Plasma Proteins Modified by Advanced Glycation End Products (AGEs) Reveal Site-specific Susceptibilities to Glycemic Control in Patients with Type 2 Diabetes*

Protein glycation refers to the reversible reaction between aldoses (or ketoses) and amino groups yielding relatively stable Amadori (or Heyns) products. Consecutive oxidative cleavage reactions of these products or the reaction of amino groups with other reactive substances (e.g. α-dicarbonyls) yield advanced glycation end products (AGEs) that can alter the structures and functions of proteins. AGEs have been identified in all organisms, and their contents appear to rise with some diseases, such as diabetes and obesity. Here, we report a pilot study using highly sensitive and specific proteomics approach to identify and quantify AGE modification sites in plasma proteins by reversed phase HPLC mass spectrometry in tryptic plasma digests. In total, 19 AGE modification sites corresponding to 11 proteins were identified in patients with type 2 diabetes mellitus under poor glycemic control. The modification degrees of 15 modification sites did not differ among cohorts of normoglycemic lean or obese and type 2 diabetes mellitus patients under good and poor glycemic control. The contents of two amide-AGEs in human serum albumin and apolipoprotein A-II were significantly higher in patients with poor glycemic control, although the plasma levels of both proteins were similar among all plasma samples. These two modification sites might be useful to predict long term, AGE-related complications in diabetic patients, such as impaired vision, increased arterial stiffness, or decreased kidney function.

AGEs have been localized in biological tissues by immunohistochemistry (27)(28)(29) and quantified with ELISA using monoclonal antibodies recognizing methylglyoxal-derived hydroimidazolone (30), imidazolone (31), pentosidine (32), pyrraline (33), and N ⑀ -carboxymethyl lysine (34,35). Alternatively, individual modifications were assayed by their intrinsic fluorescent properties (36 -38) or after derivatization with a fluorophore (37,39). AGEs can be detected likewise after chromatographic separation by mass spectrometry without derivatization in the multiple reaction monitoring mode, allowing their quantification with high sensitivity (40,41) and robustness (42). Characterization of amide-AGEs in proteomes will help with understanding the development of glycation-associated pathologies by identifying the affected protein (function) and providing access to subtissue level changes of individual residues prone to such modifications. We recently reported a two-step procedure to detect AGE-modified residues in plasma samples using specific and sensitive precursor ion scans (43,44). Here, we extended this approach by using gas phase fractionation to detect amide-bound AGEs that have been previously identified in lens homogenate (41) and as free adducts in human plasma and urine and found to be increased in uremic dialysis patients (45). These peptides were relatively quantified in plasma samples obtained from obese type 2 diabetes mellitus (T2DM) patients with good or poor glycemic control and compared with healthy lean and obese persons.
Plasma Samples-Blood was obtained from male healthy, normoglycemic persons (HbA 1c Ͻ 5.6%; fasting plasma glucose Ͻ 6.0 mmol/liter) categorized as lean (body mass index Ͻ 25 kg/m 2 , n ϭ 6) or obese (body mass index Ͼ 30 kg/m 2 , n ϭ 5), as well as from obese T2DM patients (HbA 1c Ͼ 6.5%, fasting plasma glucose Ͼ 6.0 mmol/liter) being under poor (HbA 1c Ͼ 7.5%, n ϭ 5) or good glycemic control (HbA 1c ϭ 6 -7%, n ϭ 6). Cohorts were matched for age (50 years) and, in case of obese cohorts, for body mass index (supplemental Table S1). All blood samples were collected between 8 and 9 am after a 12-h fast in EDTA-treated tubes, blood cells were removed from plasma by centrifugation for 10 min at 2,000 ϫ g, and the separated plasma was stored at Ϫ80°C. The study was approved by the ethics committee of the University of Leipzig (approval no. 159-12-21052012), and all subjects gave written informed consent before taking part in the study.
Nano UPLC-ESI-Linear Ion Trap-Orbitrap-MS-Plasma samples (75 ng) were loaded on a nanoAcquity UPLC Symmetry TM trap column using a flow rate of 5 l/min for 5 min and separated on a nanoAcquity UPLC BEH130 TM column (30°C) using a nanoAcquity TM UPLC System equipped with an Acquity sample manager (10 l of injection volume, full loop injection) and a nanoAcquity UPLC binary solvent manager (Waters, Eschborn, Germany). Eluents A and B were water and acetonitrile, respectively, both containing formic acid (0.1% v/v). The peptides were eluted with a linear, two-step gradient (3 3 50% eluent B in 45 min, 50 3 85% eluent B in 2 min) at a flow rate of 0.4 l/min. The column was connected via a PicoTip online nano-ESI emitter to a nanoESI-Orbitrap-MS (LTQ Orbitrap XL ETD) operated in positive ion mode and controlled by Xcalibur 2.0.7 software (Thermo Fisher Scientific, Bremen, Germany). Mass spectra were acquired in the Orbitrap at a resolution of 60,000. Tandem mass spectra (MS/MS) were recorded for the six most intense signals (z Ն 2) of an Orbitrap survey scan by data-dependent acquisition in the linear ion trap. Tandem mass spectra were analyzed with Sequest against the SwissProt database using Proteome Discoverer 1.1.0.263 (Thermo Fisher Scientific). Amide-AGE-modified peptides identified with an Xcorr below 2.20 for doubly and 3.75 for triply protonated ions were manually confirmed (49). Relative, label-free quantification relied on the integration of extracted ion chromatograms (m/z Ϯ 0.02) for each of the three technical replicates acquired per sample (supplemental Table S2).
Statistics-Cohort comparison, statistical calculations, and graphical display were performed with the software GraphPad Prism 5.02 testing the technical replicates' averages within a cohort first for normal distribution (Kolmogorov-Smirnov test, ␣ ϭ 0.05). If both cohorts to be compared showed normal distribution, a t test was performed to assess the significance of differences. Otherwise, the nonparametric Mann-Whitney test was used (both two-sided, ␣ ϭ 0.05). Correlation analysis was performed using linear regression.
Quantification of Amide-AGEs in Plasma-Seventeen of the identified amide-AGE-modified peptides could be quantified by the peak areas displayed in the extracted ion chromatograms in all plasma samples with relative standard deviations among the technical replicates typically below 60%, except for DRQC diox K acetyl YIW ox GQK, which was not detected in samples 17, 19, and 22. Peptides EFAK acetyl EIDISCVK and LK acetyl C ox DEW diox SVNSVGK were detected in all or only seven samples, respectively, but always too weak for quantification (signal to noise ratio Ͻ 10). The physiological state of the subjects (normoglycemic lean or obese, well or poorly controlled T2DM) did not affect the quantities of 15 of the 17 peptides (p Ͼ 0.05, average differences Ͻ 50%; supplemental Fig. S1). In contrast, LK acetyl C cam ASLQK (HSA 222-229, processed: 198 -205) allowed distinguishing poorly controlled diabetic patients from lean healthy persons (p ϭ 0.017, without highest data point: p ϭ 0.038; Fig. 3A). Moreover, SK formyl -EQLTPLIK from apolipoprotein A-II (Fig. 3B) was able to discriminate poorly controlled patients with T2DM from all other cohorts (all p Ͻ 0.002). To ensure that these higher levels in poorly controlled patients with T2DM were not associated with an elevated level of the protein (i.e. similar modification rate, but higher protein quantities), nonmodified peptides corresponding to both proteins were quantified as well. HSA quantification relied on 21 non-AGE-modified peptides identified with high and one peptide (LKC cam ASLQK) with medium confidence scores (49) (all peptides ranked at position 1), whereas five peptides of medium confidence were chosen for apolipoprotein A-II (supplemental Table S3). Because the plasma levels of both proteins (assessed via the majority of the non-AGEmodified peptides; supplemental Fig. S2) did not differ between the groups, the increase in modified peptides can be attributed solely to elevated modification degrees at both sites. The modification rate of LKC cam ASLQK (HSA) was even so high that the quantity of the unmodified peptide decreased in the poorly controlled patients with T2DM relative to all other cohorts albeit not significantly (p Ͼ 0.05; supplemental Fig. S2G). Most likely, the lower content of the unmodified peptide was not attributed completely to the acetyl-AGE, but to other AGE modifications as well. The obtained differences in the cohort averages among the background of a limited cohort size (n) and the observed intracohort standard deviations () were statistically evaluated for the minimum meaningful difference (d). The minimum n required to demonstrate meaningful d with typical confidence levels (␣ ϭ 0.05, ␤ ϭ 0.2) was calculated with the transposed form of formula n ϭ 16 ϫ 2 /d 2 . Indeed, the minimum d calculated for SK formyl EQLTPLIK between T2DM_LT and T2DM_MT was by a factor of 0.75 smaller than the observed differences of the cohort averages.
Correlation analysis with the patients' clinical data (supplemental Table S1) showed either none or only weak associations to body mass index, plasma protein concentration, or HbA 1c except for SK formyl EQLTPLIK (apolipoprotein A-II), which strongly correlated to HbA 1c (R 2 ϭ 0.79; Fig. 4).

Discussion
Detection of Amide-AGEs in Diabetic Plasma-The structures of peptides comprising N ⑀ -acetylated, formylated, and glycerinylated lysine residues can be easily determined by tandem mass spectrometry, because all three modifications are stable against collision-induced dissociation (50). Thus, standard MS-based proteomic approaches can be applied to identify accordingly modified sites. Gas phase fractionation of pooled plasma samples was sufficiently sensitive to identify 19 modified sites in 11 plasma proteins. Most common were acetylation and formylation sites among the identified amide-AGEs, which is in accordance with the ratio of free AGE-modified amino acids in healthy and uremic plasma (45). Other modification sources apart from nonenzymatic glycosylation might be enzymatic reactions or acetylating agents (e.g. acetylsalicylic acid). Lys 223 is the first site that was identified in HSA to be acetylated by acetylsalicylic acid (51) and was, together with Lys 549 , detected in "pure" HSA products (52). Additionally, acetylated Lys 88 was detected in nondiabetic plasma even prior to incubation with acetylsalicylic acid (53), but none of the other sites observed after HSA or plasma incubation with acetylsalicylic acid were found (52,53). Enzymatic acetylation (and formylation) are regulatory functions in nucleus, liver cell cytosols, and mitochondria (54), but such activities have not been reported to our knowledge in plasma. In contrast, all HSA and apolipoprotein A-II sites identified here as modified by amide-AGEs are known to be glycated in plasma from patients with T2DM (55,56), indicating that these sites are susceptible to glycation finally leading to AGEs.
Relative Label-free Quantification of Amide-AGE Peptides in Plasma-Fifteen peptides (90%) were present in comparable quantities in all plasma samples indicating a relatively constant basic level of AGEs in human plasma that is not influenced by obesity and diabetes. This view is supported by studies at the amino acid level of plasma protein hydrolysates that detected similar global AGE levels in healthy persons (45,57).
Most detected AGE modification levels were not influenced by the patients' glycemic status considering the HbA 1c level. This is probably attributed to the different formation pathways of the N ␣ -(fructosyl) moiety, because hemoglobin is directly modified by glucose, whereas the studied N ⑀ -AGEs require first glucose degradation prior to alkylation of lysine residues. Further parameters probably contributing to the lower sensitivity in reflecting the blood glucose level are the different protein localizations (intracellular versus plasma), protein life times, and the site-specific susceptibilities to undergo modification. Thus, N ⑀ -AGEs are likely long term markers. Indeed, two amide-AGE modification sites were present at elevated levels in diabetic patients with poor glycemic control, most likely representing highly susceptible residues. Additionally, the generally

Amide-AGE-modified peptides identified via nanoUPLC-ESI-Orbitrap-MS/MS in pooled plasma of five T2DM patients under poor glycemic control
No. poor correlation of the modification degrees among different sites within one protein (data not shown) emphasizes the sitespecific susceptibilities. This might be attributed to nearby residues or more generally the local environment, i.e. reagent accessibility and nearby functional groups affecting the pK a of the Lys residue or acting as local acid/base catalysts in glycation reactions (56,58,59). Furthermore, the differences were relatively moderate (3-6-fold compared with the other cohorts) and were detected only in one peptide per modification type, which suggests that the changes would not be detectable with an analytical approach relying on protein hydrolysis. Hence, AGE modification sites appear to be highly relevant as biomarkers and to understand the pathology of diabetes. Compared with the glycemically poorly controlled patients with T2DM, the well controlled cohort showed SK formyl -EQLTPLIK levels comparable with healthy controls (Fig. 3B), illustrating on the molecular level that glycemic control does not only reduce protein glycation levels but also prevents elevated advanced glycation degrees at sites prone to AGE modifications. Thereby, the T2DM-associated risk to develop AGErelated severe complications can be reduced, which has been previously supported by clinical studies (60). The two amide-AGE modification sites identified here closely resemble poorly controlled patients with T2DM and thus might be helpful to monitor the risk of late complications in diabetes. However, the three individuals showing much higher LK acetyl C cam ASLQK levels compared with all other participants (patients 3, 8, and 14), underwent prescribed acetylsalicylic acid treatment, which most likely caused the modification increase and thus may impact its biomarker value.
In conclusion, we demonstrate that bottom-up MS-based proteomics is sufficiently sensitive to detect protein-bound amide-AGE modifications in human plasma without the need FIGURE 2. A, total ion current (gas phase fractionation for m/z 550 -600) of plasma pooled from five T2DM patients under poor glycemic control. The zoomed region shows the extracted ion chromatogram (m/z 592.82-592.86) with signals identified as SK formyl EQLTPLIK (m/z 592.8457, z ϭ 2, t R ϭ 25.6 min, apolipoprotein A-II), IQNILTEEPK (m/z 592.8289, z ϭ 2, t R ϭ 24.0 min, serum paraoxonase/arylesterase 1), and DVEDAHSGLLK (first isotopic peak of m/z 592.3051, z ϭ 2, t R ϭ 26.3 min, N-acetylglucosamine-1phosphotransferase subunits ␣/␤). B, the MS/MS and the assigned b and y signals of SK formyl EQLTPLIK. Shown are average peak areas from three technical replicates (dots) and cohort averages Ϯ standard deviation (lines). Significance levels were determined by unpaired, two-sided t tests or nonparametric Mann-Whitney tests (depending on whether values were normally distributed within a cohort or not) and are indicated with asterisks. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001; ****, p Ͻ 0.0001. . Correlation analysis of SK formyl EQLTPLIK quantity with HbA 1c level. Average peak areas from three technical replicates Ϯ standard deviation are shown for a total of 22 plasma samples together with a linear regression curve (R 2 ϭ 0.79) and its 95% confidence band.
for AGE enrichment strategies, which simplifies the analysis and improves the robustness. Because of the site-specific response to the glycemic status in patients, which were most likely subtissue level changes, we conclude that site-specific detection and quantification of AGEs may prove very useful in AGE-related biomarker discovery and pathology research. However, validation of these results in larger cohorts and with more robust quantification methods (e.g. LC-multiple reaction monitoring) is required in future. Those studies could probably reveal whether and which AGE sites could be suitable as biomarkers reflecting the disease severity in a graded manner, which could be useful to trace or even predict disease progression toward severe complications. As further amide-AGE sites were detected, albeit at too low confidence, further fractionation via orthogonal separation techniques like hydrophilic interaction chromatography might extend the range of detectable modification sites even further.
Author Contributions-M. B. designed the cohorts, provided all plasma samples, and contributed to data interpretation. A. F. acquired the LC-MS and contributed to everything. U. G. conducted data analysis and interpretation and wrote the manuscript together with R. H. R. H. planned the studies, received funding, contributed to study design and data interpretation, and wrote the manuscript together with U. G. All authors reviewed the results and approved the final version of the manuscript.