Glycosylation and Sialylation of Macrophage-derived Human Apolipoprotein E Analyzed by SDS-PAGE and Mass Spectrometry

Apolipoprotein E (apoE) is a 34-kDa glycoprotein secreted from various cells including hepatocytes and macrophages and plays an important role in remnant lipoprotein clearance, immune responses, Alzheimer disease, and atherosclerosis. Cellular apoE and plasma apoE exist as multiple glycosylated and sialylated glycoforms with plasma apoE being less glycosylated/sialylated than cell-derived apoE. Some of the glycan structures on plasma apoE are characterized; however, the more complicated structures on plasma and cellular/secreted apoE remain unidentified. We investigated glycosylation and sialylation of cellular and secreted apoE from primary human macrophages by one- and two-dimensional gel electrophoresis and mass spectrometry. Our results identify eight different glycoforms with (HexNAc)2-Hex2-(NeuAc)2 being the most complex glycan detected on Thr194 in both cellular and secreted apoE. Four additional glycans were identified on apoE(283–299), and using β-elimination/alkylation by methylamine in vitro, we identified Ser290 as a novel site of glycan attachment. Comparison of plasma and cellular/secreted apoE from the same donor confirmed that cell-derived apoE is more extensively sialylated than plasma apoE. Given the importance of the C terminus of apoE in regulating apoE solubility, stability, and lipid binding, these results may have important implications for our understanding of apoE biochemistry.

creted by most cells including hepatocytes, smooth muscle cells, neuronal cells, and macrophages (1-3) and demonstrates extraordinary functional diversity. It has important roles in remnant lipoprotein clearance, the immune response, Alzheimer disease, cell proliferation, and lymphocyte activation (4,5). More recent studies suggest that elevated plasma apoE precedes elevation of C-reactive protein and confers increased risk of cardiovascular death in the elderly (6). Proteomics-based approaches have identified elevated high density lipoprotein (HDL)-apoE as being associated with coronary disease (7). In contrast, macrophage-specific expression of apoE protects against atherosclerosis in mice (8,9). The mechanisms by which macrophage apoE is antiatherogenic may include stimulating the removal of excess cholesterol from macrophage foam cells as well as anti-inflammatory, antiproliferative, and immunomodulatory properties (4, 5, 10 -12). An accurate understanding of the structure of apoE secreted from macrophages is important for our understanding of its properties and its role in the atherosclerotic process.
Structural studies on apoE have provided important insights into its biological properties (13). Crystallography has demonstrated that the N-terminal domain is structured in a globular four-helix bundle with the helices orientated in an antiparallel alignment (14). The structure of the C terminus has not been resolved by crystallography, but circular dichroism spectroscopy indicates it to be highly ␣-helical (14). Recently, NMR studies of monomeric, full-length human apoE indicated that the C-terminal domain in the intact protein adopts a more defined structure than it does as an isolated fragment (15). Lipid binding occurs at the C terminus (residues 244 -272), resulting in unfolding of the molecule into a helical hairpin with the binding region for the low density lipoprotein (LDL) receptor contained within the N terminus at its apex (16).
Mucin-type O-glycosylation is a particularly common, complex, and important post-translational modification of secreted and cell surface glycoproteins (17,18) that is difficult to accurately characterize; however, several recent reports have facilitated analysis (19,20). Cellular apoE and plasma apoE exist as multiple glycoforms, which vary in charge because of variable sialylation. The initial analysis of the carbohydrate content of plasma very low density lipoprotein (VLDL)-apoE by colorimetric methods and gas chromatography demonstrated that the major unmodified hexose in apoE was galactose and that N-acetylglucosamine, N-acetylgalactosamine, and sialic acid were present (21,22). Two-dimensional gel electrophoresis  identified up to six sialylated apoE (Es) glycoforms in cells for any given genotype and fewer sialylated glycoforms in plasma (22). ApoE does not contain the consensus sequence (NX(T/S/C)) required for N-linked glycans, and carbohydrate residues are attached to apoE via an O-linkage to residue Thr 194 (23)(24)(25). More recent studies using 2-DE and MALDI-TOF/TOF (23) confirmed previous results and identified five glycosylated glycoforms of apoE in plasma VLDL with the most complex sugar structures containing two sialic acid residues (HexNAc-Hex-NeuAc-NeuAc). There were more negatively charged glycoforms present on 2-DE than were distinguished by MALDI-TOF/TOF, raising the possibility that complex structures containing more than two sialic acid residues may be inherently unstable during MS analysis. Importantly, this recent study did not analyze apoE glycoforms in, or secreted from, cells.
The purpose of this study was to undertake the first detailed characterization of the glycan structures of apoE from primary human macrophages by 1-DE, 2-DE, and mass spectrometry. We found that cellular and secreted apoE in human macrophages has at least eight different glycoforms with (HexNAc) 2 -Hex 2 -(NeuAc) 2 being the most complex glycan identified. We extend previous studies by the identification of a novel site of glycan attachment on Ser 290 near the functionally important apoE C terminus in addition to glycosylation of Thr 194 and show that a major glycoform is present in each of the spots separated by 2-DE.

Isolation and Culture of Human Monocyte-derived Macrophages (HMDMs)
Human monocytes were isolated from white cell buffy coat concentrates from healthy donors using density gradient centrifugation after layering on Ficoll-Paque Plus (GE Healthcare). Purified monocytes were differentiated in 6-well Primaria plates (BD Biosciences) by culturing in RPMI 1640 medium containing 50 units/ml penicillin G, 50 g/ml streptomycin, 2 mM L-glutamine, 10% heat-inactivated human serum, and 25 ng/ml macrophage colony-stimulating factor (PreproTech) for 3 days followed by culturing in the same medium without macrophage colony-stimulating factor for up to 7 days. After differentiation, the cells were washed and enriched with cholesterol by incubation in RPMI 1640 medium including 10% lipoprotein-deficient serum and 50 g/ml acetylated LDL for 2 days. After enrichment, the cultures were washed twice with prewarmed RPMI 1640 medium and incubated in RPMI 1640 medium for between 1 and 24 h. At the indicated time points, the cells and medium samples were harvested. Cells were lysed using radioimmune precipitation assay buffer (50 mM Tris-Cl, pH 7.5, 150 mM NaCl, 0.1% SDS, 1% Triton X-100, 0.5% deoxycholate, and protease inhibitors). White cell buffy coat concentrates and human serum were supplied by the New South Wales Red Cross blood transfusion service, Sydney, Australia. Donors were genotyped for apoE by the laboratory of Prof. D. Sullivan, Royal Prince Alfred Hospital, Sydney, Australia, by restriction enzyme analysis (24).

Isolation and Culture of Human Monocytes and Preparation of Human Plasma Proteins
Blood samples in EDTA-containing tubes were obtained from a healthy volunteer with an apoE3/3 genotype. Monocytes were isolated as described above. After density gradient centrifugation, plasma supernatant was collected. Total plasma proteins were prepared as described (25). Briefly, 12 l of plasma was mixed with 20 l of a 10% SDS and 2.3% DTT solution boiled at 95°C for 5 min. The sample was diluted to 500 l with rehydration buffer (9 M urea, 2 M thiourea, 4% CHAPS, and trace bromophenol blue). 30 l of the sample was separated by 2-DE, and apoE was detected by Western blot.

Immunoprecipitation
To isolate apoE from cholesterol-enriched HMDMs, cell lysates and medium were immunoprecipitated using a goat antibody to human apoE and protein A-Sepharose. 1.2 mg of cell lysates and medium samples was precleared for 30 min by the addition of 50 l of protein A-Sepharose, then mixed with 5 l of goat anti-apoE antibody, and incubated for 1 h with rotation. After 1 h, 50 l of protein A-Sepharose was added, and the samples were incubated for another 1 h with rotation. Beads were spun down and washed five times with radioimmune precipitation assay buffer. ApoE was eluted using rehydration buffer.

Two-dimensional Electrophoresis
To detect individual apoE glycoforms, 40 l of immunoprecipitated apoE was subjected to 2-DE. For the first dimension, isoelectric focusing was performed with a ZOOM IPGRunner system (Invitrogen) using 7-cm, pH 4 -7 strips at 2000 V-h at room temperature. Samples were then reduced in 1ϫ NuPAGE sample reducing agent for 15 min and alkylated with 125 mM iodoacetamide for 15 min after which second-dimension SDS-PAGE was performed using NuPAGE Novex 4 -12% Bis-Tris ZOOM gels. After electrophoresis, the gels were fixed, and protein spots were visualized using a SilverQuest silver staining kit (Invitrogen) and SimplyBlue SafeStain (Invitrogen) for mass spectrometry analysis. Preliminary experiments confirmed that this separation clearly distinguished apoE 2, 3, and 4 glycoforms as described (26), consistent with calculated pI values of 5.65 (apoE3), 5.81 (apoE4), and 5.52 (apoE2) (ExPASy Compute pI/Mw tool). For all experiments described herein, apoE3/3 donor macrophages were used exclusively.

In-gel Digestion
Protein spots were excised from one-or two-dimensional gels and destained to remove Coomassie stain by incubation with 100 mM NH 4 HCO 3 in CH 3 CN for 1 h. Reduction and alkylation were then performed to maximize digestion efficiency. For reduction, gel spots were incubated with 10 mM DTT in 20 mM NH 4 HCO 3 at 37°C for 1 h, and this was followed by incubation with 25 mM iodoacetamide in 20 mM NH 4 HCO 3 at 37°C for 1 h for alkylation. Gel spots were dehydrated with 100 l of 100% CH 3 CN for 10 min. Sequencing grade trypsin (Promega) was added to the dehydrated gel spots at 10 ng/l in 15 mM NH 4 HCO 3 , and samples were incubated overnight at 37°C. Digested peptides were extracted from gel spots with 1% formic acid for 10 min followed by incubation with 100% CH 3 CN for another 10 min. Peptide mixtures were evaporated in a SpeedVac for 1 h and finally dissolved in 0.05% heptafluorobutyric acid and 1% formic acid.

Mass Spectrometry
Q-Tof Ultima-Digest peptides were separated by nano-LC using a CapLC and autosampler system (Waters). Samples (5 l) were concentrated and desalted onto a micro-C 18 precolumn (500 m ϫ 2 mm; Michrom Bioresources, Auburn, CA) with H 2 O:CH 3 CN (98:2, 0.05% heptafluorobutyric acid) at 15 l/min. After a 4-min wash, the precolumn was switched (Valco 10-port valve, Dionex) in line with a fritless nanocolumn (75 m ϫ ϳ10 cm) containing C 18 medium (5 m, 200 Å; Magic Bioresources, Michrom) manufactured according to Gatlin et al. (27). Peptides were eluted using a linear gradient of H 2 O:CH 3 CN (98:2, 0.1% formic acid) to H 2 O:CH 3 CN (65:35, 0.1% formic acid) at ϳ300 nl/min over 30 or 60 min using a CapLC system (Waters). The precolumn was connected via a fused silica capillary (25 m ϫ 10 cm) to a low volume tee (Upchurch Scientific) where high voltage (2400 V) was applied, and the column tip was positioned ϳ1 cm from the Z-spray inlet of a Q-Tof Ultima API hybrid tandem mass spectrometer (Micromass, Manchester, UK). Positive ions were generated by electrospray, and the Q-Tof Ultima was operated in datadependent acquisition (DDA) mode. A TOF MS survey scan was acquired (m/z 350 -1700; 1 s), and the two largest multiply charged ions (counts Ͼ20) were sequentially selected by Q1 for MS/MS analysis. Argon was used as the collision gas, and an optimum collision energy was chosen (based on charge state and mass). Tandem mass spectra were accumulated for up to 2 s (m/z 50 -2000). Peak lists were generated by MassLynx (version 4.0 SP4, Micromass) using the Mass Measure program and submitted to the database search program Mascot (version 2.2, Matrix Science, London, UK). Search parameters were as follows: precursor and product ion tolerances were Ϯ0.25 and 0.2 Da, respectively; Met(O) and carboxyamidomethyl-Cys were specified as variable modifications; enzyme specificity was trypsin; one missed cleavage was possible; and the non-redundant protein database from NCBI (March 2008) was searched.
LTQ-FT Ultra-Digest peptides were separated by nano-LC using an Ultimate 3000 HPLC and autosampler system (Dionex, Amsterdam, Netherlands) as described above. Peptides were eluted using a linear gradient of H 2 O:CH 3 CN (98:2, 0.1% formic acid) to H 2 O:CH 3 CN (64:36, 0.1% formic acid) at 350 nl/min over 30 or 60 min. High voltage (1800 V) was applied to a low volume tee (Upchurch Scientific), and the column tip was positioned ϳ0.5 cm from the heated capillary (T ϭ 200°C) of an LTQ-FT Ultra (Thermo Electron, Bremen, Germany) mass spectrometer. Positive ions were generated by electrospray, and the LTQ-FT Ultra was operated in DDA mode.
A survey scan (m/z 350 -1750) was acquired in the FT ICR cell (resolution ϭ 100,000 at m/z 400 with an accumulation target of 1,000,000 ions). Up to six of the most abundant ions (Ͼ3000 counts) with charge states Ͼ2ϩ were sequentially isolated and fragmented within the linear ion trap using collisionally induced dissociation with a normalized collision energy of 25 V, activation q of 0.25, and activation time of 30 ms at a target value of 30,000 ions. m/z values selected for MS/MS were dynamically excluded for 30 s. Peak lists were generated using Mascot Daemon/extract_msn (Matrix Science, Thermo) using the default parameters and submitted to the database search program Mascot (version 2.2, Matrix Science). General search parameters were as follows: precursor tolerance was 4 ppm, product ion tolerance was Ϯ0.4 Da, Met(O) and Cys carboxyamidomethylation were specified as variable modifications, enzyme specificity was trypsin, one missed cleavage was possible, and the non-redundant protein database from NCBI or Swiss-Prot (March 2008) was searched. Additional searches with variable modifications N-acetylation, Cyssulfenic acid, Cys-sulfonic acid, deamidation, and phosphorylation were performed. All peptides assigned by Mascot had Mowse scores Ͼ20, and spectra of glycopeptides were interpreted and validated manually. Extracted ion chromatograms (XICs) were derived from either MS or MS/MS spectra from the calculated monoisotopic mass of each ion (Ϯm/z 0.1) using the QualBrowser in XCalibur (version 2.07), and abundances were calculated from the area of each extracted ion.

␤-Elimination and Alkylation of Glycopeptides
Peptide digests were treated with NH 2 CH 3 vapor as described (20). Briefly, a portion of the digests (ϳ15%) was dried (SpeedVac), tubes were placed in scintillation vials together with a microcentrifuge tube containing NH 2 CH 3 (50 l), and the vial was flushed with N 2 . After capping, the vials were left at 70°C for 60 min. Peptides were solubilized and analyzed by nano-LC MS/MS using the Q-Tof Ultima as described above.

Analysis of Ser 290 on Lipid-bound ApoE
A high resolution EPR structure of lipid-bound apoE was recently published, and the Protein Data Bank file was a kind gift from Prof K. Weisgraber (31). Analysis of the glycosylation site on Ser 290 was performed using Discovery Studio 2.0 (Accelrys Software Inc.) on this structure.

Identification of O-Glycosylated Peptides of ApoE Using
Q-TOF MS-To analyze the sugar structures of apoE glycoforms, cellular and secreted apoE was immunoprecipitated and separated by 1-DE ( Fig. 1, inset). After Coomassie staining, 34-kDa apoE bands were excised, destained, and treated with trypsin, and peptides were analyzed by nano-LC datadependent tandem mass spectrometry. Total ion chromatograms (TICs) and XICs derived from all MS/MS spectra using m/z 204.08, 292.09, and 366.13 for glycan oxonium ions (32) were obtained from Q-Tof Ultima mass analysis (not shown). These were used to indicate likely glycopeptides from which MS/MS spectra were manually interpreted. Recently, HexNAc-Hex oxonium ions (calculated mass [M ϩ H] ϩ ϭ 366.13) were identified in the sugar structures of apoE (25). Cellular apoE and secreted apoE showed very similar XIC profiles, and two different potential glycosylated peptides were readily identified (not shown). An MS/MS spectrum of the [M ϩ 2H] 2ϩ precursor ion m/z 1223.07 (single charged m/z 2445.12) was obtained. This was identified as modified AATVGSLAGQPLGER (apoE(192-206)) and is consistent with glycosylation of Thr 194 (25). The delta mass between 2445.12 and 1497.80 is consistent with HexNAc-Hex-(NeuAc) 2 glycosylation of this peptide (ExPASy GlycoMod tool). Four double charged glycan fragment ions (m/z 850.93, 932.45, 996.47, and 1077.50) were detected in the spectrum derived from this protonated peptide, and the compositions of sugar structures of these fragments are described ( Fig. 1a and legend).
Identification of O-Glycosylated Peptides of ApoE Using LTQ-FT MS-To further confirm the Q-Tof Ultima MS/MS results and detect additional glycopeptides, we performed similar experiments and analysis using a more sensitive LTQ-FT Ultra mass spectrometer. In these data-dependent experiments, the linear ion trap was used to collect MS/MS spectra, and the ICR cell was used for MS. Fig. 2 shows the partial TIC (Fig. 2a) and XICs (MS2 spectra) (Fig. 2, b-d) of separated apoE protonated peptides. To identify specific glycopeptides observed from the Q-TOF experiments and other glycopeptides, the XICs were extracted for the oxonium ions m/z 204.08 ϩ 292.09 ϩ 366.12 (Fig. 2b) (Fig. 2d). Despite recording these spectra in the linear ion trap with the concomitant loss of low mass ions and mass accuracy of approximately Ϯm/z 0.4, many abundant peaks characteristic of oxonium ion or apoE glycopeptides were observed, and the glycosylation patterns were characterized (Table I and Fig. 3).
The precursor and product ion measurements in MS/MS spectra from the Q-Tof Ultima were all within ϳ50 ppm of the calculated values, and with a resolution of ϳ10,000, each product charge state was accurately determined, allowing confident assignment of glycan fragmentation patterns (Fig.  1). Fig. 3 shows MS/MS spectra of the most extensively glycosylated peptides identified from the LTQ-FT. The two glycoforms detected were triple charged ions, and these were readily resolved in the MS scan. An [M ϩ 3H] 3ϩ precursor ion, m/z 937.08 (Ϯ2 ppm, single charged, m/z 2810.28), was identified for apoE(192-206) (Fig. 3a); the delta mass between 2810.28 and 1497.80 corresponds to apoE(192-206)-(HexNAc) 2 -Hex 2 -(NeuAc) 2 , and the glycosylation pattern was consistent with the observed fragment ions (Fig. 3a). An [M ϩ 3H] 3ϩ precursor ion, m/z 856.67 (Ϯ2 ppm, single charged, m/z 2568.14), was identified for apoE(283-299) (Fig. 3b) and was consistent with apoE(283-299)-HexNAc-Hex-(NeuAc) 2 (Fig. 3b). Although MS/MS spectra measured within the LTQ were without isotopic resolution, many abundant glycosylated peptide fragment ions and oxonium ions were detected in both MS/MS spectra, allowing confident prediction of the glycan structure. The determined compositions of sugar structures of these fragment ions are described in Fig. 3.
The relative abundances of each ionized glycoform and unmodified peptides were determined by calculating monoisotopic XICs using the summed area of the 2ϩ and 3ϩ charge states (Ϯm/z 0.1) of potential precursor ions from all LTQ-FT MS spectra (supplemental Fig. 1). Separation of each of the glycoforms was observed with no apparent co-elution that could be attributed to in-source fragmentation or detectable loss of sialic acid (supplemental Fig. 1). No estimates of the relative amounts of each form within the samples can be and secreted (f) apoE normalized to unmodified apoE(76 -90) peptide. Data were obtained by automatic peak area detection from double and triple charged ion XICs (see text for details), and to bring out the large differences in normalized abundance, the y axis is in log scale. made because of potential differences in ionization efficiencies from these types of analyses, but reliable comparisons of the relative amounts between samples can be made because any differences in ionization remains the same in each measurement and glycoform. A scatter plot was generated using the area of the double and triple charged ions of apoE(76 -90) as a control. We assumed this procedure would normalize the variable amount of apoE in each extract. For each peptide, the relative abundance of each glycan was similar in secreted and cellular apoE with the exception of HexNAc, which was more abundant in cellular than secreted apoE (Fig. 2e). Interestingly, the apoE(192-206) peptide showed proportionately more glycosylation than the apoE(283-299) peptide, suggesting preferential glycosylation of the apoE(192-206). The figure also shows that more complex sugars detected in apoE(192-206) peptides were not detected in apoE(283-299) peptides and that the relative abundance of sugars such as HexNAc-Hex-NeuAc and HexNAc-Hex-(NeuAc) 2 was greater in apoE(192-206) peptides than in apoE(283-299) peptides.

Analysis of ApoE Glycoforms Synthesized in HMDMs Using 2-DE-Cholesterol-enriched
HMDMs were washed and incubated in serum-free medium, and media and cells were collected at 1 and 24 h. ApoE was immunoprecipitated from lysates and from culture media, separated by 2-DE (IP-2-DE), and detected by silver staining and Western blotting (Fig. 4). Cellular apoE and secreted apoE were separated within a pI range of 5.1-5.8. One major asialylated apoE (E) glycoform (also designated glycoform 1) was detected, and at least seven Es glycoforms were detected in cellular and medium samples (Fig. 4). Glycoforms detected in cell lysates and in culture medium were sequentially numbered 1-14 for later MS/MS analysis. On the basis of apparent molecular weight and pI, cellular glycoforms 1, 2, 4, 5, 6, and 7 appeared to correspond to secreted glycoforms 8, 9, 10, 11, 12, and 14.
Between 1 and 24 h, cellular apoE glycoforms did not change (Fig. 4, a versus b), whereas apoE secreted into medium changed significantly over this time with increased abundance of E and EЈ relative to Es glycoforms (Fig. 4, c  versus d). EЈ was more prominent in medium than in cells (Fig.  4, b versus d), whereas glycoform 3 was more prominent in cells than in medium (Fig. 4, a versus c). Similar results were obtained with silver staining (Fig. 4, a-d) and Western blot (Fig. 4, e-g) using macrophages from two independent donors. Consistent with previous studies (22,(33)(34)(35), apoE in cell lysates and cell culture media were more extensively sialylated than was plasma apoE even when the cells and plasma were from the same donor (Fig. 4, e and g versus h).
Although 2-DE had indicated the presence of up to seven sialic acid residues per molecule of apoE, 1-DE and MS/MS detected only up to two sialic acid residues per glycan. To investigate in more detail the identity of each glycoform and whether the apparent differences in charge of apoE glycoforms identified by 2-DE corresponded to differences in sialylation and/or glycosylation, immunoprecipitated cellular

LTQ-FT MS analysis of O-glycosylated apoE peptides isolated from macrophages
The 34-kDa band of apoE was excised after 1-DE (see legend to Fig 3)  apoE and secreted apoE were separated by 2-DE, destained, and digested, and DDA MS analysis using the LTQ-FT was performed (Table II). As observed after MS analysis of 1-DE, the same two distinct glycosylated peptides were detected in many of the apoE spots with the same glycosylation patterns observed. In earlier studies, glycoform 1 was identified previously as asialylated and non-glycosylated (25,36). Our analysis confirmed the absence of sialic acid residues but identified HexNAc in glycoform 1 as well as a non-glycosylated peptide, suggesting that this spot is a mixture of non-sialylated, glycosylated, and non-glycosylated apoE. Glycoform 3 had a similar molecular weight but was more highly charged than glycoform 1, suggesting either failure to detect sialic acid residues due to glycan instability or an alternative source of negative charge to explain its pI relative to glycoform 1. Mascot searches were repeated including phosphorylation, de-amidation, and Cys-sulfonic acid variable modifications as these are hypothesized to change pI values. The Mascot search result of glycoform 3 (supplemental Fig. 2) shows a peptide containing Cys-sulfonic acid, and this peptide was absent from the search of glycoform 1, indicating that the likely cause of the pI shift of glycoform 3 was conversion of Cys to Cys-sulfonic acid (supplemental Fig. 2). Glycoforms of increasing charge (e.g. glycoforms 2, 4, 5, 6, and 7) demonstrated increasing glycan complexity but not beyond that of (HexNAc) 2 -Hex 2 -(NeuAc) 2 (Table II), suggesting the presence of multiple sites of glycosylation on apoE or some unidentified post-translational modifications to explain their difference in charge. The presence of multiple glycans per spot may also be due to instability during sample preparation or LC/MS analysis (Table II) 2 and shows the same fragment ions observed with the spectrum acquired using the Q-Tof Ultima (Fig. 1b). The fragment ions indicated by ϩϩ are assumed to be double charged. HNAc, HexNAc. 299) glycopeptides in each separated 2-DE spot demonstrated similar glycan distribution patterns in cellular and secreted apoE for each peptide (Fig. 5). For all 2-DE glycoforms of apoE, the relative proportion of glycosylated apoE (192)(193)(194)(195)(196)(197)(198)(199)(200)(201)(202)(203)(204)(205)(206) peptides appeared greater than the proportion of glycosylated apoE(283-299) peptides (Fig. 5), and the more highly glycosylated/sialylated forms of apoE occurred at higher pI values.
Characterization of Sialic Acid Modification in ApoE Glycoforms-To identify the types of sialic acid linkage of apoE, lectin blots were performed using MAA (which detects terminal ␣-(233)-sialic acid linkage) and SNA (which detects terminal ␣-(236)-sialic acid linkage) (37). Secreted apoE was used for 1-DE analysis in preference to cellular apoE because preliminary studies demonstrated extensive cross-reactivity of lectins with a number of cell proteins in cell lysates, causing smearing (not shown). Fetuin protein, which has both ␣-(233)and ␣-(236)-sialic acid linkages, was stained by both MAA and SNA. Several bands were detected by each lectin in secreted apoE samples (Fig. 6). Comparison of the SNA lectin blot (middle three lanes) and apoE Western blot (far right lanes) demonstrated that the bands identified in the SNA lectin blot do not correspond to apoE. In contrast, a 34-kDa band corresponding to apoE was detected by MAA lectin blot.
To confirm that cellular apoE also demonstrated ␣-(233)sialic acid linkage, IP-2-DE lectin blots were performed on cellular IP-2-DE apoE to eliminate smearing (Fig. 6b). MAA weakly stained two Es glycoforms in cell lysates corresponding to glycoforms 2 and 4. However, there were only minor changes in apoE glycoform distribution after treatment with ␣-(233)-neuraminidase and substantial elimination of Es glycoforms after ␣-(233, 6,8,9)-neuraminidase treatment, suggesting that linkages other than ␣-(233) were present (Fig.  6c). The loss of glycoforms 4 -7 in cell lysates and glycoforms 10 -14 from medium samples after neuraminidase treatment confirmed the presence of sialic acid residues in cell and secreted apoE. Interestingly, EЈ was generated after treating apoE from cell lysates and media with ␣-(233,6,8,9)-neuraminidase, suggesting that EЈ arises from the desialylation of Es glycoforms. ApoE(283-299)-The newly identified glycosylated tryptic peptide (apoE(283-299)) has three potential O-linked glycosylation sites (two Ser and one Thr in VQAAVGTS-AAPVPSDNH). To identify which amino acid represented the site of O-linked glycan attachment, peptide digests were treated with methylamine vapor, and under these conditions, glycopeptides underwent ␤-elimination of the glycan and alkylation, forming stable methylamine derivatives of either Ser or Thr with a mass addition of m/z 13 (20).

Determination of Location of Second Site of O-Glycosylation in
After nano-LC and Q-Tof Ultima DDA MS analysis, Mascot database searches were performed including variable modifications Ser-methylamine and Thr-methylamine together with the standard search parameters. Both Thr 194 residue, which is present in the apoE(192-206) glycopeptide, and Ser 290 residue, which is present in apoE(283-299), were detected as modified by addition of methylamine. The Mascot search assignment was confirmed by manually interpreting the MS/MS spectra of the two modified peptides. The MS/MS spectra of apoE(192-206) peptide exposed to methylamine identified modification of Thr 194 only (supplemental Fig. 3). Fig. 7 shows the MS/MS spectrum of double charged ions of apoE(283-299) (Fig. 7a) and methylamine-derivatized FIG. 4. Two-dimensional gel electrophoresis of apoE glycoforms in HMDMs. HMDMs from two independent donors (a-d, donor 1; e-h, donor 2) were cholesterol-enriched by incubation for 2 days with 50 g/ml AcLDL, washed, and incubated in serum-free and BSA-free RPMI 1640 medium for up to 24 h. At the indicated times, apoE was immunoprecipitated from cell lysates (a, b, and e) and media (c, d, f, and g), separated by 2-DE, and detected by silver staining (a-d) or Western blot (e-h). The distribution of cellular apoE glycoforms did not change between 1 (a) and 24 h (b), whereas there was an increase in the abundance of glycoforms E and EЈ relative to Es glycoforms between 1 and 24 h in the medium (c versus d and f versus g). Western blots show that plasma apoE (h) comprises less charged apoE glycoforms than secreted (g) or cellular apoE (e) from the same donor. On the basis of mass and pI, the following glycoforms appear to be the same: 1, E, and 8; 2 and 9; 4 and 10; 5 and 11; 6 and 12; and 7 and 14.
apoE(283-299) (Fig. 7b). Comparison of the y-ion series of product ions present in Fig. 7, a and b, showed a mass difference of m/z 13 in y-ions 13, 12, 11, and 10. The mass of the ions in both spectra (i.e. y-ions 9, 8, 7, and 6) were identical after Ser 290 , thereby confirming that the methylamine-modified residue and the major glycosylation site was Ser 290 within apoE(283-299). A similar analysis of the frag-mentation spectrum containing Thr 194 confirmed that this was the major glycosylated amino acid in apoE(192-206) (supplemental Fig. 3). DISCUSSION We undertook the first detailed analysis of the glycosylation of apoE from primary human macrophages. Our results con-  firm previous studies identifying that apoE is extensively glycosylated and sialylated but extend these previous observations by demonstrating multiple glycosylation sites on apoE and identifying Ser 290 on the apoE C terminus as a novel site of glycosylation. Given the importance of the C terminus of apoE in regulating apoE solubility, stability, and lipid binding, these results may have important implications for our understanding of apoE biochemistry. Mucin-type O-glycosylations as occur on apoE are common, complex, and important post-translational protein modifications (17,18) that are difficult to accurately characterize (19,20). They are typically initiated in the Golgi with attachment of an N-acetylgalactosamine (GalNAc) residue to the side chain of an exposed Ser or Thr residue. Stepwise elongation generates up to eight core structures before further modification by processes such as sialylation. O-Linked sugars are frequently located in clusters in short regions of peptide chains containing repeating units of Ser, Thr, and Pro. Consistent with this, the peptide apoE(283-299) is rich in Ser and Thr and is the most highly predicted site for O-glycosylation in apoE (NetOGlyc 3.1 Server), supporting our identification of glycosylation on the apoE C terminus.
Our results confirm the complementary nature of MS and 2-DE approaches to identify apoE glycoforms. We identified at least seven consistently observed glycoforms of apoE in cells and in media by 2-DE, but analysis of these glycoforms by MS did not precisely distinguish between more highly charged and less charged glycoforms because multiple glycopeptides were detected in each 2D spot. Mancone et al. (25), who studied apoE associated with VLDL, also identified multiple glycoforms on 2-DE but reported no more than two sialic acid residues per glycoform. As variations in apoE charge on 2-DE were largely eliminated by treatment with neuraminidase, our data indicate that there may be up to six sialic acid residues per molecule of cell-derived apoE, and we propose that this variation relates to both the number of amino acid residues with O-linkage and variation in the number of sialic acid residues (between zero and two) on each O-linkage. Our data indicate that glycoform 3 differs from glycoform 1 as a result of conversion of Cys to Cys-sulfonic acid, which in other systems has been associated with both loss and gain of function (38). For the other spots, it is possible that there may be glycosylation sites on apoE in addition to both Thr 194 and Ser 290 . Preliminary analysis of lower abundance ions and MS/MS spectra (data not shown) suggests the presence of glycosylated residues other than Ser 290 within apoE(283-299), and candidates include Thr 289 and Ser 296 . Other modifications of apoE may also contribute to differ- ences in glycoform charge. ApoE can be phosphorylated by protein kinase CK2 in vitro on Ser 296 (39). Mascot searches, however, did not identify other common modifications including phosphorylation of macrophage apoE in these studies.
The functional consequences of apoE sialylation are likely to be complex. ApoE was secreted normally from UDP-galactose/UDP-N-acetylgalactosamine 4-epimerase-deficient ldlD CHO cells expressing a reversible defect in protein O-glycosylation (36,40), implying that glycosylation is not essential for secretion. However, sialylated apoE was reportedly released from the plasma membrane of macrophages more efficiently than apoE that was not sialylated (41). Desialylation of apoE markedly decreased binding to HDL (42), which may be critical for facilitating HDL-mediated recycling of apoE and removal of cellular cholesterol. Protein solubility and stability can be markedly enhanced by glycosylation (43)(44)(45) with a linear dependence on the number of carbohydrate residues attached to the protein (46). In the case of apoE, we expect that C-terminal glycosylation may protect against self-association, spontaneous aggregation, and fibril formation as occurs in Alzheimer disease and atherosclerotic plaques (47,48). Ser 290 is situated just after the last C-terminal ␣-helix at the start of an unstructured C-terminal tail. Therefore, glycosylation of Ser 290 is unlikely to interfere with the interhelical interactions (supplemental Fig. 4). More likely, the presence of highly negatively charged sialic acid residues on Ser 290 could modify interactions between apoE and phospholipid head groups or affect interactions between apoE molecules.
All sialylation and glycosylation of plasma VLDL-derived apoE have been attributed to O-linkage to residue Thr 194 (36), which resides within the peptide AATVGSLAGQPLQER (apoE (192-206)). In our studies, secreted apoE(192-206) peptides were more commonly glycosylated and contained larger glycan structures than apoE(283-299) peptides. As previous . e, sequence-specific b-and y-type ions identified in both spectra. Intense product ions indicate conversion of Ser 290 to Ser 290 -methylamine, indicating that the O-glycosylation site was predominately on Ser 290 (product ions containing Ser 290 were shifted by ϩm/z 13). Spectra were recorded using a Q-Tof Ultima with precursor/production ion resolution ϳ10,000 and a root mean square error of ϳ50 ppm. Rel Abun, relative abundance. studies found that mutagenesis of Thr 194 precluded apoE glycosylation (36), we propose that Thr 194 provides an initial and necessary site for apoE glycosylation and that this is followed by glycosylation of other sites on the C terminus. Such a stepwise pathway has been described for MUC1 (49).
Although it has been postulated that apoE undergoes postsecretory desialylation by plasma neuraminidases (50), VLDL-apoE reinjected in humans was not desialylated (51), and the mechanism by which apoE becomes desialylated is unknown. Our 2-DE analysis indicates that the glycoform distribution of secreted apoE changes over time with less sialylated forms (E and EЈ) appearing in medium during longer incubations. This suggests either that macrophages desialylate apoE extracellularly or that sialic acid-poor apoE is more slowly secreted and accumulated in the medium than is sialylated apoE. Desialylation appears the more likely mechanism as the glycoform EЈ is generated by treatment of cellular or secreted apoE with neuraminidase. Our studies suggest the possibility that neuraminidase activity at the macrophage surface or in adjacent extracellular space acts on secreted apoE to generate lower molecular weight glycoforms after secretion. Importantly, recent studies have described endogenous sialidases Neu 1 and Neu 3 in human macrophages (52). The ability of cells other than macrophages, such as hepatocytes, to similarly generate sialic acid-poor apoE requires further investigation.
Sialylation most commonly occurs through ␣-(233) and ␣-(236) linkages to galactose. Our data based on lectin blotting indicate that macrophage-derived apoE has some terminal ␣-(233)-sialic acid linkages. However, the almost complete elimination of ES glycoforms with ␣-(233, 6,8,9)neuraminidase and weak effect of ␣-(233)-neuraminidase strongly suggest that other sialic acid linkages such as ␣-(236), ␣-(238), and ␣-(239) are likely to be present. Some ␣-(236)-sialic acid linkages may be below the limit of detection by SNA lectin blot especially if these are non-terminal sialic acid residues (29). The site specificity of O-mucins is tissue-specific as the GalNAc-transferases have tissue-specific expression patterns. It will therefore be interesting to compare the sialic acid linkages of macrophage-secreted, hepatocyte-secreted, and plasma apoE as differences may allow robust distinction between hepatocyte-derived, lipoprotein-derived, and macrophage-derived apoE in various tissues such as the arterial wall.