation of sugars from the media is beneficial for an E. coli glycoprotein production strain, providing precursors for glycan construction. Moreover, using the strains constructed for this study, further conservation of carbon through increased flux through the glyoxylate pathway is a complementary modification to the cell. Unexpectedly, overexpression of dxs was detrimental here, even eliminating glycoprotein detection completely in some strains. Although it has been hypothesised that phenotypes related to increased dxs expression require similar optimisation of associated pathway enzymes, dxs is involved in isoprenoid production and therefore is used in diverse cellular processes beyond Und-P production. Although Und-P abundance could be a bottleneck in N-glycosylation, further studies would need to be undertaken to evidence this prior to identifying a gene target or tuning dxs expression that can work in tandem with ptsA and icl overexpression. Figure 5. Average µg glycoprotein per litre of cell culture expressing IFN-α2b along with the metabolic engineering genes and the pgl2 machinery. CLM24 pEC (IFN-α2b) pACYC (pgl2) is included for a comparative control. Quantity was calculated through measuring the density of glycosylated bands of IFN-α2b compared to a loading control of 0.5 µg of purified AcrA. "I" denotes IFN-α2b expression and asterisks (*) above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05).
Variations in glycosylation efficiency in different proteins, for example, AcrA and IFN-α2b, are expected. These proteins are expressed to different levels and have very different structures, potentially affecting the accessibility of the PglB oligosaccharyltransferase. Moreover, AcrA has two bacterial N-glycosylation sites, N123 and N273, where the nitrogen on asparagine acts as the site for covalent attachment to the glycan, compared to just one site engineered into IFN-α2b at N104. Both the construction of glycans through glycosidic bonds and attachment of the hexasaccharide construct to polypeptide backbones requires energy in the form of ATP [33] and, therefore, less sites would reduce energy demands on the cell. However, both target proteins showed higher glycosylation efficiency when both ptsA and icl were over expressed. As suggested previously, the enhanced uptake and phosphorylation of sugars from the media is beneficial for an E. coli glycoprotein production strain, providing precursors for glycan construction. Moreover, using the strains constructed for this study, further conservation of carbon through increased flux through the glyoxylate pathway is a complementary modification to the cell. Unexpectedly, overexpression of dxs was detrimental here, even eliminating glycoprotein detection completely in some strains. Although it has been hypothesised that phenotypes related to increased dxs expression require similar optimisation of associated pathway enzymes, dxs is involved in isoprenoid production and therefore is used in diverse cellular processes beyond Und-P production. Although Und-P abundance could be a bottleneck in N-glycosylation, further studies would need to be undertaken to evidence this prior to identifying a gene target or tuning dxs expression that can work in tandem with ptsA and icl overexpression.

Conclusions
The physiology and genetics of E. coli are the most understood of all prokaryotic host cell factories for the production of recombinant proteins. Although the addition of glycans to proteins is not a native process in E. coli, there are advantages of incorporating this capability into the system, such as improving the stability of industrial enzymes through to diagnostic applications. Little is known about how the Campylobacter jejuni N-glycosylation system impacts on metabolism in E. coli, but here we aimed to combine gene modification targets from previous studies, based on the hypothesis that central carbon metabolism requires optimisation to improve glycosylation efficiency. Although combining the overexpression of genes involved in sugar uptake and phosphorylation with pathways that enable carbon conservation did not show large differences in glycan precursor production through a cell surface display screen, significant increases in glycosylation efficiency on recombinant proteins were revealed. AcrA and IFNα2b glycosylation increased 1.69-fold and 2.2-fold, respectively. However, not all modifications were positive (dxs overexpression), highlighting the need to test multiple cell engineering targets together and also the requirement for tuning the overexpression of individual genes for optimisations, providing a more comprehensive understanding of cellular constraints.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2306-5354/6/1/27/s1, Figure S1: A: GalNAc specific lectin peroxidase screen against the pgl2 glycan represented by E. coli MC4100 cells containing pACYC(pgl2) and the various metabolic engineering plasmids. Target protein not induced. B: Graph showing the percentage change at the relative intensity of the colonies when compared to the control without the metabolic engineering genes. "A" denotes acrA expression and asterisks above the bars indicate strains of significant difference from the control (Unpaired t-test with Welch's correction, n = 3; P < 0.05), Figure S2: A: Western blots of the seven constructs and the control expressing AcrA along with the pACYC (pgl2) machinery. His-tag antibody was used for detection of the target protein. The three bands for each strain represent the three biological replicates. B: A demonstration of Western blot binning for densitometry analysis. C: Raw densitometry data, Figure S3: Metabolic pathways for dxs (left) and icl (right), Figure S4: Western blots of control and engineered strains expressing IFN α2b along with the pACYC (pgl2) machinery. His-tag antibody was used for detection of the target protein. The three bands for each strain represent the three biological replicates. A: Lane 1: Novex protein marker, Lanes 2-4: IFN_pgl2, Lanes 5-7: IFN_ptsA_pgl2, Lane 8: 0.5 g AcrA. B: Lane 1: Novex protein marker, Lanes 2-4: IFN_ptsA_ICL_pgl2, Lanes 5-7: IFN_ptsA_ICL_dxs_pgl2, Lane 8: 0.5 g AcrA, Table S1: Sequence of the newly designed multiple cloning site and the primers used to amplify it, Table S2: The primers used to amplify the three genes of interest and the restriction sites used to insert them into pEC (acrA_MCS).