Identification of a New Cross-link and Unique Histidine Adduct from Bovine Serum Albumin Incubated with Malondialdehyde*

Malondialdehyde, acetaldehyde, acrolein, and 4-hydroxynonenal are all products of fatty acid oxidation found in the fatty streaks of atherosclerotic arteries due to a lack of antioxidants and an increase in glycation products. Previously identified cross-links derived from these molecules have nearly always required more than one molecule of each type, although this is physiologically less likely than a reaction involving a single molecule. Here we provide indirect but strong evidence for a malondialdehyde-derived cross-link requiring just one malondialdehyde molecule to link arginine and lysine, giving 2-ornithinyl-4-methyl(1ϵ-lysyl)1,3-imidazole following a 4-day incubation of albumin with 8 mm malondialdehyde. This cross-link was identified as its partial degradation product Nϵ-(2-carboxyl,2-aminoethane)-Nϵ-methanoyl-lysine by NMR and mass spectrometry. Analysis of plasma from treated diabetic patients revealed that one patient levels had as high as 0.46%, 0.67% of their lysine/arginine residues modified by this cross-link, although others had lower levels. Alkaline hydrolysis of serum albumin also revealed two acid-labile malondialdehyde adducts of histidine in significant quantities, the isomers 4- and 2-ethylidene-histidine. These constituted up to 0.93% of the histidines in treated diabetic patients. Although collagen is readily cross-linked by malondialdehyde, none of these particular products could be found in incubations of collagen with malondialdehyde.

Glycated collagen promotes the oxidation of polyunsaturated fatty acids to a myriad of reactive aldehydes (1,2). Many of these have only one functional group, leaving the most toxic major products as 4-hydroxynonenal and malondialdehyde (MDA), 1 the latter being the focus of this paper. These oxidation products, mainly derived from the fats carried by lipoprotein, are implicated in the progression of atheroma by reacting with the collagen arterial wall.
Compared with many aldehydes, MDA is relatively nonreactive at neutral pH as it has a pK a value of 4.46, above which it favors the enolate salt form (3) stabilized by a conjugated ⌸-bond system. Resonance stabilization reduces the elec-trophilicity of MDA, decreasing its reactivity with proteinbased amine groups such as lysine and arginine, for example, in comparison with reaction with glutaraldehyde. Thus, MDA has a half-life in vitro with 10 mM lysine of approximately 2 days (3). However, free MDA is removed from the bloodstream much faster, having a half-life of ϳ2 h in rats (4), with 2-propenal adducts being detected in urine (2). Nonetheless, a significant proportion of MDA has a longer half-life in vivo because it binds to protein amine groups mainly as a temporary and unstable 3-iminoprop-1-en-1-ol adduct, ␤-lysylaminoacrolein (5). The presence of these labile MDA adducts is supported by various ways of measuring MDA, which indicate that at least 80% MDA in tissues is bound reversibly to protein in vivo (5).
Incubation of collageneous rat tail tendon in a 10 mM MDA solution for 24 h at 37°C and pH 7.4 induces structural changes in collagen, increasing both the mechanical brittleness and insolubility of the fibers and forcing the conclusion that extensive cross-linking has occurred. We have shown that MDA and acetaldehyde can react with collagen to give a potential cross-linking dihydropyridine product (3), and others (6,7) have confirmed this finding. However, under the same conditions, 70 mM MDA hardly cross-links 25 mg/ml bovine serum albumin (BSA) (8), 2 as demonstrated by the absence of an observable dimer band on SDS-PAGE gels after this time period. On the other hand, heavy cross-linking of BSA is observed when the reaction is carried out in the presence of acetaldehyde but not acetaldehyde alone, suggesting synergy between the two reagents (8). Several MDA cross-links have been proposed involving synergy with other reactive agents, but MDA crosslinks of collagen may also involve reaction with the existing enzymatic cross-links (3).
Other potential cross-links and MDA-derived adducts beyond ␤-LAA commonly require more than one MDA molecule being required to form the proposed structure, even in the presence of acetaldehyde (3,6,7,9). Such a reaction would be less likely under physiological conditions because MDA is at very low serum concentrations, and even in a "glycation hotspot," the environment of an atherosclerotic plaque (10) would be competing with many other oxidized fatty acid or sugar products for reactive lysine. In addition, blood clearance rates of even lightly glycated proteins or erythrocytes are considerably faster than normal (11,12). This finding suggests that either the scavenger receptors will remove MDA-adducted proteins long before a second MDA can react to form any cross-linking structure or that other molecules will react with bound MDA first rather than a second MDA molecule. Therefore, physiologically important MDA cross-links are most likely to involve just one MDA molecule. It is also possible that the rapid formation of MDA cross-links only occurs in the presence of a higher concentration secondary metabolite such as acetaldehyde. Well characterized protein cross-links formed in vivo follow these patterns. For example, pentosidine and DOGDIC (N 6 -{2-{[(4S)-4-ammonio-5-oxido-5-oxopentyl]amino}-5-[(2S,3R)]-2,3,4-trihydroxybutyl]-3,5,dihydro-4H-imidazol-4-ylidene}-L-lysinate) are derived from a single ribose or glucose molecule cross-linking lysine and arginine (13,14), whereas methylglyoxal-lysine dimer and glyoxal-lysine dimer are derived from two lysines, methylglyoxal or glyoxal, and a secondary metabolite, possibly acetaldehyde (15).
Previous studies employing acid hydrolysis did not reveal a cross-link involving a single MDA molecule under pseudo-physiological conditions in vitro (16). Here, we report some success in an attempt to isolate an MDA cross-link using alkaline hydrolysis of MDA reacted with both BSA and collagen.

MATERIALS AND METHODS
Reagents-All of the chemicals were purchased from Sigma or Aldrich unless otherwise specified. High pressure liquid chromatography grade solvents were all used pre-filtered (0.22 m) and were obtained from Rathburn Chemicals (Walkerburn, Peebleshire, Scotland, United Kingdom).
Excess fat and hair were scraped from the skin of 9-month-old guinea pigs. The skin was then shredded in a Moulinette homogenizer to yield a crude collagen preparation. Fats were extracted by stirring in 1 liter of 2:1 chloroform:methanol overnight, the slurry was centrifuged, the supernatant was removed, and the remaining material was washed three times in methanol and freeze-dried prior to reaction with MDA. BSA (fat-free, 96%ϩ) was purchased from Sigma (A-6003).
Reaction of MDA with Proteins-Reactions were carried out without prior sodium borohydride reduction and in a metal-free environment to minimize cross-link and oxidation artifacts (5,18,19) including the oxidation of MDA itself. Samples were stored frozen prior to subsequent analysis.
Fat-free BSA (15 g) or crude collagen (15 g) was reacted with 8 mM MDA in 150 ml of 100 mM Chelex-treated phosphate buffer (pH 7.4) (19,20) in the presence of 0.05% sodium azide and 1 mM diethylenetriaminepentaacetic acid under sterile conditions for 48 h at 37°C. After 48 h, another 8 mM MDA was added and the reaction continued for an additional 48 h.
MDA was removed from the BSA by a 3-day dialysis at 4°C against 2 ϫ 5 liters of 0.05% sodium azide in water and removed from the collagen by five washes of water containing 0.05% sodium azide. Control incubations of BSA in the absence of MDA for 0 or 4 days were carried out to ascertain whether BSA contains or forms non-standard amino acids under these conditions. At the same time, 1 g of BSA or collagen in 10-ml buffer was reacted using 450 KBq of 14 C 2 MDA.
Sample Identification-Hydrolyzed samples were assigned a threeletter code according to their nature. The first letter denoted whether the sample was from collagen (C) or BSA (B), the second letter denoted whether it was a control (C) or MDA-incubated sample (M), and the last letter denoted whether it was acid (A) or alkaline hydrolyzed (L). Thus, eight samples were generated named CCL, CCA, CML, CMA, BCL, BCA, BML, and BMA (Scheme 1). The additional sample BMLm refers to a larger scale alkaline hydrolysis of the incubation of BSA (15 g) with non-radioactive MDA.
Alkaline-hydrolyzed plasma samples from five diabetic patients were labeled Pa, Pb, Pc, Pd, and Pe. Specific fractions from semi-preparative ion exchange column separations were then appended to this to give nomenclature such as Pa 56 -57.
Acid or Alkaline Hydrolysis-Untreated plasma (4 ml) or protein samples from in vitro work were hydrolyzed at 15 mg/ml in 1 M barium hydroxide at 108°C for 48 h. The alkali were allowed to cool to just below boiling point and neutralized by the addition of a slight excess of cold 2 M ammonium hydrogen carbonate. The barium carbonate solid was removed by centrifugation, whereas the excess ammonium hydrogen carbonate and secondary ammonium carbonate product were removed by freeze-drying. For comparison, samples from in vitro incubations were also hydrolyzed at 10 mg/ml in 6 M HCl at 110°C for 18 h and then freeze-dried. Residues were re-dissolved in 0.1 pyridine/formate buffer (pH 2.9) at 20 mg/ml prior to ion exchange chromatography.
Semi-preparative Ion Exchange Chromatography-Hydrolysates derived from 100 mg of protein were loaded onto a 600 ϫ 12-mm Duolite 225 carboxymethyl polystyrene resin cation exchange column at 60°C and employed pyridine/formate buffers to separate components by charge as described previously (21). When using radioactivity, the column fractions were counted using a LKB Wallac 1219 Rack Beta scintillation counter and radioactive peaks located in relation to standard amino acids as described earlier (16).
All ϳ320 mg of the protein from alkaline hydrolysis of 4-ml human plasma were loaded onto the column in one run. Plasma sample fractions were freeze-dried, and fractions of interest were analyzed directly by NMR, mass spectrometry, and amino acid analysis.
Preparative Ion Exchange Chromatography-Sample BMLm was loaded at 15 ml/min onto a 500 ϫ 50-mm Duolite 225 cation exchange column pre-equilibrated in the same 0.1 M pyridine/formate buffer (pH 2.9) as the semi-preparative system. 1 liter of this buffer followed by 2 liters at pH 3.7 (0.75 M pyridine), 2 liters at pH 5.2 (1 M pyridine), and finally 3 liters at pH 8.0 (2 M pyridine) buffers were used to elute amino acids in a stepped gradient. Fractions of 1 liter were taken, and the pyridine/formate was removed by repeated rotary evaporation.
Fractions of interest were dissolved in the pH 2.9 pyridine/formate buffer, and the pH was re-adjusted to 2.9. These fractions were then re-chromatographed using the smaller ion exchange column (600 ϫ 12 cm) for improved resolution of the neutral and basic amino acids while incorporating a small amount of radioactive sample as a tracer. Pyridine/formate was again removed by rotary evaporation, and the sample was re-dissolved in 0.1% heptafluorobutyric acid for reversed-phase high pressure liquid chromatography (see below).
Analytical Amino Acid Analysis-Aliquots from the initial protein hydrolysis samples and purified products from C18 reverse-phase column chromatography were analyzed on an Alpha Plus automatic amino acid analyzer (Amersham Biosciences) using a standard buffer gradient with a Dionex AI 450 data collection system.
C18 Reverse-phase Column Chromatography-Radioactive peaks selected for further investigation were applied to a 250 ϫ 21.5-mm Spherisorb ODS1 C18 semi-preparative column at 10 ml/min using a gradient increasing from 0 to 30% acetonitrile over 60 min. Column effluent was monitored by UV at 214 nm, 10-ml fractions were collected, and radioactive peaks were identified for freeze-drying.
NMR and Mass Spectrometer Analysis-The three final fractions C1, C2, and C3 from high pressure liquid chromatography separations of in vitro incubations were prepared in 600 l of 10% D 2 O with 20 M trimethylsilylpropionic acid as an internal NMR standard. Using water suppression by a long presaturation pulse, a series of 128 or 256-scan one-dimensional runs were done on a 500-MHz Jeol NMR spectrophotometer in neutral (pH 7.4, 70 and 20°C) and acidic conditions (pH 3.7, 20°C). These were supplemented by two-dimensional total correlation (TOCSY, mixing time 60 ms, pH 7.4, 20°C) and correlated spectroscopy experiments (COSY, pH 7.4, 20°C) for spin-system identification. Electrospray mass spectroscopy was carried out on a Micromass instrument using acetonitrile as the carrier solvent.
The ion exchange purified plasma sample Pa 56 -57 contained the most material eluting at similar position on the ion exchange column to fraction C. It was prepared in 550 l of 10% D 2 O with trimethylsilylpropionic acid as stated above. Using a gradient pulse for water suppression, a two-dimensional total correlation experiment was run with an accompanying one-dimensional experiment on a 600-MHz Bruker NME spectrophotometer in slightly acidic (mixing time 60 ms, pH 5.7, 20°C) conditions. This and several other plasma samples were subsequently analyzed using a ThermoFinnigan LCQ classic ion trap mass spectrometer.

Identification of the Products from Reactions of MDA with BSA or Collagen
The procedures undertaken and results outlined below are summarized in Scheme 1.
Semi-preparative Ion Exchange Chromatography- Fig. 1 shows the results of semi-preparative cation exchange analysis of hydrolysates of radioactive products from samples BML and CML. In addition to a major peak of unbound radioactivity (peak A), two products were reproducibly identified: peak B, which was found in both collagen, and BSA, whereas peak C was unique to BSA. Peak B was not pursued further, because it co-eluted with the bulk of the amino acids in the preparative run and no adduct of lysine, arginine, or histidine has been seen to elute so early (concurrent with glycine and alanine) on an ion exchange column. Therefore, further separations focused on peak C, which eluted between tyrosine (fractions 46 -47) and lysine (fractions 68 -69), in the region where most basic amino acid-derived cross-links have been identified. Controls (not shown) showed that the previously characterized N-pyrimidinylornithine from acid hydrolysis (16) eluted at fractions 48 -49 and thus cannot account for any of peaks A-C.
Amino Acid Analysis of Hydrolyzed Proteins- Fig. 2a, bottom three traces, shows a comparison between acid (sample BMA) and alkal (sample BML) hydrolyses of BSA reacted with MDA before separations plus an amino acid standard for reference (STD). Sample BML contained a new peak at 69.0 min (Fig. 2a, arrow) slightly earlier than the peak of histidine in the standard, which is destroyed by the alkaline hydrolysis protocol. Histidine is seen in sample BMA as a peak at 69.7 min. Apart from the new peak at 69.0 min, no significant differences between sample BML and the control sample BCL (data not shown) could be seen at this stage of the analysis. Furthermore, no significant differences could be observed between samples CML and CCL by amino acid analysis. Identification of MDA adducts from acid hydrolysis has already been done (16), and repeat amino acid analysis of samples BCA, CCA, and CMA showed no unexpected peaks.
From peak integration, the new amino acid observed in sample BML at 69.0 min is equivalent to 4.25 residues/BSA molecule, assuming that its ninhydrin equivalent is the same as histidine (cf. 1.08, leucine). BSA contains 17 histidines/molecule. From its elution position, this 69-min peak could be the same as the radioactive peak C in Fig. 1. Peak B in Fig. 1 would probably elute somewhere in the 30 -45-min region of these amino acid traces. The remaining amino acid traces in Fig. 2, SCHEME 1. Overview of separation protocols. Thick lines show procedures undertaken to generate and purify the components from MDA reactions and human plasma. Thin lines link these procedures to diagrams from sample analysis shown later in the paper. AAA, amino acid analysis; M/S, mass spectroscopy; dns, data not shown. See the section on sample identification for other abbreviations. FIG. 2. Amino acid analysis of purified protein hydrolysates and isolated products. a, analyses from BMA4 and BML4 incubations. These are shown together with analyses of the radioactive components C1-C3 later separated out by high pressure liquid chromatography (HPLC) (see Fig. 3). The letters CVMILYFHK refer to standard amino acids. Or, ornithine; Hylys, hydroxylysine. b, analysis of 5:1 (C1i) and 1:5 (C2i) mixture ratios of C1:C2 shows that they can be resolved using a modified gradient (20). c, analyses from selected ion exchange chromatography fractions of plasma samples Pa and Pb.
shown here for easy comparison, are from purified products and will be referred to later.
Preparative Ion Exchange Chromatography of Sample BMLm-Fractions from the preparative ion exchange column run were evaluated by amino acid analysis. The first of two fractions eluted by the pH 5.2 buffer contained tyrosine and lysine, respectively (data not shown). Because peak C eluted between tyrosine and lysine on the analytical column ( Fig. 1), these two fractions, which between them should contain peak C, were freeze-dried and further separated on the semi-preparative ion exchange column to obtain larger quantities of fraction C. The fractions were spiked with radioactivity from the BML sample.
C18 Reverse Phase Column Chromatography-All of the fractions containing peak C from incubations BML and BMLm were loaded onto a 250 ϫ 21.5-mm C18 semi-prep column over two runs. Three peaks of radioactivity were detected in peak C (Fig. 3, trace a), giving peaks C1, C2, and C3. An N-pyrimidinylornithine standard (16) detected by light absorbance gave a single peak in a region distinct from C1, C2, and C3 (Fig. 3,  trace b). Peaks C1, C2, and C3 were prepared for analysis by NMR and mass spectrometry.

NMR and Mass Spectrometer Analysis of Peaks C1-C3
Peak C1-Mass spectroscopy of this sample gave a single ion at 182.05 (M ϩ 1) molecular weight (Fig. 4, left traces). The NMR spectrum (Fig. 5) was almost identical to histidine but with one of the 4H or 2H ring protons missing and a new CH-CH 3 spin system as identified by two-dimensional COSY experiments (data not shown). Replacement of the 2H or 4H proton of histidine with a ϭ CH-CH 3 group plus the removal of a ring N-H proton gives a molecule, either 2-ethylidene histidine (2EH) or 4-ethylidene histidine (4EH), respectively (Fig.  6), that exactly fits both NMR and mass spectroscopy data. The remaining 2/4H proton was observed to shift in a manner akin to a standard histidine proton when the pH was adjusted to pH 3.7 (data not shown).
2EH and 4EH may be artifacts of alkaline hydrolysis of an MDA-histidine adduct rather than being a direct product from the reaction among the MDA degradation product, acetaldehyde, and histidine (Fig. 6). All of these structures are more stable Schiff bases than the lysine adduct ␤-LAA because of additional resonance stabilization from the double bonds in the histidine ring.
Peak C2-This fraction contained three sets of peaks by NMR (Figs. 5 and 7) as identified by integrals of their proton peaks. The major one was the other variant of the histidine adducts described above. The ring proton is further downfield (7.83 ppm, peak rHЈ in Fig. 5) than the molecule in peak C1 (7.72ppm), making this likely to be 2EH and leaving the Peak C1 adduct as 4EH, although these ring proton shifts are closer together compared with unmodified histidine (8.12 and 7.14 ppm). Similar to peak C1, the mass spectrum of C2 showed a strong peak at 182.05 (Fig. 4). By NMR, the total yield for the both types of histidine adduct was 2.1% of available histidine before the MDA reaction started, of which 31% is the isomer seen in fraction C2; however, there will inevitably have been significant losses during purification. The original amino acid compositional analysis of BML (Fig. 2) suggests that up to 25% of the histidines could have been modified.
The second set of peaks was a very small lysine impurity (ϳϽ3%) seen only when separated out by two-dimensional NMR in Fig. 7. This lysine impurity was not a significant feature on the mass spectrum of C2 (Fig. 4) in comparison with other peaks.
Unaccounted by either of these first two peak sets, peak C2 contained also two mass spectroscopy peaks at m/z of 243.91 and 262.10 (Fig. 4). The final set of NMR peaks correlating to these masses in peak C2 was observed in several groups, which can be seen more clearly in the total correlation two-dimensional spectrum of this fraction (Fig. 7). Two sets of correlated shifts with equivalent proton intensities could be identified as N ⑀ -adducted lysine side chains. These sets had parts/million shifts of 3.73 (␣) , 1.88 (␤) , 1.49/1.42 (␥) , 1.77 (␦) , and 3.14 (⑀) annotated as 2LЈ␣-⑀ (Fig. 5) and 3.72 (␣) , 1.87 (␤) , 1.38/1.32 (␥) , 1.68/ 1.64 (␦) , and 3.44/3.32 (⑀) annotated as 2L␣-⑀. They also had a complex CH-CH 2 spin system (Fig. 5, 2L, 2LЈ, ) with fine structure that could only be resolved by repeating the onedimensional spectra at 70°C (Fig. 5, inset). The coupled proton shifts of 3.98 (H) and 3.83 (2H) from this CH-CH 2 spin system are a triplet and doublet at 70°C. For this to occur, there must be two isomeric forms of the molecule in slow exchange (Ͻ2 s Ϫ1 ) on the NMR time scale at the lower temperature that go into fast exchange at 70°C (Ͼ10 s Ϫ1 ). In other words, these two sets of peaks denoted 2L and 2LЈ are two isomers of the same molecule. Finally, two CH singlets were observed in the aromatic region of the spectrum, the one at 8.15 ppm (Fig. 5,  2L, 2LЈ).
Daughter ion spectra were obtained from the mass spectroscopy peaks initially observed at m/z 244 and 262 (Fig. 4, M ϩ 1). In the same manner that the two sets of NMR peaks (2L and 2LЈ were from the same molecule), it can be seen that the initial two peaks with masses of 261.99 and 244.31 are probably derived from the same molecule. Only the ion ratios differ in their fragmentation patterns.
Lysine, the recognizable backbone of this new molecule in peak C2, has a mass of 146 daltons (C 6 H 14 N 2 O 2 ). To obtain the observed m/z values of 244 and 262, the masses of 97 and 115 need to added to lysine, respectively, along with the ionizing H ϩ . This indicates that an extra nitrogen atom must have been added in addition to the radioactive 14 C 2 MDA, which gives the observable radioactive peak. Furthermore, there cannot have been any contribution from the reactive MDA breakdown product acetaldehyde because that would result in methyl groups being present in the final product, whereas no methyl groups are observed in peak C2. MDA must therefore be responsible for the CH-CH 2 spin systems seen at 3.98/3.83 ppm, which implies that the third carbon of the MDA has been oxidized to a carboxyl group. From this finding, it can be deduced that the 244 ion involves the addition of C 4 H 3 NO 2 to lysine, whereas the 262 ion fragment includes an additional water molecule. The fourth carbon is required to give the observed aromatic CH proton.
Three of the additional carbons in this lysine adduct come from MDA; thus, possible sources for the fourth carbon that has a downfield proton attached (8.15/8.02 ppm depending on isomer) and the extra nitrogen were investigated. Protein mainchain nitrogen atoms are unreactive, so the nitrogen must  Fig. 6. Peaks denoting the two stereoisomers of NNL shown in Scheme 2 are seen in fractions C2 and C3 and correspond to Greek annotations prefixed by 2L or 3L. Each stereoisomer of NNL has two observable forms depending on whether the amide bond shown in Scheme 2 is cis or trans, the minor form being annotated with an apostrophe. Between 4.4 and 4.8 ppm, fractions C1 and C2 are shown from NMR runs done at the different temperatures shown to separate the proton from bulk water. An inset of sample C2 from 3.65 to 4.05 ppm demonstrates the fast-exchange averaging of the cis and trans forms (2L and 2LЈ) of NNL as temperature is increased. Peaks separated by a bigger parts/million shift remained in slow to intermediate exchange. For the aromatic region of the spectrum (7.6 -8.3 ppm), Peak Q is probably the ring proton from NNL (Scheme 2) but cannot be confidently assigned. The singlet peak at 3.35 ppm in all of C1-C3 is a small unidentified impurity. All of the NMR peaks from the fractions are shown apart from water, the standard, and the two other pyridine impurity peaks in fraction C3. come from a second lysine, histidine, or arginine side chain, indicating that the MDA had cross-linked two amino acids at one point, the cross-link being damaged by alkaline hydrolysis. Of these amino acids, only arginine has both the required properties of a single carbon that could conceivably be separated from the rest of the side chain and the potential to form a stable cross-link via nitrogen attack on a carbon double bond as shown in Scheme 2. This cross-link, 2-ornithinyl-4-methyl(1⑀-lysyl)1,3-imidazole (OMLI), can then rearrange and fragment to form N ⑀ -(2-carboxy,2-aminoethane)-N ⑀ -methanoyllysine (NNL), a stable diamino acid that has the required mass of 261 for peak C2 (Scheme 2). Scheme 3 shows that positive ions can be derived through chemical ionization fragmentation pathways of this molecule to fit the fragmentation spectra of C2 in Fig. 4. The accompanying neutral fragments are omitted for simplicity.
Several properties of the NMR spectra also support the conclusion that the new molecule in peak C2 is the structure NNL. The central methanoyl group is part of an amide functional group, and slow amide rotation around the C-N bond to give cis/trans-isomers can explain the two distinct structures seen by NMR at low temperature. The singlet peaks at 8.15 and 8.04 ppm can be two forms for the proton in the methanoyl group. The ratio of the two CH-CH 2 groupings could not be measured at 25°C, but it is interesting that these groups very close in part/million shift at 25°C (Ͻ0.01 ppm) became averaged at 70°C. Raising the temperature has increased the frequency of rotation enough for these peaks to report intermediate to fast exchange on the NMR time scale, whereas other paired peaks further apart in shift remained in a slow exchange system that reports both structures (data not shown).
Because mass spectroscopy and NMR data match well and can explain some unusual features, NNL can be assigned as the molecule in fraction C2. Working back, MDA must have crosslinked lysine specifically to arginine during incubation with protein because only arginine can react with ␤-lysyl aminoacrolein to form a stable link that can breakdown to NNL on alkaline hydrolysis. Therefore, an MDA cross-link is OMLI, which is destroyed by acid hydrolysis and damaged by alkaline hydrolysis to form NNL.
From the mass spectroscopy, the mass ions from peak C2,  7. Two-dimensional NMR total correlation spectra of peak C2 (left) and fraction Pa 56 -57 (right). Assignments from peak C2 are given for the cis/trans forms from one isomer of NNL (2L, 2LЈ, 2L, and 2LЈ) and also show one isomer of the histidine adduct (␣Ј, ␤ 1 Ј, and ␤ 2 Ј). Identical assignments could be made for fraction Pa 56 -57 where there are several additional peaks due to incomplete purification. The gradient suppression affected peaks near the water peak much more in the Pa 56 -57 spectrum, resulting in weaker ␣Ј to ␤ 1 Ј/␤ 2 Ј correlations. m/z 244 and 262, were again observed in peak C3 (Fig. 4). Therefore, one can conclude that peaks C3 and C2 are the two diastereoisomers of NNL where the chiral carbon is shown in Scheme 2. Each diastereoisomer also has a cis and trans form. Total yields of this damaged cross-link NNL after purification were 0.33% of the original lysine and 0.85% of the original arginine. Of that, 76% is the diastereoisomer found in peak C3. If losses of this molecule were similar to those of 2EH/4EH, the initial yields may have been an order of magnitude higher.
Amino Acid Analysis-Peaks C1, C2, and C3 were analyzed on the automatic amino acid analyzer, (Fig. 2a, traces C1-C3) in order to locate them with respect to other amino acids and glycation products. The 2EH and 4EH adducts in peaks C1-C2 gave a peak at 69 min as seen before in sample BML, distinct from the known histidine peak location at 69.7min. 2EH and 4EH could be separated by a different protocol designed to separate collagen cross-links (Fig. 2b, C1i and C2i) (22).
Amino acid data for the NNL are harder to interpret. All of the remaining peaks in the 60 -75-min range in fractions C2 and C3 are only showing trace quantities. One would expect a NNL peak at least 50% the size of 2EH/4EH from rough quantifications by NMR and mass spectroscopy, eluting very close to the adduct given its almost identical elution position on the semi-preparative ion exchange column. Conversely, in frac-tions C2 and C3 (Fig. 2a), there is an unassigned doublet peak at 41 min. This doublet could be acid-degradation product(s) of NNL, because any MDA cross-link will be acid-labile (16), and initial conditions in the amino acid analysis column are pH 2.2 and 70°C. These same acid degradation products of NNL may also be responsible for peak B in Fig. 1.

Identification of Alkaline Hydrolysis Products from
Human Plasma Table I quantifies the 2EH/4EH calculated by amino acid analysis as found in fractions from the semi-preparative ion exchange separations of 4-ml alkaline-hydrolyzed diabetic plasma samples Pa-Pe. Amino acid traces from four paired fractions are shown in Fig. 2c (Pa 56 -57, Pa 54 -55, Pb 52-53, and Pb 54 -55) where all four fractions have the 41-min and histidine adduct peaks along with some other peaks of which the major ones will be ornithine, lysine, and/or ammonia. The 41-min peak seen in fraction Pa 56 -57 (Fig. 2c), the largest observed for all of the samples, represents at least 0.46/0.67% of the total lysine/arginine residues from the Pa plasma sample should it be an NNL derivative. Unfortunately, because it is apparently an NNL degradation product, it could not be reliably quantified, although such degradation will mean that its concentration will always be underestimated.
Mass spectroscopy traces of fractions Pa 54 -55 and Pa 56 -57 are considerably more complex than those of fractions C1-C3 (Fig. 4). This is expected due to the fewer separation steps, the large number of small metabolites in plasma, and the extra peaks seen in the amino acid analysis (Fig. 2c). However, the mass ion m/z 182.1 shift position from 2EH and/or 4EH is the largest peak in fraction Pa 54 -55, whilst the mass ion m/z 262.1 mass from NNL is the largest peak in fraction Pa 56 -57. To further prove that OMLI and 2EH/4EH occur in vivo, the significant similarities between the pair of two-dimensional total correlation spectra in Fig. 7 show that molecules within plasma fraction Pa 56 -57 (right) are essentially the same as those seen in fraction C2 (left), simply differing in their relative quantities.
Of secondary interest are the masses 357.1 and 234.1 seen in the mass spectrum of fraction Pa 56 -57. The mass 357.1, although only seen as a very small peak, happens to be the right mass (356 ϩ 1) for the intact cross-link. The MS-MS fragmentation pattern, although not confirming the intact cross-link structure, is consistent with it where the main ions can be accounted for by release of H 2 O, OH Ϫ , CO 2 , and NH 2 . The mass m/z 234 is consistent with a lysine adduct similar to the NNL and is of interest because its fragmentation pattern is also similar to NNL. The 198ϩ fragment of m/z 234 is shown in Scheme 3, whereas the 130ϩ fragment could easily be a variant of the 129ϩ fragment also shown, simply with an extra proton. The mass m/z 234 could be the 198ϩ molecule with two extra water molecules added, the ring being open before MS-MS bombardment. These molecular masses will be further investigated in future work along with an attempted synthesis of the molecules described here. DISCUSSION The use of alkaline hydrolysis to search for MDA cross-links sensitive to acid hydrolysis revealed distinct differences between collagen and bovine serum albumin in their reaction to MDA. An alkali degradation product of a cross-link formed from MDA reacting with lysine and arginine was identified together with a new reaction product of MDA with histidine. SCHEME 2. Proposed mechanism for the formation of the MDA cross-link by reaction with lysine and arginine followed by its partial degradation by alkaline hydrolysis to form NNL. The damaged cross-link, NNL, was obtained in two forms with mass ions 243 and 261. The intact precursor is 2-ornithyl-4-methyl(1 ⑀ -lysyl)1,3imidazole (the designated cross-link). Protons with Greek annotations are assigned by NMR in Fig. 5. *, this bond causes the cis/transisomerization denoted by the apostrophe or lack of it in Figs. 5 and 6. The major (non-apostrophe) form is probably the trans form, but the two are not yet distinguished. **, this bond is a chiral center that becomes racemic during the alkaline hydrolysis, giving the two stereoisomers of NNL denoted as 2L and 3L. It is unknown which stereoisomer denotes each set of assignments.
Despite the known ability of MDA to extensively cross-link collagen, neither of these new compounds was present in the alkaline hydrolysate of collagen incubated with MDA, possibly because of steric constraints imposed by the helical structure of the molecule.
The Alkali-Degraded Cross-link-The identification of the alkali partial degradation product NNL is consistent with an MDA cross-link formed by reacting one molecule of MDA with lysine and arginine to give the original cross-link OMLI (see Scheme 2).
NNL was not isolated from collagen alkaline hydrolysis, but some cross-linking must have occurred to cause the collagen stiffening that follows MDA incubation (16). Our earlier work using tritiated borohydride reduction techniques (16) showed that the enzymatic cross-links in collagen reacted strongly with MDA, which could account for extra cross-linking. OMLI is also likely to be mainly an intraprotein cross-link, because incubation between MDA and BSA does not result in significant dimerization of BSA (8). Cross-linking possibilities for MDA also exist between DNA bases and may result in molecules related to OMLI.
Histidine Adducts-Although lysine and arginine have frequently been implicated in glycation cross-linking, there have been relatively few reports on histidine-based cross-linking and adducts. One related molecule, carnosine or ␤-alanyl histidine, is found in brain and long-lived tissues at concentrations up to 20 mM (23) where it has been suggested to have anti-aging and cardioprotective effect. However, its primary mechanism of action is in dispute because it apparently scavenges aldehydes and glycation products in vitro and in vivo by reacting with them (24) as well as by chelating copper to inhibit glycoxidation (25). 4-Hydroxynonenal also reacts with histidine to form an adduct (26), and recent evidence showed that the presence of histidine can block the formation of the stable 4-hydroxynonenal-induced cross-link between two lysine molecules (27). We provide evidence here that significant amounts of proteinbound histidine can also react with MDA and probably other aldehydes to form adducts. Again, type I collagen-based histidine does not appear to react, implying that the fibrillar structure of this collagen protects histidine from glycation while other exposed histidines such as those in BSA or carnosine can act as a sponge for aldehydes. This collagen effect may also be partly because of the fact that two of the already small number of collageneous histidines, approximately seven per chain depending on collagen type, are involved in formation of the mature collagen cross-link histidino-hydroxylysinonorleucine by an spontaneous reaction with a proximal immature crosslink, dehydrohydroxylysinonorleucine (28).
In vivo, the results show that five diabetic patients had between 0.22 and 0.93% of their histidine modified to either 2EH or 4EH, demonstrating that this is a potentially useful biomarker of lipid-derived chemical modification of proteins in SCHEME 3. The partial fragmentation pathways of NNL during mass spectrometry. diabetes. Further controls involving measurement of patient lipid status need to be done to verify this finding. There is no obvious reason why acetaldehyde or other aldehydes should not react with the histidine ring if MDA can. Reaction Mechanisms-Similar molecules such as acrolein or 4-hydroxynonenal cannot form cross-links related to OMLI. Reaction patterns for these molecules show that histidine, lysine, and cysteine and by inference arginine preferentially react with the C3 double bond to form Michael adducts. Lysine can subsequently react with the aldehyde group (2, 27, 29 -33). Unlike MDA, there is then no possibility of ring closure to give a stable aromatic ring-based cross-link.
One can construct schemes where any aldehyde can prevent OMLI (or other MDA-derived cross-link) formation before the second nitrogen of the arginine closes the ring simply by reacting with the free amino of the cross-link precursor first (Scheme 2), ironically acting as a cross-link inhibitor (34). Only a potentially unlikely second reaction with MDA opens the possibility of cross-linking again, probably via a dihydropyridine-type structure reported previously (3,6,8). A variety of synthetic experiments are being carried out to investigate the potential reaction mechanism proposed in Scheme 2. In particular, it will be important to investigate whether the in vivo metabolite responsible for the formation of 2EH and 4EH is likely to be MDA or acetaldehyde.