Residues Comprising the Enhanced Aromatic Sequon Influence

Aug 18, 2017 - We devised an algorithm for prediction of N-glycosylation efficiency using the SAS software, employing the 120 sequences studied as a t...
0 downloads 23 Views 4MB Size
Article pubs.acs.org/JACS

Residues Comprising the Enhanced Aromatic Sequon Influence Protein N‑Glycosylation Efficiency Yen-Wen Huang,†,‡ Hwai-I Yang,†,§ Ying-Ta Wu,† Tsui-Ling Hsu,† Tzu-Wen Lin,† Jeffery W. Kelly,*,∥,⊥ and Chi-Huey Wong*,†,⊥ †

Genomics Research Center Academia Sinica, Taipei 115, Taiwan Institute of Biochemical Sciences, National Taiwan University, Taipei 106, Taiwan § Institute of Clinical Medicine, National Yang-Ming University, Taipei 112, Taiwan ∥ Department of Molecular Medicine and ⊥Department of Chemistry, The Scripps Research Institute, La Jolla, California 92037, United States ‡

S Supporting Information *

ABSTRACT: N-Glycosylation is an important co- and/or posttranslational modification that occurs on the vast majority of the one-third of the mammalian proteome that traverses the cellular secretory pathway, regulating glycoprotein folding and functions. Previous studies on the sequence requirements for N-glycosylation have yielded the Asn-X-Ser/Thr (NXS/T) sequon and the enhanced aromatic sequons (Phe-X-Asn-X-Thr and Phe-X-X-AsnX-Thr), which can be efficiently N-glycosylated. To further investigate the influence of sequence variation on N-glycosylation efficiency in the context of a five-residue enhanced aromatic sequon, we used the human CD2 adhesion domain (hCD2ad) to screen the i−2, i−1, i+1, and i+2 residues flanking Asn at the i position. We found that aromatic residues, especially Trp, and sulfur-containing residues at the i−2 position improved N-glycosylation efficiency, while positively charged residues such as Arg suppressed Nglycosylation. Thiol, hydroxyl, and aliphatic-based side chains at the i−1 position had higher N-glycosylation efficiency, and Cys, in particular, compensated for the negative effect of Arg at the i−2 position. Small residues and Ser at the i+1 position increased the likelihood of N-glycosylation, and Thr is better than Ser at the i+2 position. We devised an algorithm for prediction of Nglycosylation efficiency using the SAS software, employing the 120 sequences studied as a training set. We then introduced the optimized-enhanced aromatic sequons into other glycoproteins and observed an enhancement in N-glycan occupancy that was further supported by modeling the high-affinity interaction between the optimized sequence on hCD2ad and a human oligosaccharyltransferase (OST) subunit. The findings in this study provide useful information for enhancing or suppressing Nglycosylation at a site of interest and valuable data for a better understanding of OST-catalyzed N-glycosylation.



INTRODUCTION

Prior studies demonstrate that not all consensus Nglycosylation sequons are glycosylated,15−17 suggesting a poor understanding of the requirements for OST-catalyzed Nglycosylation. Statistical analysis of sequons that are utilized by OST indicates that only 65% of all NXS/T sites are Nglycosylated,18 70% of which are NXT and 30% are NXS sequons. Non N-glycosylated NXS/T sequences include 41% NXT and 59% NXS sequons. It was shown that a higher glycosylation frequency was observed when aromatic residues (Phe, Tyr, or Trp) occupied the i−2 position, upstream of Asn at position i. Small nonpolar residues (such as Gly, Ala, or Val) were shown to be preferred at the i+1 position, and bulky hydrophobic amino acids were favored at the i+3 and i+5 positions.18 A glycoproteomic analysis of a rat N-glycopeptide

N-Glycosylation is a major co- and/or post-translational protein modification that can influence protein stability and often impacts functions associated with cellular or intercellular recognition.1−3 Membrane-bound and secreted proteins are generally N-glycosylated to modulate their interaction with ligands.4−6 N-Glycosylation often occurs on the nascent chain cotranslationally during its insertion by the ribosome into the lumen of the endoplasmic reticulum. In the process an Asn-XSer/Thr sequence or sequon (X stands for any amino acid except Pro) is recognized by oligosaccharyltransferase (OST), which catalyzes the transfer of (GlcNAc2Man9Glc3) from its lipid conjugate onto the side-chain amide NH2 group of Asn.7−10 N-Glycosylated proteins have a dedicated proteostasis network to make the quality control decision, i.e., to fold and secrete the client N-glycoprotein or to degrade it.11−14 © 2017 American Chemical Society

Received: April 17, 2017 Published: August 18, 2017 12947

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

residue enhanced aromatic sequon, employing human CD2 adhesion domain (hCD2ad), a good model for such a study, as the protein has only one glycosylation site and there is no proline or disulfide group to interfere with folding. In addition, the protein’s amenability to folding and stability studies21 makes it an ideal model protein to screen the i−2, i−1, i+1, and i+2 residues flanking Asn at position i, employing saturation mutagenesis at the i−2, i−1, and i+1 positions or limited mutagenesis at the i+2 position (S/T) to evaluate N-glycan occupancy (Scheme 1). A major goal of this study was to enhance N-glycosylation efficiency of secreted proteins while maintaining protein production at near wild-type (wt) levels. We wanted to test the hypothesis that optimizing the enhanced aromatic sequon glycan occupancy for hCD2ad would generate rules that would be predictive in identifying the best fiveresidue enhanced aromatic sequon for glycoprotein engineering of other secreted N-glycoproteins.

library to identify the sites of N-glycosylation reveals that over 95% of the glycopeptides are found to have the NXS/T sequon, with very few having nonstandard sequons, such as NXC, NGX, and NXV.19 The NXS/T sequon is often found within a reverse turn or a loop structure in N-glycoproteins, although there is little difference between sequons that are occupied versus unoccupied in terms of the protein structures harboring them. Since sequon glycan occupancy is generally lower in α-helical structures and higher in β-sheet structures, in comparison with analogous unoccupied Asn sequons, it is tempting to speculate that secondary and/or tertiary structure nearby the Nglycosylation site might influence glycan occupancy. However, it seems unlikely that a topologically complex β-sheet structure would be significantly formed cotranslationally. In summary, around 35% of NXS/T sequons are not N-glycosylated, and thus the guiding principle of sequon glycan occupancy remains unclear. The consensus sequence for N-glycosylation in bacteria is D/ E-X1-N-X2-S/T.20 In contrast, mammalian cells do not prefer acidic residues at the i−2 position. Instead, mammalian cells appear to prefer an aromatic residue such as Phe at the i−2 position, in the context of an enhanced aromatic sequon: sequences that are more efficiently N-glycosylated by cells and which stabilize the native state of proteins that they are incorporated into as a consequence of the Phe residue interacting with the hydrophobic α-face of GlcNAc1 in appropriate reverse-turn contexts (Scheme 1; right



RESULTS Aromatic and Sulfur-Containing Residues at the i−2 Position Enhance N-Glycosylation Efficiency, whereas Positively Charged Residues Reduce Efficiency. To further investigate the influence of sequence variation on Nglycosylation efficiency in the context of a five-residue enhanced aromatic sequon, we employed the human CD2 adhesion domain to screen the effect of i−2, i−1, i+1, and i+2 position mutations flanking Asn at position i (Phe63Lys64Asn65Gly66Thr67 is the wild-type sequon). Published NMR-based structural studies reveal that the N-glycosylated Asn65 residue occupies the i position of a type I β-turn harboring a G1 β-bulge,21 also classified as a 3:5 hairpin28 (Scheme 1). This turn spans residues Phe63 (i−2 position of the β-bulge turn) to Thr67 (i+2 position of the β-bulge turn), with Gly66 occupying the i+1 or the bulge position of the βbulge turn. These NMR data show that the Phe63 phenyl side chain interacts in a face-to-face fashion with the hydrophobic αface of GlcNAc1 attached to Asn65, forming a compact structure that is stabilized primarily through the hydrophobic effect and CH−π interactions dominated by dispersion forces (Scheme 1).22 This enhanced aromatic sequon with an aromatic residue at the i−2 position has been shown to enhance the glycosylation efficiency by OST, as well as augment glycoprotein stability as a consequence of the type I β-bulge turn conformation adopted by the N-glycosylated sequence (Scheme 1).21,22,29 We do not expect the β-bulge reverse turn to be significantly populated during cotranslational N-glycosylation by OST; thus factors influencing N-glycosylation efficiency by OST and the physcical chemistry underlying native state N-glycoprotein stability are likely of different origin. Saturation site-directed mutagenesis of each of the i−2, i−1, and i+1 residues or the presence of S or T at the i +2 position of wt-hCD2ad, followed by Western blotting analysis, enables the extent of N-glycosylation to be quantified for each sequence studied (Scheme 1 and Figure 1a). The cDNA encoding residues 27−105 of hCD2ad were fused to an N-terminal FLAG-tag to aid purification. Saturation mutagenesis was first performed at the i−2 position (F63 in the wild-type sequence) of the five-residue enhanced aromatic sequon to generate hCD2ad variants for analysis of glycan occupancy (Figure 1a). N-Glycosylated and nonglycosylated hCD2ad were resolved by sodium dodecyl sulfate polyacrylamide (12%) gel electrophoresis (SDS-PAGE). Densitometry was used to quantify gels imaged by both silver staining and

Scheme 1. Strategy for Identification and Optimization of Efficiently N-Glycosylated Sequonsa

a

Saturation mutagenesis of the i−2, i−1, and i+1 positions of the enhanced aromatic sequon or limited mutagenesis at the i+2 position (S/T), followed by screening of the influence of each amino acid change on N-glycan occupancy in hCD2ad, was expected to afford a sequence that is efficiently N-glycosylated. The N-glycosylation efficiency is defined by the ratio of the N-glycosylated to the nonglycosylated sequon (G/N ratio). The structure on the right shows the CH−π interaction between the GlcNAc residue attached to Asn65 and the aromatic ring of Phe63 in the enhanced aromatic sequon in the context of a β-bulge reverse turn comprising hCD2ad (PDB code 1GYA). The side chain of Lys (position i−1) is omitted for clarity.

panel).21,22 Other studies demonstrate that aromatic side chains (Phe, Trp, Tyr, or His) at the i−2 position of the enhanced aromatic sequon can increase the N-glycosylation efficiency and also regulate the homogeneity of the glycan profile.22−27 However, the influence of the sequence on Nglycosylation efficiency in the context of a five-residue enhanced aromatic sequon remains unclear, nor do we understand why some enhanced aromatic sequons are not Nglycosylated. In this study, we investigate the influence of sequence variation on N-glycosylation efficiency in the context of a five12948

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

Figure 1. N-Glycosylation ratio of each hCD2ad variant harboring different guest amino acids at the i−2 position (amino acid 63) of the five-residue enhanced aromatic sequon. (a) The i−2 variants of hCD2ad exhibited differences in N-glycosylation efficiency. hCD2ad variants encoding different amino acids at the i−2 position were generated by saturation mutagenesis. To enhance hCD2ad protein expression levels, the DNA sequence was codon usage optimized for human cell expression. Expression plasmids of the hCD2ad variants were constructed with an N-terminal FLAG-tag for immunoprecipitation preceded by the N-terminal preprotrypsin leader sequence (PPTLS) for secretion from 293T cells followed by Western blotting analysis of the conditioned media. N65Q-hCD2ad was used as a nonglycosylation control; G, glycosylated hCD2ad; N, nonglycosylated hCD2ad. X: one of 20 amino acids. (b) Aromatic residues or S-containing residues at the i−2 position showed higher glycan occupancy. The ratio of N-glycosylated to nonglycosylated (G/N ratio) hCD2ad variant was calculated based on the relative band intensity of glycosylated hCD2ad and nonglycosylated hCD2ad. The open bars, black solid bars, and gray solid bars indicate polar uncharged, hydrophobic nonpolar, and charged amino acid residues, respectively, assuming average pKa’s. Error bars indicate standard deviations (SDs) from ≥3 replicates. Bars extending beyond the vertical black line indicate i−2 residues exhibiting higher N-glycosylation efficiency in comparison to wt-hCD2ad, whereas the dashed line indicates 50% N-glycosylation of wt-hCD2ad (*, p < 0.05; comparison with wt).

Figure 2. Identification of the optimal amino acid residue at the i−1 position for N-glycosylation in the hCD2ad five-residue enhanced aromatic sequons. (a) The i−2 position was fixed as 63F (wt), 63W (high glycosylation), and 63R (low glycosylation), and the i−1 position (aa64) was substituted with each of the 20 amino acids in each of the three contexts to determine the N-glycosylation efficiency of each i−1 mutation. Protein expression levels were analyzed and quantified by Western blotting, as described in Figure 1b, c, and d. Cys at the i−1 position improved Nglycosylation efficiency. The bars extending beyond the black line indicate higher glycosylation than found in the wt-hCD2ad variant, and the bars not reaching the dashed line indicate less than 50% glycosylation as compared to wt-hCD2ad. Error bars indicate SDs from three replicates (*, p < 0.05; **, p < 0.01; ***, p < 0.001; comparison with K at the i−1 position).

and unglycosylated hCD2ad, respectively, was confirmed by liquid chromatography tandem mass spectrometry (LC-MS/ MS). The MS/MS spectrum indicated that the glycan oxonium ions (204.1 or 366.1 Da) were observed in the spectrum of the 18−20 kDa band, but not in the spectrum of the 15 kDa band.

Western blotting as described in the Supporting Information. The molecular weight (MW) of native nonglycosylated hCD2ad carrying the FLAG-tag is 15 kDa, whereas Nglycosylated hCD2ad migrated from 18 to 20 kDa. The identification of the upper and lower bands as N-glycosylated 12949

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

Figure 3. Identification of optimal residues at the i+1 position of the five-residue sequon. (a) In the context of the 63W 64C-hCD2ad (T at i+2) sequence, a small residue at the i+1 position showed enhancement of N-glycosylation efficiency, as discerned by Western blotting (upper panel) and statistical analysis (lower panel). (b) The sequon with R/D at the i−2/i−1 positions of hCD2ad showed low N-glycosylation efficiency. The data in both (a) and (b) showed that Pro at the i+1 position was nonglycosylated. Error bars indicate SDs from more than three replicates (**, p < 0.01; ***, p < 0.001; compared to G).

was replaced by C, Y, H, or S (Figure 2c). Results to this point indicate that the best combination of residues at the i−2 and i− 1 positions of a five-residue enhanced aromatic sequon for maximal N-glycosylation is W and C, respectively, and for poor N-glycosylation, R and D, respectively. Hydroxyl or Small Amino Acids at the i+1 Position Enhance Glycan Occupancy. The highly N-glycosylated W63 C64-hCD2ad sequence was utilized to identify which residue at position 66 (the i+1 position) would reduce or increase N-glycosylation efficiency. The poorly N-glycosylated R63 D64-hCD2ad variant was used to identify which i+1 residue would restore the N-glycosylation efficiency (Figure 3a). Most W63 C64 X66-hCD2ad variants exhibited high glycosylation efficiency, with the exceptions being a W residue at the i+1 position, which significantly reduced the Nglycosylation, and more dramatically a P at the i+1 position, which prevented N-glycosylation (Figure 3a, upper panel). Most R63 D64 X66-hCD2ad variants showed poor glycosylation, and no glycosylation was observed when P or W occupied the i+1 position (Figure 3b). Statistical analysis showed that a Q, S, G, or A residue in the i+1 position of the W63 C64-hCD2ad five-residue enhanced aromatic sequon enhanced N-glycosylation efficiency. An S or T occupying the i +1 position of the R63 D64-hCD2ad sequence could partially reverse the poor glycosylation associated with this five-residue sequon (Figure 3a and b lower panel). Effect of the i+2 Position on Glycosylation Efficiency. We next utilized the five-residue FKNGT/S and WCNGT/S sequons in hCD2ad to investigate the effect of the i+2 position on N-glycosylation. Not surprisingly, the result showed that Thr dramatically increased the N-glycosylation ratio compared to Ser in both FKNGX and WCNGX five-residue enhanced

We employed the N65Q hCD2ad mutant that cannot be glycosylated as a negative control, which exhibited a MW of 15 kDa exclusively (Figure 1a). Western blotting also revealed the relative level of protein expression of each hCD2ad variant purified from conditioned media after substitution of 20 different amino acids at the i−2 position (Figure 1a). The Nglycosylation efficiency (G/N ratio) was calculated based on the band intensity of N-glycosylated hCD2ad divided by that of nonglycosylated hCD2ad analyzed statistically (Figure 1b). The aromatic residues (Y, W, and F) and the sulfurcontaining residues (C and M) at the i−2 position exhibit the highest N-glycosylation efficiency. In contrast, charged residues (e.g., R and K) substantially reduced the N-glycosylation efficiency. The top four hCD2ad i−2 variants ranked in order, each with more than 4.5-fold enhancement in N-glycosylation efficiency as compared to wt-hCD2ad (i−2 residue = F) are W, C, M, and Y (Supporting Information; Table 1). A G/N ratio of less than 2 was defined as poor N-glycosylation. Cys at the i−1 Position Increases N-Glycosylation Efficiency. To evaluate the influence of the i−1 residue on Nglycosylation efficiency, the i−2 residue was fixed to one of three residues: Phe (wt-hCD2ad), Trp63 (high N-glycosylation ratio), or Arg63 (low N-glycosylation ratio). Saturation mutagenesis at the i−1 position allowed comparisons in these three contexts (Figure 2a). The N-glycosylation efficiency of each of the 60 combinations was analyzed by Western blotting, using wt-hCD2ad as a reference (Figure S1). The statistical analysis showed that Cys at the i−1 position significantly increased N-glycosylation efficiency in comparison to wthCD2ad, which harbors a Lys residue at the i−1 position (Figure 2b, K64). Analysis of each variant showed that a higher glycosylation efficiency was observed when K64 in wt-hCD2ad 12950

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

enhancing N-glycosylation occupancy when at the i+1 position. Furthermore, the chemical properties of the 20 amino acids were further divided into eight groups (Figure S3d). Aromatic and sulfur-based side chains enhanced N-glycan occupancy more so than other groups at the i−2 position, and the sulfurcontaining group was the highest group and aliphatic groups exhibited above average ratios at the i−1 position. The hydroxyl group showed an enhanced N-glycosylation ratio at both i+1 OS and i+1 PS. The following equation is our predictive algorithm: Predicted glycosylation ratio = −0.5070 + 3.9373 (if POU residues at the i+1 position) + 3.6061 (if POU residues at the i−1 position) + 3.0451 (if residue at the i−1 position is Cys, Met, Ser, Thr, Ala, Val, Ile, Leu, or Gly) + 1.3292 (if nonpositive residues at the i−2 position) + 2.1615 (if residue at the i−2 position is Phe, Tyr, Trp, Cys, or Met) POU residues include Gly, Tyr, Asn, Ser, Cys, Gln, and Thr; nonpositive residues are those except for His, Arg, and Lys. Although the peptide hydrophobicity was significantly associated with N-glycosylation efficiency in the univariate regression analysis, this variable was no longer statistically significant in predicting the glycosylation ratio when added to the above equation (data not shown). In the internal validation test, the N-glycosylation efficiency predicted by the model correlated well with the observed N-glycosylation ratio (Pearson correlation coefficient R = 0.78) (Figure 5a). To

aromatic sequons (Figure 4a and b). Furthermore, the WCNGS sequon showed a higher N-glycosylation ratio than

Figure 4. Thr at the i+2 position is better than Ser for N-glycosylation. (a) Thr at the i+2 position of hCD2ad showed higher glycan occupancy. hCD2ad variants with FKNGT (wt), FKNGS, WCNGS, and WCNGT sequons were examined as described already. (b) Residues at the i−2/i−1 positions of hCD2ad played an important role in N-glycosylation efficiency, and all sequences revealed that T is better than S at the i+2 position. Error bars indicate SDs from more than three replicates (***, p < 0.001; compared to wt; #, p < 0.05 compared to the FKNGS sequon).

the FKNGS sequon, indicating that a proper combination of residues at the i−2 and i−1 positions affords a substantial improvement in glycan occupancy (Figure 4b). Based on prior results, it is not surprising that S or T at the i+2 position strongly influences N-glycosylation efficency, even when the five-residue enhanced aromatic sequon harbors W and C at the i−2 and i−1 (63 and 64) positions. Even in the context of an optimized enhanced aromatic sequon, the lack of S or T at the i +2 position (WCNGA) results in undetectable N-glycosylation (Figure S2). In summary, this systematic screening of the peptapeptide enhanced aromatic sequon sequence space has identified optimized, poor, and nonglycosylatable sequons for N-glycosylation of hCD2ad. Equation to Predict Five-Residue Enhanced Aromatic Sequon N-Glycosylation Efficiency. To develop a fiveresidue enhanced aromatic sequon N-glycosylation occupancy predictor, we utilized various parameters to classify neighboring residues of the N-glycosylated Asn residue. We focused on 120 hCD2ad five-residue sequon variants to create a database for deep analysis. The five-residue sequons were classified into i−2, i−1, i+1 (optimized sequence (OS) WCNXT) and i+1 (poor sequence (PS) RDNXT) to analyze different properties at each site. The hydrophobicity of the five-residue sequons was predicted by the Sequence Specific Retention Calculator. Results showed a positive correlation between peptide hydrophobicity and N-glycosylation occupancy at the i−2, i− 1, and i+1 positions (Figure S3a). The molecular weight of each of 20 residues at different sites was classified into four categories (I, II, III, IV). Small residues at i+1 OS (category II) improved the N-glycosylation ratio, whereas small residues at other positions did not enhance sequon occupancy (Figure S3b). Large residues at the i−1 and i+1 positions were not favored for improvement of N-glycosylation. In addition, the 20 residues were classified into four groups by their charge characteristics: negatively charged residues in black, positively charged residues in red, polar uncharged (POU) residues in white, and hydrophobic nonpolar (HNP) residues in gray (Figure S3c). As shown, POU and HNP amino acids at the i−2 position were better for N-glycosylation occupancy than charged residues, POU residues obviously showed a higher N-glycosylation occupancy than other groups at the i−1 position, and POU residues also showed a preference for

Figure 5. Correlations between predicted N-glycosylation efficiency estimated by the predictive model and the observed N-glycosylation efficiency for various hCD2 five-residue sequons. (a) Scatter plot of algorithm predicted vs observed G/N ratio for 118 five-residue sequon variants that were used to generate the model. The Pearson’s correlation coefficient R = 0.78. (b) Scatter plot of algorithm predicted vs observed glycosylation ratios in 10 additionally designed hCD2ad variants that were not used for model derivation (R = 0.895).

further validate the prediction model, a total of 10 other acceptors that were not included in the hCD2ad library were designed and the N-glycosylation ratios were determined (Figure S4). As shown in Figure 5b, the predicted glycosylation ratios correlated well with observed ones (R = 0.895), which confirmed the accuracy and reliability of the prediction model in predicting N-glycosylation efficiency. Extension of the Predictive Sequons to Other NGlycoproteins. According to the five-residue model developed for predicting N-glycosylation efficiency in hCD2ad, we predicted that BCMA, a small glycoprotein containing one glycosylation site at Asn42, should have a high probability of Nglycosylation. The five-residue sequon of BCMA, YCNAS, containing an aromatic residue at the i−2 position (+2.1615 and +1.3292), a sulfur residue at i−1 (+3.6061 and +3.0451), and non-POU at i+1 (+0) should be a good acceptor for Nglycosylation (total value: +9.6349). However, BCMA exhibited poor N-glycosylation efficiency, even when the i−2 position was changed to W, affording the best two-residue combination (WC) at the i−2 and i−1 positions (Figure 6a). It 12951

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

+11.4107, respectively). Glycosylation of the former is related to protease resistance and secretion and of the latter to protein dimerization.32,33 Western blotting showed three types of IFNγ, distinguished by their molecular weight as lower (no glycan), middle (glycan at one site), and higher (glycan at both sites) bands (Figure 6c). After replacing the 23rd/24th residues with the optimized sequon residues W/C, the product showed a decrease of the lower band, indicating improvement of Nglycosylation efficiency. However, when the optimized amino acids were installed into the 95th/96th amino acids of IFN-γ as W and C, there was marked inhibition of protein expression (media detection). So placing the optimized sequon (WCNGT) at both sites eliminated the middle and lower bands of IFN-γ; however, the protein expression level was decreased (Figure 6c). Analysis of total cell lysate demonstrated increased N-glycosylation of intracellular IFN-γ variants after replacement at 25N, 97N, and 25N/97N with optimized sequences. According to these results, even if the sequon is optimized for OST, the local protein structure may change through disulfide formation and influence its glycosylation or the protein stability may change, leading to intracellular degradation by the proteasome or autophagy. Computer Modeling of the Optimized N-Glycosylation Sequon and a Human OST Subunit. To understand the mechanism of peptide preferences for OST-catalyzed Nglycosylation on hCD2ad, a theoretical structural model of human OST subunit STT3A was built based on the structure of bacterial OST PglB, as described in the experimental procedures. According to the complex model (Figure 7a), the atoms OD2 of Asp49, OD1 and OD2 of Asp167, and OE2 of Glu351 interact with the Mg2+ ion, and the residues Asp49 and Glu351 could form hydrogen bonds with the amide group of the ligand Asn(0), resulting in the carboxamide twisted activation of asparagine as previously proposed.34 As illustrated by the model (Figure 7b), the i−2 pocket is a deep gorge encompassed by Val404, Met407, Pro155, Tyr483, and Arg 160. The two major interactions that the ligand side chain could have in the pocket are a cation−π interaction with Arg160 and an aromatic−π interaction with Tyr 483. Therefore, aromatic amino acids, such as Trp, are favored at the i−2 positon. The negative charge of the carboxyl group in the i−1 pocket is due to Glu47, so Asp at the i−1 position is not favored, perhaps due to electrostatic repulsion. On the other hand, Cys and Ser are better residues for a ligand to position well in the i−1 pocket. The i pocket is a burrow, such that the ligand for Asn glycosylation could point to the catalytic site as described above (Figure 7a). The i+1 position has a narrow opening pinched by Phe48 and Ser350, such that Gly and Ala are favored, but Trp or Pro is not due to the increased steric hindrance caused by the bulky or rigid side chain. Although Asp527 in the pocket could provide hydrogen bonding to Ser and Thr at the i+2 position, the hydrophobic gorge encompassed by Val349, Trp525, and Trp526 does favor Thr at the i+2 position. In line with the experimental data, we concluded that, whereas the side chain of most amino acids can be accommodated in the i−2 and i−1 positions, Trp at i−2 and Cys at the i−1 position would favor binding of peptide in the catalytic sites, favoring activation of N-glycosylation. In contrast, Arg at i−2 and Asp at the i−1 position would be the least favored arrangement in the sequon, probably due to the steric hindrance of Arg in the i−2 pocket (Figure 7c) and electrostatic repulsion caused by the carboxyl group of Asp at the i−1 position.

Figure 6. Incorporation of optimal glycosylation sequons into other glycoproteins. (a) BCMA variants were expressed in 293T cells, and the cell lysate was analyzed by Western blot. The endogenous fiveresidue N-glycosylation sequon, YCNAS, was replaced by WCNGT and other sequences for comparison. (b) Bar graph showing the statistical analysis and comparison of the N-glycosylation ratio between BCMA variants. (c) IFN-γ contains two N-glycosylation sites, 25N/97N, exhibiting three glycosylation states: nonglycosylated (N), glycosylation at one site (G), and glycosylation at two sites (GG). The optimized sequence WCNGT was installed into the 25N/97N sites to examine the N-glycosylation efficiency. Because the protein expression levels of IFN-γ variants were different, total input of (1) was collected for IP: wt (0.05), WCNGT-25N (0.05), −97N (0.5), −25N/97N (0.5), and the whole lysate of each variant was loaded. Error bars indicate SDs from more than three replicates (**, p < 0.01; ***, p < 0.001).

is reported that BCMA contains three disulfide bonds between amino acid residues 8 and 21, 24 and 37, and 28 and 41.30 We assumed that the disulfide bond formation could influence local conformation and, thereby, the N-glycosylation efficiency. Moreover, the disulfide bond between Cys28 and Cys41 is next to the glycosylation site at Asn42. We therefore mutated Cys28 to Ser to break the disulfide formation nearest the Nglycosylation site. C28S-BCMA exhibited a significant increase in glycan occupancy, and a similar pattern was observed after installing the optimized sequon “WCNGT” in BCMA without the C28S mutation (Figure 6b). Furthermore, WCNGTBCMA with the C28S mutation showed the highest Nglycosylation ratio compared with other BCMA variants. The three variants of BCMA, labeled No. 4, 5, and 6, with no disulfide effect exhibited predicted values of +9.6449, +13.5722, and −0.507, respectively, and also showed a similar pattern of N-glycosylation efficiency in comparison with the model (Figure 6b). The poor sequon, RD, at the i−2/i−1 position of BCMA still showed a lower G/N ratio, even when the disulfide effect was eliminated. This result demonstrated that the WCNGT sequon is highly accepted as an OST substrate before Cys28−Cys41 disulfide bond formation. We also examined the disulfide formation in wild-type BCMA by mass spectrometry and found that BCMA with or without Nglycosylation appeared to have three disulfide bonds (data not shown). In addition, interferon gamma (IFN-γ), an FDAapproved biological, contained two glycosylation sites at Asn25 and Asn9731 and the five-residue sequons of these two sites are ADNGT and LTNYS (model predicted value: +4.7595 and 12952

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

Figure 7. Predicted catalytic site of human OST with peptide ligands. (a) The red ribbon represents the ligand for Asn glycosylation. The theoretical structural model of human glycosyltransferase subunit STT3A (dolichyl-diphosphooligosaccharide, STT3A_HUMAN P46977) was built by SWISSMODEL SERVER (https://swissmodel.expasy.org). (b) Ligand-binding pocket of human OST (STT3A) and the predicted interactions of the residues with the ligand WCNGT. (c) The steric hindrance of Arg at the i−2 position (left) could lead to an improper binding of the residue at i−1 and activation position of Asn compared to the optimized ligand (right).



DISCUSSION N-Glycosylation is an important co- and/or post-translational modification used by eukaryotes to enhance the in vivo activity of proteins, and it enables their folding through a dedicated lectin-based proteostasis network. N-Glycosylation also inhibits the rapid clearance of N-glycoproteins. For example, recombinant human erythropoietin (rHuEPO), containing three Nglycosylation sites and one O-glycosylation site, has been engineered to harbor two additional N-glycosylation sites at N14/18 (darbepoetin alfa) to improve its half-life and activity in animal studies35 without eliciting an immune response in clinical trials.36,37 In addition, leptin, a nonglycosylated protein involved in the control of body weight,38 has been constructed with five additional N-linked glycosylation sites. This variant was shown to reduce body weight to a greater extent in mice in comparison with nonglycosylated leptin.35 However, these biologics with designed N-glycosylation sequons to alter Nglycosylation efficiency could face another challenge. Someday the regulatory agencies may require the manufacturer to sell homogeneous glycoforms. Herein, we assessed the hypothesis that optimizing the enhanced aromatic sequon glycan occupancy for hCD2ad would generate rules that would be predictive for other secreted N-glycoproteins. We found that optimal five-residue hCD2ad enhanced aromatic sequons exhibited different side chain amino acid preferences at the i−2, i−1, and i+1 positions. As shown in the results, while the i−2 position of hCD2ad appeared to prefer aromatic residues (Trp, Phe, and Tyr) or sulfur-containing residues (Cys and Met), the i−1 position prefers polar uncharged residues and the i+1 position favors hydroxyl- or amide-containing side chains for enhanced Nglycosylation. The Cys residue at the i−1 position obviously

enhanced the efficiency of N-glycosylation. A charged residue at the i−2 or i−1 position or side chains composed of an aromatic or imino functionality at the i+1 position of the five-residue sequon showed N-glycosylation suppression. We further analyzed the frequency of the five-residue sequon in a glycopeptide database (Figure S5) that contains experimentally verified N-linked glycosylation sites on human proteins. Detailed analysis showed that L/V/G/S/I/A residues appear with greater than 50% frequency at the i+1 position (G/S/A residues also appear as optimal in our study, Figure 3), L/G/S/ A/V/T residues appear with 40% frequency at the i−1 position (these residues are also optimal in our study, Figure 2), and L/ A/S/V/G/F residues appear with 40% frequency at the i−2 position (only the F residue appears as optimal in our study, Figure 1). Although there was some overlap with our data serving as a predictive model, the N-glycan occupancy still cannot be explained through frequency analysis. Previous studies had identified the consensus sequon for Nglycosylation through mass spectrometry, analysis of databases, and glycoengineering of model proteins. The first two methods could analyze only the standard consensus sequence, NXS/T, and the frequency of residues near the glycosylated Asn of glycoproteins.18,19 Studies on the N-glycosylation efficiency of sequons with variations at the i−2 and i+1 sites individually have shown that aromatic residues at the i−2 position of the Nglycosylation sequon of hCD2ad would enhance N-glycosylation.25 In another study that focused on the i+1 position of rabies virus glycoprotein, all 20 amino acids were studied at the i+1 position. A fully glycosylated sequon was observed with the G/R/H/S variants, whereas inefficient glycosylation was exhibited by the W/D/E/L variants.39 Libraries of glycosylated peptides created by site-directed mutagenesis at i−2, i−1, and i 12953

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

not known). If the co- and post-translational N-glycosylation ratio differs from protein to protein, which seems likely, then for the hCD2ad results to be predictive for what would occur in other proteins, the sequence preferences and the conformation of the enhanced sequon sequence in OST harboring the STT3A catalytic subunit and OST harboring the STT3B catalytic subunit would have to be similar (which is not known). The mechanism of native state stabilization by Nglycosylation and the mechanism by which the OST(s) efficiency is altered by the exact sequence being N-glycosylated are distinct. The optimized enhanced aromatic sequon of hCD2ad appears to be recognized in a largely extended conformation by OST harboring the STT3A catalytic subunit based on the model described in Figure 7. Thus, it is reasonable to hypothesize that the enhanced aromatic sequon increases Nglycosylation efficiency by modulating the reaction catalyzed by the OST enzyme(s) on the sequon substrate, i.e., by increasing kcat/Km of OST(s). In other words by a different mechanism than the native state stabilizing CH−π interaction. We recognize that some N-glycosylation of hCD2ad may occur post-translationally by OST comprising the STT3B catalytic subunit. It is unclear how the sequence and conformational preferences of the enhanced aromatic sequon affect the balance between co- and post-translational N-glycosylation, which may be the reason that no one protein is a perfect predictor for the optimized enhanced aromatic sequon for all N-glycoproteins. The N-glycosylated five-residue enhanced aromatic sequon within hCD2ad is harbored within a β-bulge turn in the native state. The N-glycosylated sites in BCMA and IFN-γ are in an αhelix and coil region in the native conformations, respectively (Scheme 1 and Figure S8). Since we do not know the preferred conformations of the co- and post-translational OSTs, it is not clear what relevance, if any, the native conformations play in Nglycosylation efficiency. We also examined the glycans of the hCD2ad variants (Figure S9). The major glycoforms on hCD2ad variants expressed by 293T cells were bisecting glycans with terminal sialic acid, fucose, and lactose (Figure S9a), similar to those found on hCD2ad variants expressed by 293 cells (Figure S9b). Although N-glycosylation efficiency can be improved by installing the optimized sequon to the glycoprotein of interest, the glycan compositions and sequences should be further examined, as they may change as a function of the enhanced aromatic sequon. In summary, we used site-directed saturation mutagenesis at the i−2, i−1, or i+1 positions of the N-glycosylated enhanced aromatic sequon of hCD2ad to identify sequons with improved or suppressed N-glycosylation efficiency, to develop a predictive model for N-glycosylation efficiency. Although the stepwise approach used to generate this predictive model was not based on all combinations of five-residue sequons, this predictive model showed its usefulness in forecasting Nglycosylation efficiency, successfully in the two examples pursued (Figure S10). Finally, the modeling of STT3A structurally rationalized why some sequons suppress Nglycosylation while others enhance N-glycosylation. Although this predictive model of N-glycosylation efficiency is developed based on a limited number of example sequences in one protein (hCD2ad) and may not be generally applicable to all proteins for the reasons described above, it could be useful for the design of N-glycoproteins as pharmaceuticals. Thus, using this predictive model as a design tool will scrutinize its utility over

+1 positions demonstrated that the optimal acceptor was PYNVTK for the single-subunit OST of Pyrococcus f uriosus.40 However, they only replaced the i−2/i−1 positions with several amino acids and paired these with 20 amino acids at the i+1 position to identify sequons with improved N-glycosylation efficiency. Another group used an in vitro translation/ glycosylation system to examine the residue at i−1 and showed that when the i−1 position was occupied by G/L/S/F/C, Nglycosylation efficiency was high, compared to our model as C/ G/Y/S/H.41 Since nonhuman cells produce biologics with different N-glycans, which can cause an immune response, also influencing the potency of the products,10,42,43 we used human cells in our study of N-glycoprotein production in order to identify the five-residue enhanced aromatic sequon preferences for the human OST(s). We assessed the N-glycan occupancy of 120 hCD2ad variants to identify the sequences exhibiting the best N-glycan occupancy; we did not consider unconventional sequons (NXC and NXV as shown in Figure S6). We used this data to create a predictive algorithm referred to as the SAS program. We then extended this model to predict the glycosylation of additional glycoproteins. Because changes of N-glycosylation by gene mutation have been shown to cause an unexpected pathogenesis,43,44 our N-glycosylation prediction model could be used to predict whether a mutation around the glycosylated Asn residue would change the N-glycosylation efficiency. We examined the ratio of secreted N-glycosylated to secreted nonglycosylated hCD2ad or IFN-γ as a function of various combinations of five-residue enhanced aromatic sequons, to discern whether this effort was useful for glycoengineering, i.e., for maintaining the protein expression yield while enhancing Nglycosylation efficiency. After the disulfide bond proximal to the N-glycosylation site in BCMA was removed, the optimized enhanced aromatic sequon from hCD2ad increased the efficiency of BCMA N-glycosylation in the cell lysate, while maintaining the expression level. Although the exact sequence of events could differ for different N-glycoproteins, both Nglycosylation by OST(s) and disulfide bond formation mediated by the oxidizing environment and the disulfide isomerases occur in parallel in the endoplasmic reticulum (ER), and disulfide bond formation will likely more strongly influence N-glycosylation efficiency post-translationally than cotranslationally. Even though the optimized enhanced aromatic sequon increased the glycosylation efficiency in IFN-γ, the secretion yield dropped considerably. Although the optimized sequon improved the N-glycosylation efficiency of the hCD2ad variants, the protein expression level of these variants showed variability (Figure S7), indicating that N-glycosylation efficiency is not the only issue to be addressed. Protein stability could also be influenced by sequon optimization, which can influence the ER quality control decision in the ER. A limitation of our study is that it was not designed to analyze the pathway to the final secreted N-glycosylated protein; for example, the extent of intracellular degradation of N-glycosylated and nonglycosylated proteins (i.e., quality control) was not assessed. Whether the optimized five-residue sequence affording maximal N-glycosylation of hCD2ad will do so for other N-glycoproteins generally is an interesting open question. For the optimized sequence from hCD2ad to be predictive for what will happen in other N-glycoproteins, the ratio of co- and post-translational Nglycosylation by OST/STT3A and OST/STT3B, respectively, probably needs to be similar, assuming these two OSTs have different sequence preferences for N-glycosylation (which is 12954

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955

Article

Journal of the American Chemical Society

(20) Kowarik, M.; Young, N. M.; Numao, S.; Schulz, B. L.; Hug, I.; Callewaert, N.; Mills, D. C.; Watson, D. C.; Hernandez, M.; Kelly, J. F.; Wacker, M.; Aebi, M. EMBO J. 2006, 25, 1957. (21) Culyba, E. K.; Price, J. L.; Hanson, S. R.; Dhar, A.; Wong, C. H.; Gruebele, M.; Powers, E. T.; Kelly, J. W. Science 2011, 331, 571. (22) Chen, W.; Enck, S.; Price, J. L.; Powers, D. L.; Powers, E. T.; Wong, C. H.; Dyson, H. J.; Kelly, J. W. J. Am. Chem. Soc. 2013, 135, 9877. (23) Price, J. L.; Culyba, E. K.; Chen, W.; Murray, A. N.; Hanson, S. R.; Wong, C. H.; Powers, E. T.; Kelly, J. W. Biopolymers 2012, 98, 195. (24) Price, J. L.; Powers, D. L.; Powers, E. T.; Kelly, J. W. Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 14127. (25) Murray, A. N.; Chen, W.; Antonopoulos, A.; Hanson, S. R.; Wiseman, R. L.; Dell, A.; Haslam, S. M.; Powers, D. L.; Powers, E. T.; Kelly, J. W. Chem. Biol. 2015, 22, 1052. (26) Aguila, S.; Martinez-Martinez, I.; Dichiara, G.; GutierrezGallego, R.; Navarro-Fernandez, J.; Vicente, V.; Corral, J. PLoS One 2014, 9, e114454. (27) Hang, I.; Lin, C. W.; Fleurkens, S.; Villiger, T. K.; Soos, M.; Morbidelli, M.; Woods, R. J.; Gauss, R.; Aebi, M. Glycobiology 2015, 12, 1335. (28) Sibanda, B. L.; Blundell, T. L.; Thornton, J. M. J. Mol. Biol. 1989, 206, 759. (29) Hsu, C. H.; Park, S.; Mortenson, D. E.; Foley, B. L.; Wang, X.; Woods, R. J.; Case, D. A.; Powers, E. T.; Wong, C. H.; Dyson, H. J.; Kelly, J. W. J. Am. Chem. Soc. 2016, 138, 7636. (30) Hymowitz, S. G.; Patel, D. R.; Wallweber, H. J.; Runyon, S.; Yan, M.; Yin, J.; Shriver, S. K.; Gordon, N. C.; Pan, B.; Skelton, N. J.; Kelley, R. F.; Starovasnik, M. A. J. Biol. Chem. 2005, 280, 7218. (31) Rinderknecht, E.; O’Connor, B. H.; Rodriguez, H. J. Biol. Chem. 1984, 259, 6790. (32) Sareneva, T.; Pirhonen, J.; Cantell, K.; Julkunen, I. Biochem. J. 1995, 308, 9. (33) Sareneva, T.; Pirhonen, J.; Cantell, K.; Kalkkinen, N.; Julkunen, I. Biochem. J. 1994, 303, 831. (34) Lizak, C.; Gerber, S.; Michaud, G.; Schubert, M.; Fan, Y. Y.; Bucher, M.; Darbre, T.; Aebi, M.; Reymond, J. L.; Locher, K. P. Nat. Commun. 2013, 4, 2627. (35) Elliott, S.; Lorenzini, T.; Asher, S.; Aoki, K.; Brankow, D.; Buck, L.; Busse, L.; Chang, D.; Fuller, J.; Grant, J.; Hernday, N.; Hokum, M.; Hu, S.; Knudten, A.; Levin, N.; Komorowski, R.; Martin, F.; Navarro, R.; Osslund, T.; Rogers, G.; Rogers, N.; Trail, G.; Egrie, J. Nat. Biotechnol. 2003, 21, 414. (36) Glaspy, J.; Jadeja, J. S.; Justice, G.; Kessler, J.; Richards, D.; Schwartzberg, L.; Rigas, J.; Kuter, D.; Harmon, D.; Prow, D.; Demetri, G.; Gordon, D.; Arseneau, J.; Saven, A.; Hynes, H.; Boccia, R.; O’Byrne, J.; Colowick, A. B. Br. J. Cancer 2001, 84, 17. (37) Vansteenkiste, J.; Pirker, R.; Massuti, B.; Barata, F.; Font, A.; Fiegl, M.; Siena, S.; Gateley, J.; Tomita, D.; Colowick, A. B.; Musil, J. J. Natl. Cancer Inst. 2002, 94, 1211. (38) Brennan, A. M.; Mantzoros, C. S. Nat. Clin. Pract. Endocrinol. Metab. 2006, 2, 318. (39) Kasturi, L.; Chen, H.; Shakin-Eshleman, S. H. Biochem. J. 1997, 323, 415. (40) Igura, M.; Kohda, D. Glycobiology 2011, 21, 575. (41) Bano-Polo, M.; Baldin, F.; Tamborero, S.; Marti-Renom, M. A.; Mingarro, I. Protein sci 2011, 20, 179. (42) Larkin, A.; Imperiali, B. Biochemistry 2011, 50, 4411. (43) Ghaderi, D.; Taylor, R. E.; Padler-Karavani, V.; Diaz, S.; Varki, A. Nat. Biotechnol. 2010, 28, 863. (44) Vogt, G.; Chapgier, A.; Yang, K.; Chuzhanova, N.; Feinberg, J.; Fieschi, C.; Boisson-Dupuis, S.; Alcais, A.; Filipe-Santos, O.; Bustamante, J.; de Beaucoudrey, L.; Al-Mohsen, I.; Al-Hajjar, S.; AlGhonaium, A.; Adimi, P.; Mirsaeidi, M.; Khalilzadeh, S.; Rosenzweig, S.; de la Calle Martin, O.; Bauer, T. R.; Puck, J. M.; Ochs, H. D.; Furthner, D.; Engelhorn, C.; Belohradsky, B.; Mansouri, D.; Holland, S. M.; Schreiber, R. D.; Abel, L.; Cooper, D. N.; Soudais, C.; Casanova, J. L. Nat. Genet. 2005, 37, 692.

time as the experimental N-glycosylation efficiencies are reported.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/jacs.7b03868. Supplementary data and experimental procedures (PDF)



AUTHOR INFORMATION

Corresponding Authors

*[email protected] *[email protected] ORCID

Jeffery W. Kelly: 0000-0001-8943-3395 Chi-Huey Wong: 0000-0002-9961-7865 Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We thank Ms. M.-H. Pan for technical assistance in SAS analysis, Dr. H.-W. Huang for the BCMA construct, Mr. C.-H. Chen for analysis of the glycan profile along with the Mass Spectrometry Core Facility at the Genomics Research Center, Academia Sinica, and Ms. C.-S. Suen for analyzing the frequency of the five-residue sequon from the glycopeptide database at the Institute of Biomedical Science, Academia Sinica. This research was supported by the National Science Council and Academia Sinica, Taiwan, and by NIH GM051105 (J.W.K.).



REFERENCES

(1) Fuster, M. M.; Esko, J. D. Nat. Rev. Cancer 2005, 5, 526. (2) Hauselmann, I.; Borsig, L. Front. Oncol. 2014, 4, 28. (3) Hanson, S. R.; Culyba, E. K.; Hsu, T. L.; Wong, C. H.; Kelly, J. W.; Powers, E. T. Proc. Natl. Acad. Sci. U. S. A. 2009, 106, 3131. (4) Moyle, W. R.; Lin, W.; Myers, R. V.; Cao, D.; Kerrigan, J. E.; Bernard, M. P. Endocr. J. 2005, 26, 189. (5) Takahashi, M.; Yokoe, S.; Asahi, M.; Lee, S. H.; Li, W.; Osumi, D.; Miyoshi, E.; Taniguchi, N. Biochim. Biophys. Acta, Gen. Subj. 2008, 1780, 520. (6) Liu, Y. C.; Yen, H. Y.; Chen, C. Y.; Chen, C. H.; Cheng, P. F.; Juan, Y. H.; Chen, C. H.; Khoo, K. H.; Yu, C. J.; Yang, P. C.; Hsu, T. L.; Wong, C. H. Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 11332. (7) Kornfeld, R.; Kornfeld, S. Annu. Rev. Biochem. 1985, 54, 631. (8) Gavel, Y.; von Heijne, G. Protein Eng., Des. Sel. 1990, 3, 433. (9) Helenius, A.; Aebi, M. Annu. Rev. Biochem. 2004, 73, 1019. (10) Schwarz, F.; Aebi, M. Curr. Opin. Struct. Biol. 2011, 21, 576. (11) Parodi, A. J. Annu. Rev. Biochem. 2000, 69, 69. (12) Parodi, A. J. Biochem. J. 2000, 348, 1. (13) High, S.; Lecomte, F. J.; Russell, S. J.; Abell, B. M.; Oliver, J. D. FEBS Lett. 2000, 476, 38. (14) Hebert, D. N.; Lamriben, L.; Powers, E. T.; Kelly, J. W. Nat. Chem. Biol. 2014, 10, 902. (15) Petrescu, S. M.; Branza-Nichita, N.; Negroiu, G.; Petrescu, A. J.; Dwek, R. A. Biochemistry 2000, 39, 5229. (16) Wormald, M. R.; Petrescu, A. J.; Pao, Y. L.; Glithero, A.; Elliott, T.; Dwek, R. A. Chem. Rev. 2002, 102, 371. (17) Xu, C.; Ng, D. T. Nat. Rev. Mol. Cell Biol. 2015, 16, 742. (18) Petrescu, A. J.; Milac, A. L.; Petrescu, S. M.; Dwek, R. A.; Wormald, M. R. Glycobiology 2004, 14, 103. (19) Zielinska, D. F.; Gnad, F.; Wisniewski, J. R.; Mann, M. Cell 2010, 141, 897. 12955

DOI: 10.1021/jacs.7b03868 J. Am. Chem. Soc. 2017, 139, 12947−12955