To create the E. coli expression constructs, portions of the GI protein coding region were amplified by PCR using the primers listed in Supplementary Table S1 and cloned between the BamHI and NotI sites of pET32a. This region of GI was produced with N-terminal S- and His-tags. To create the pET32-CT5, the region encoding GI amino acids 789–893 (GenBank: AAT80910.1) ( Supplementary Figure S1 ) was cloned between the NcoI and XhoI sites of pET32a. The protein expressed from this construct was named CT5. Site-directed mutagenesis of pET32-CT5 was performed using the QuikChange Site-Directed Mutagenesis Kit following the instructions of the manufacturer (Stratagene, La Jolla, CA, USA) using the primers listed in Supplementary Table S1 .

SEC and different portions of GI were co-expressed for 2 h in BL21-AI essentially as described previously (Scott et al., 2006). Cells were harvested from 1.5 mL of culture by centrifugation and resuspended in 50 µL of phosphate-buffered saline (PBS) and 17 µL of 4× sodium dodecyl sulfate (SDS) gel-loading buffer. The cells were lysed by boiling for 5 min, and the sample proteins were resolved using SDS-PAGE, blotted to Immobilon-P membrane (Millipore, Bedford, MA, USA), and the blots probed to detect the GlcNAc-modified proteins using a radioactive galactosyl transferase assay that capped GlcNAc with ³H-galactose (Scott et al., 2006). To detect labeled proteins, the membrane was sprayed with EN³HANCE (Perkin-Elmer Life Science, Boston, MA, USA), air-dried, and exposed to a pre-flashed BioMax XAR film (Eastman Kodak, Rochester, NY, USA) at −80°C. S-tagged proteins were detected on duplicate blots using horseradish peroxidase-conjugated anti-S antibodies (Novagen, Madison, WI, USA) as recommended by the supplier. Horseradish peroxidase was detected using Super Signal West Pico (Pierce, Rockford, IL, USA) with exposure to X-OMAT™ Blue XB-1 film (Eastman Kodak, Rochester, NY, USA).

The purified O-GlcNAc-modified CT5 was prepared as described previously (Kim et al., 2011). SEC and CT5 were co-expressed from pET32-CT5 and pACYC-Mal-SEC in E. coli BL21-AI. Of a Luria–Bertani (LB) medium, 500 mL was inoculated with 10 mL of an overnight culture and grown at 22°C. Arabinose was added to 0.2% (w/v) when the culture reached an optical density (OD₆₀₀) of 0.4. After 1 h, isopropyl-β-d-thiogalactoside was added to 1 mM and the cells harvested after 2 h by centrifugation. The cells were resuspended in 20 mL of 50 mM sodium phosphate and 500 mM sodium chloride (pH 8.0) and broken using a French press (Milton Roy, Ivyland, PA, USA) at 10,000 psi. CT5 was purified using the ProBond Purification System (Invitrogen Life Technologies, Carlsbad, CA, USA) following the manufacturer’s instructions. The purified proteins were concentrated using an Amicon Ultra-15 (Millipore Corporation, Billerica, MA, USA). The purified CT5 was then reduced in 2 mM Tris-(2-carboxyethyl)phosphine at room temperature for 1 h and alkylated in 10 mM iodoacetamide in the dark at room temperature for 1.5 h. The alkylated CT5 was digested with MS grade trypsin (40:1; Promega, Madison, WI, USA) or endoproteinase Lys-C (100:1; Roche, Penzberg, Germany) overnight at 37°C, desalted using a Sep-Pak C₁₈ cartridge (Waters, Milford, MA, USA), and then dried using a Speed-Vac. GlcNAc was capped with galactose using galactosyl transferase, creating the disaccharide N-acetyl-d-lactosamine (LacNAc). Peptides bearing LacNAc were enriched by RCA I affinity chromatography (Hayes et al., 1995; Haynes and Aebersold, 2000). The peptides were dissolved in 100 μL PBS and loaded onto a 1.2-m-long RCA I agarose column (Vector Laboratories, Burlingame, CA, USA) constructed in a Teflon tubing (1.55 mm i.d. × 1.2 m length, with a 0.5-μm end frit). The column was washed at room temperature with 3 mL PBS at a flow rate of 50 uL/min and 100 uL fractions were collected. The bound material was then eluted with 0.2 M lactose in PBS. Peptides were detected by measuring the absorbance at 280 nm. The pool was dried in a Speed-Vac, then desalted using OMIS C₁₈/100 μL (Varian, Palo Alto, CA, USA), and dried again (Kim et al., 2011).

For β-elimination, the desalted and dried peptides were resuspended in 1.5% triethylamine and 0.15% NaOH and then incubated at 52°C for 1.5 h. The peptides were then desalted using Sep-Pak C₁₈, dried, and stored at −20°C.

The peptide samples were purified with a C₁₈ ZipTip (Millipore, Bedford, MA, USA). Approximately 1.2 µL of the eluted peptides was mixed on the target with the matrix (10 mg/mL α-cyano-4-hydroxycinnamic acid in 50% acetonitrile and 0.1% trifluoroacetic acid) and analyzed in reflector mode on a matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) MS using Bruker Reflex III (Bruker Daltonics, Billerica, MA, USA). Processing of the spectra and data analysis were performed with the Bruker Daltonics XTOF 3.1.

Online reversed-phase nanoflow HPLC with electrospray ionization MS, which used an LTQ-OrbitrapXL equipped with electron transfer dissociation (ETD) or collision-induced dissociation (CID) sources (Thermo Scientific, Waltham, MA, USA), and analysis of the MS data were performed as described previously (Bandhakavi et al., 2009; Kim et al., 2013). The peptides were eluted using an acetonitrile gradient of 2%–40% over 60 min. Survey scans from 360 to 1,800 m/z (mass-to-charge) were acquired using the Orbitrap analyzer at 60,000 resolution. The data-dependent settings included selection of the top five most abundant ions in each survey scan for tandem mass spectrometry (MS/MS), excluding 1+ or undetermined charge states; dynamic exclusion was enabled for 20 s. Fragment ions were detected in the linear ion trap for both CID and ETD activation modes. For experiments using ETD fragmentation, the scan parameters were as follows: precursor ion isolation window of 3 m/z units, precursor ion automatic gain control of 2 × 10⁴ charges, precursor injection time of 100 ms, fluoranthene reagent ion, reagent ion automatic gain control of 400,000, reagent ion injection time of 50 ms, and reagent ion reaction time of 100 ms. Fragmentation with CID was performed as described previously (Bandhakavi et al., 2009) with 35% normalized collision energy. The CID and ETD data were analyzed using Scaffold version 3 (www.proteomesoftware.com) and OMSSA version 2.1.7, respectively, and confirmed via manual inspection.

Prior to database searching, raw CID data were extracted and converted to the mzXML format using the MS Convert software from ProteoWizard. Data were searched using SEQUEST version 27, revision 12, against the National Center for Biotechnology Information (NCBI)-derived reference sequence Arabidopsis database from September 2009, which included common contaminants from http://www.thegpm.org/crap/index.htm and CT5. The precursor and fragment ion tolerances for the database searches were set at 10 ppm and 0.8 amu. Semi-tryptic specificity was selected with up to two missed cleavage sites. Since the samples had been alkylated, the addition of 57.02146 amu to Cys was set as a fixed modification, and variable modifications included the addition of 15.9949 amu to Met, to allow for its oxidation, and loss of 18.0106 amu from either serine or threonine, which would occur when the modification of lactosamine is removed by β-elimination. The probabilities of the peptide candidate identifications being correct (Keller et al., 2002) were calculated using Scaffold version 3. Protein identifications were filtered using the following criteria: 10 ppm precursor mass tolerance, more than 95% peptide probability, and full trypsin specificity prior to confirmation by manual inspection.

When the ETD data were analyzed using OMSSA,.dta files were generated using the DTA generator (https://github.com/coongroup/Compass?search=1). The MS/MS spectra were searched against all Arabidopsis proteins in the NCBI non-redundant database and CT5. The parameters for the search were the same as those described above, with the addition of allowing variable modification of 365.1322 amu to serine (S) and threonine (T).

As we were unable to co-express the full-length GI protein with SEC in E. coli, smaller segments that collectively span GI were co-expressed and analyzed for O-GlcNAc modification ( Figure 1A ). All fragments containing amino acids 828–840 were modified, suggesting that this region had been modified ( Figures 1B, C ).

An E. coli expression construct termed pET32-CT5, which encodes GI amino acids 789–893, was expressed well, and the protein it encodes, named CT5, was highly modified when co-expressed with SEC ( Supplementary Figure S2A ). Therefore, this construct was used for further mapping studies. The serine (S) and threonine (T) residues of the region spanning amino acids 825–840 were individually mutated to alanine (A), and the mutant proteins were examined to determine whether they were modified by SEC ( Figure 2 ). The T834A and T837A mutations reduced the modification, while modification of the T829A mutant was not detectable. Since the mutations might affect the protein structure or the interaction with SEC, or create an ectopic modification site, MS mapping was employed to unambiguously map the modification site(s).

The O-GlcNAc-modified CT5 peptides produced by digestion with Lys-C or trypsin were enriched by RCA I lectin affinity chromatography. MALDI-TOF analysis of the enriched peptides demonstrated successful enrichment of the modified peptides produced with either trypsin (m/z 2,573) or Lys-C digestion (m/z 3,736) ( Supplementary Figure S2 ). When the RCA I-enriched Lys-C peptides were analyzed by ETD MS, a quintuple-charged precursor ion (m/z 751.3275) with the sequence QENTCASTTCFDTAVTSASRTEMNPRGNHK was observed to have c₈ and z₂₃ ions with a 365-Da (LacNAc) mass increase, which supports the O-GlcNAc modification of T829 ( Figure 3 ).

To confirm that the modification was O-linked, the RCA I-enriched trypsin peptides were subjected to β-elimination, which removed the O-linked modifications, and analyzed by CID MS. This analysis detected a double-charged precursor ion (m/z 1,003.9561) corresponding to the dehydrated QENTCASTTCFDTAVTSASR peptide, which is consistent with the removal of an O-linked modification. The b₈ and y₁₃ ion masses indicated that T829 was dehydrated ( Figure 4 ).