Characterisation and quantification of protein oxidative modifications and amino acid racemisation in powdered infant milk formula.

Modification of proteins in infant milk formula (IF) is of major concern to the dairy industry and consumers. Thermal treatment is required for microbiological safety, but heat, light, metal-ions and other factors may induce oxidative damage, and be a health risk. In this study protein modifications in IFs were quantified. IFs contained both reducible (disulphide) and non-reducible (di-tyrosine, lanthionine, lysinoalanine) protein cross-links. Dehydroalanine and the cross-linked species lanthionine and lysinoalanine were detected. Protein carbonyls were detected predominantly on high molecular mass materials. Oxidation products of phenylalanine (m-tyrosine), tryptophan (N-formylkynurenine, kynurenine, 3-hydroxykynurenine), tyrosine (di-tyrosine) and methionine (methionine sulphoxide) were detected, consistent with amino acid modification. Higher levels of most of the markers of protein modification were present in the hydrolysed protein brand, when compared to the conventional IF samples, indicative of increased damage during additional processing. Significant levels of racemised (D-) amino acids were present. These data indicate that amino acids in proteins in IFs are modified to a significant extent during manufacture, with hydrolysed IF being particularly prone.


Introduction
The quality of the proteins present in dairy products, and especially infant milk formula, is a major concern of the dairy industry and consumers. Thermal treatment processes during the production of powdered milk, such as pasteurisation, sterilisation and spray drying, are required to ensure microbiological safety and prolong shelf life. However, exposure to the high temperatures and other factors (light, metal ions, shear stress) during processing can contribute to protein unfolding and result in oxidation-induced changes [1,2]. Free radicals, excited states and two-electron oxidants (e.g. peroxides, singlet oxygen) can generate oxidative modifications, with proteins being especially susceptible due to their high abundance and reactivity with oxidants [2]. These reactions can result in protein fragmentation, aggregation via covalent cross-linking or hydrophobic interactions, and changes in amino acid composition and conformation [2]. Some of these reactions can result in undesired colour, flavours and impaired digestibility of the protein, its bioavailability and nutritional value, with potential adverse effects on infant growth and development [3], though this is controversial [4]. Of particular concern is the (disputed) link between infant formulas (IF) use and the development of Type 1 diabetes and other autoimmune diseases [5].
IFs, most of which are based on bovine milk, are manufactured foods that should fulfil the nutritional requirements for the normal growth and development of newborn babies. IFs provide both macronutrients, including proteins, fats and carbohydrates, and micronutrients (minerals and vitamins), that can act as a substitute for human milk [6]. Oxidation of lipids can occur during manufacturing/processing and storage, and this may contribute to, or exacerbate diseases of prematurity associated with oxidative stress [7]. However, the nature and extent of protein oxidation, structural modification and amino acid isomerisation in IFs is not fully understood.
A number of different types of oxidative modification have been reported on proteins subjected to radical-mediated, or two-electron oxidation, with some of the products being diagnostic of the initial damaging species [2]. Some products can arise via multiple pathways, and hence are only generic markers of damage [2]. Amino acid side-chains on proteins are major sites of modification, though limited backbone modification (fragmentation) has also been observed [2]. Peptide backbone oxidation can also give rise to stereochemical inversion of side-chains (conversion of native L-to unnatural D-amino acids), with the D-amino acids having altered biological activity and increased toxicity, when compared to the L-isomers [8].
Sulphur-containing amino acids (cysteine, Cys; methionine, Met; and cystine), though present at low abundance, are especially sensitive to oxidation [2,9]. Conversion of two Cys residues to the disulphide cystine is of particular importance, as this can result in (reversible) intermolecular cross-links, and changes in protein structure and function [9]. Heat treatment of whey proteins, in either the absence or the presence of oxidants which exacerbate such changes, can generate aggregates, with b-lactoglobulin being particularly susceptible [10]. Light exposure can also induce both reversible (disulphide) and irreversible aggregate formation [11,12].
Elevated levels of a number of these markers (e.g. di-Tyr, DOPA, protein carbonyls, alcohols) have been detected in milk powders (e.g. [1,14,15]) and also in a number of human pathologies, including neurodegenerative conditions (Alzheimer's and Parkinson's diseases, ALS), atherosclerosis, rheumatoid arthritis, cystic fibrosis, asthma, diabetes, and some cancers [2,16]. However, the relationships between the nature and concentration of modified materials present in foodstuffs (e.g. milk powders), dietary intake of these materials, plasma and tissue levels, and disease development and progression remain to be resolved.
The current study set out to quantify the extent of protein oxidation products (specific side-chain species, generic markers such as carbonyls and interprotein cross-links) in commercial powdered IFs, as well as the levels of protein aggregates and racemised amino acids. In addition, we hypothesised that proteins which had been enzymatically digested to peptides, as used in the preparation of hypoallergenic IF, would contain higher levels of amino acid modifications due to both the greater accessibility of the side-chains to oxidants, and the increased processing that these samples undergo.

Materials and methods
Reagents D-Lactate dehydrogenase and b-nicotinamide adenine dinucleotide (reduced salt) were purchased from Megazyme (Bray, Ireland). Synthetic LL-and DL-LAL were purchased from Bachem (Bubendorf, Switzerland), dityrosine and NFK from Toronto Research Chemicals (Toronto, Ontario, Canada), and the anti-dityrosine monoclonal antibody (mAb; 1C3) from JaICA (Haruoka, Japan). The anti-NFK antibody was a kind gift from R. Mason and M. Ehrenshaft (NIEHS, Research Triangle Park, NC). All other chemicals used were of reagent grade from Sigma-Aldrich (St Louis, MO) Chemie GmbH (Steinheim, Germany) unless otherwise noted. Milli-Q water was obtained from a Milli-Q V R Advantage A10 V R purification system (Millipore SAS, 67 120 Molsheim, France).

Infant milk formula samples
Four different stage 1 (for 0-6 months old infants) IFs (S-A, S-B, S-C and S-D) produced by three commercial companies, representative of the Danish market, were investigated. Three of these (S-A, S-B and S-C) were produced from bovine milk, whereas sample D was a hypoallergenic formula designed for infants allergic to intact bovine milk proteins, and according to the manufacturer contained enzymatically digested proteins. The dry powders were stored at À20 C until analysed to minimise additional changes. Compositional data are reported in Supplementary Table S1. All samples were freshly dissolved in phosphate-buffered saline (PBS, pH 7.5) immediately before analysis, and adjusted to a protein concentration of 4.0 mg mL À1 unless otherwise stated. Protein concentrations were determined using the BCA assay (Pierce, Rockford, IL).

Polyacrylamide gel electrophoresis (SDS-PAGE)
SDS-PAGE analysis of protein modifications was carried out under both reducing and non-reducing conditions as described previously with slight modifications [10]. Samples were heated (90 C, 5 min) in SDS sample buffer (Invitrogen, Paisley, UK) to denature proteins, then 5 or 10 mg of proteins were loaded per well on 12% bis-Tris gels (Invitrogen , Carlsbad, CA) and electrophoresed at 200 V for 45 min. For reducing gels, 500 mM dithiothreitol at a 10 Â concentration (Invitrogen , Carlsbad, CA) was added prior to loading. Unstained protein molecular mass markers (14.4-116 kDa, Thermo Scientific , Waltham, MA) were used to determine protein masses. Bands were visualised using silver staining, with images captured using a Canon CanoScan LIDE 220 image scanner (Canon, Tokyo, Japan).

Detection and quantification of protein carbonyls
Spectrophotometric assay of protein carbonyl levels derivatised with 2,4-dinitrophenylhydrazine (DNPH) was performed as described previously, with slight modifications [17]. Briefly, 0.2 g sample was homogenised with 10 mL phosphate buffer (10 mM, pH 7.5) and two aliquots of 0.1 mL were treated with 0.5 mL ice-cold 10% (w/v) TCA to precipitate the proteins. After centrifugation, half of the protein pellet was treated with 0.5 mL 2.0 M HCl (for protein quantification), and the other half with an equal volume of 10 mM DNPH dissolved in 2.0 M HCl to quantify carbonyls. After incubation of the latter samples (40 min, 25 C in the dark), 0.5 mL of ice cold 20% TCA was added and the samples kept at À20 C for 3 min, before centrifugation at 9000 g (15 min, 4 C; Eppendorf 5415R centrifuge). Excess DNPH was removed by three washes with 1 mL of ethanol-ethyl acetate (1:1, v/v) and the samples then dried under N 2 gas. The pellets were dissolved in 1.0 mL 20 mM phosphate buffer containing 6 M guanidine HCl, pH 7.5, and the absorbance measured at 370 nm and 25 C using a SpectraMax V R i3 Platform (Molecular Devices, Sunnyvale, CA). Blank samples which did not contain DNPH were read at 280 nm to quantify protein levels. Carbonyl concentrations (nmol/mg protein) were calculated using the equation:

Quantification of thiols
Thiols (RSH) were quantified using 5,5 0 -dithiobis(2-nitrobenzoic acid) (DTNB) based on the method in [17], with slight modifications. Samples (0.5 mg protein) were incubated (10 min, 21 C) with 1.0 mL sodium phosphate buffer (10 mM, pH 7.0) containing 3.0 M urea, then aliquots (135 mL) were incubated with 15 mL freshly prepared DTNB (1.0 mM in sample buffer) in the dark for 30 min at 25 C. The absorbance at 412 nm was then determined. Thiol concentrations were calculated, after subtraction of blank values, using a standard curve prepared using glutathione (10-100 mM). Data are reported as nmol RSH mg À1 protein.

Quantification of parent amino acids and oxidation products
The amino acid composition of acid-hydrolysed protein samples was determined using a minor adaptation of a previous method [17,18] and Pro, which are not detected by this method) and the oxidation product methionine sulphoxide (MetSO) was determined from linear plots of the peak area using standard curves constructed using authentic commercial standards.
Other oxidation products were quantified as described previously with slight modifications [17,18]. Briefly, the samples were prepared as described above under vacuum. Samples were separated at 30 C using a Kinetex V R C18 column (as above) at a flow rate of 0.8 mL min À1 , using a binary gradient with buffer A (100 mM sodium perchlorate/10 mM H 3 PO 4 ) and buffer B (80% v/ v methanol). The gradient elution profile and programed excitation and emission wavelengths (k ex /k em ) of the fluorescence detector are reported in Supplementary  Table S3. Oxidation products were identified on the basis of their retention times compared to commercial standards, and quantified on the basis of peak areas, using standard curves constructed using authentic standards. Data analysis was carried out using Shimadzu Lab Solutions Browser software. The concentrations of amino acids and oxidation products were expressed as g of amino acid 100 g À1 protein, and nmol oxidised product mg À1 protein, respectively. The data for specific amino acid modifications was calculated by determination of the concentration of the oxidised species relative to the total (modified plus parent) expressed as a percentage. For di-Tyr, the data were corrected for the presence of two Tyr residues in the oxidised product.
DHA quantification DHA was quantified as described previously with slight modifications [19]. Samples (10 mg protein) were hydrolysed at 110 C for 120 min in PicoTag hydrolysis vessels using 0.5 mL 1.5 M hydrochloric acid, a treatment that converts released DHA to pyruvic acid. The samples were then clarified using Carrez I solution (50 mL, 0.085 M potassium hexacyanoferrate) and Carrez II solution (50 mL, 0.25 M zinc sulphate) with vigorous shaking, and then neutralised with 100 mL 0.1 M sodium hydroxide. The samples were then diluted to 1.0 mL and centrifuged at 9000 g (2 min, 5 C) through 0.2 mm filters (NANOSEP MF, VWR), to remove any precipitate. An aliquot (100 mL) was then pipetted into a 96-well plate (UV-STAR V R , Greiner Bio-One, Tokyo, Japan) in triplicate, and 10 mL 1.0 M NaOH, 90 mL Tris/HCl buffer (pH 7.4, 0.50 M), and 16 mL 2.82 mM NADH (Megazyme, Bray, Ireland) were added. Pyruvic acid was then quantified using the stoichiometric conversion of pyruvic acid to D-lactic acid produced on addition of 2.0 mL D-lactate dehydrogenase (2000 U mL À1 ) in the presence of NADH, with the absorbance at 340 nm determined 3 min after addition of the enzyme. Concentrations were determined using a standard curve constructed using authentic standards (0-12 mM).

Mass spectrometric quantification of lanthionine (LAN) and lysinoalanine (LAL)
LAN and LAL were quantified in protein hydrolysates, as described previously [20] with slight modifications. IF (10 mg) was placed in a glass vial with 400 mL of 6 M hydrochloric acid. The samples were evacuated and flushed with N 2 gas three times, then evacuated, sealed and subjected to acid hydrolysis (110 C, 20 h). The samples were then filtered (0.2 mm Mini-UniPrep Syringeless filter devices; Whatman, GE Healthcare, Chicago, IL), dried under N 2 gas, then resuspended in a solution containing 1 mg mL À1 of internal standard (deuterated lysine: 4,4,5,5-d 4 -L-Lys hydrochloride, Sigma-Aldrich). Analysis employed LC-MS/MS with electrospray ionisation. Samples were separated using a reverse-phase column (Phenomenex Aeris 1.7 mm Peptide XB-C18, pore size 150 Å, particle size 2.1 mm) coupled to a ThermoScientific Q-Exactive Orbitrap spectrometer. LAL and LAN were eluted using a binary gradient of 27 min using aqueous and organic mobile phases containing perfluoropentanoic acid (PFPA; solvent A: 5 mM PFPA in water; solvent B: 5 mM PFPA in 100% acetonitrile), with the gradient structured as follows: 100% A (0-5 min) for elution of LAN, 0-35% B (5-10 min), 35-50% B (10-15 min) for elution of LAL, 50-100% (15-17 min), 100% B (17-22 min), 100% B to 100% A (22-24 min) and 100% A (24-27 min). MS data acquisition and analysis were performed using Thermo Xcalibur software (Waltham, MA). Parent ion masses were 151.14, 234.14 and 209.06 for d 4 -Lys, LAL and LAN, respectively. Quantification was based on parallel reaction monitoring, with the most abundant daughter ions for each analyte selected for quantification (88.11 for d 4 -Lys, 84.08 for LAL, 120.01 for LAN). Normalised collision energy (NCE) was set to 34 to achieve efficient MS2 fragmentation. Peak areas for LAL and LAN were normalised to the internal standard from the same run, with absolute quantification determined using peak area ratio standard curves (1-1000 ng mL À1 ). The data are expressed as nmol mg À1 protein.

Enantioseparation and quantification of D-amino acids by UPLC
Enantioseparation and detection of D-amino acids was performed on hydrolysed protein samples using a modification of a previous method [21]. Protein samples were hydrolysed using MSA, and subjected to UPLC analysis with fluorescence detection (see above), after precolumn derivatisation for 2 min of the samples (10 mL) using freshly prepared OPA (20 mL, 200 mM in methanol) containing iso-butyryl-L-cysteine (IBLC; 200 mM in 0.1 M borate buffer) at a molar ratio of 1:3, pH 9.0 (OPA-IBLC). Samples were then injection on to a BEH-C18 column (2.1 Â 150 mm, particle size 1.7 mm) and eluted at a flow rate of 0.35 mL min À1 at 30 C, using a Shimadzu Nexera system (see above). The mobile phases were solvent A (20 mM sodium acetate in water, pH 6.2) and solvent B (7% acetonitrile in methanol); the gradient profile is given in Supplementary Table S4. Identification and quantification were determined by comparison to authentic standards, with concentrations calculated by a linear regression analysis of the peak areas.

Statistical analyses
Data are presented as mean ± SD of three replicates from independent experiments, unless otherwise stated. Statistical analysis was performed using GraphPad Prism 7.03 (GraphPad Software, Inc, La Jolla, CA) and SPSS statistical software (IBM SPSS 24.0, Chicago, IL). Statistical significance was assumed at p < 0.05, using one-way ANOVA (analysis of variance), followed by post-hoc Tukey's multiple comparisons test or Dunnett's multiple comparisons test.

IF contain high molecular mass protein aggregates
SDS-PAGE analysis of IF samples showed the presence of protein bands assigned to bovine milk proteins including a-lactalbumin ($ 14.1 kDa), b-lactoglobulin ($ 18.4 kDa), caseins ($19-25 kDa) and serum albumin ($66 kDa) and other minor components (lactoferrin, $80 kDa, immunoglobulins) ( Figure 1). These bands were readily detected under reducing conditions (Figure 1(A)), whereas gels run under non-reducing conditions showing considerable smearing in the upper parts of the gels (> $70 kDa) (Figure 1(B)) consistent with the presence of significant concentrations of reducible aggregates in IF samples A-C. These cross-links are likely to be due to disulphide bonds that can be reduced by DTT. The extent of aggregate formation differed between the samples with IF samples A and B consistently showing higher quantities of high mass aggregates than sample C, which contained a higher level of b-lactoglobulin as assessed by the pixel density of the band at $ 18.4 kDa. The residual smearing at the top of the gels run under reducing conditions is consistent with the additional presence of non-reducible cross-links (see also below). Sample D did not show significant protein bands on the gels, consistent with this being a hydrolysed protein preparation containing only low molecular mass peptides and/or free amino acids.

Quantification of protein carbonyls
Significant levels of DNPH-reactive carbonyl compounds were detected on proteins and peptides precipitated from IF solutions using TCA (Figure 2(A) and Table 1). For samples A and B, $3 nmol carbonyls per mg protein were detected, with lower levels ($2.2 nmol mg À1 protein) detected in sample C. Significant levels were also detected in hydrolysed protein sample D ($0.85 nmol mg À1 protein), presumably reflecting carbonyls present on peptides that were susceptible to TCA precipitation, but too small to be retained on the SDS-PAGE gels (i.e. < 5 kDa).
The carbonyl distribution between different proteins was examined on SDS-PAGE gels using an antibody directed against the 2,4-dinitrophenylhydrazone moiety present on derivatised proteins. This procedure showed that (a) multiple proteins contained carbonyl groups in each preparation, (b) that similar classes of proteins were modified in each preparation and that (c) the extent of modification varied significantly across the different IFs (Figure 2(B)). Positive staining was detected for multiple protein bands, with the majority being present on proteins of higher mass, and particularly those at $ 25-35 kDa, 60-80 kDa and $100 kDa. These data are consistent with the presence of carbonyls on caseins, serum albumin, immunoglobulins, and/or protein aggregates (e.g. of a-lactalbumin or b-lactoglobulin). Only modest levels were detected on monomeric a-lactalbumin or b-lactoglobulin. A consistently higher level of staining was detected for sample C, compared to A and B, in contrast to the overall levels detected by direct analysis (Figure 2(A)); the reasons for this difference are unknown, but may arise from steric interactions of the (large) antibody with hydrazone antigen on the proteins, with this resulting in differential readings compared to the direct quantification of hydazones.

Amino acid analysis of proteins and peptides
Analysis of the free thiol (Cys residues) concentrations on the proteins/peptides present in the IF samples ( Figure 3) showed markedly higher levels for sample C ($135 nmol mg À1 protein) than the other IF samples (57-71 nmol mg À1 protein). Sample B had a slightly higher concentration than samples A and D. Analysis of the other amino acids present in the proteins/peptides, using amino acid analysis, showed that the profiles and the concentrations of amino acids of the four IFs were similar (Supplementary Figure 1 and Supplementary  Table S5). These data reflect only the material that is precipitated by TCA, and, therefore, only a fraction of the total amino acids present in the parent IF samples; this is particularly true for the sample D, where the majority of the amino acids are present either as low molecular mass peptides and/or as free species. The respective concentrations of precipitated versus  TCA-soluble materials are given in Supplementary Figure S2, with a much higher proportion of the total protein-derived material amino acids present in the TCA precipitates for samples A-C (69.5-76.4%) than D (35.7%).

Detection and quantification of amino acid sidechain oxidation products
The amino acid analysis method employed also allowed quantification of methionine sulphoxide (MetSO), a major oxidation product of methionine (Met) residues. For each IF sample, significant concentrations of MetSO were detected with these accounting for 33-46% of the total Met (Met þ MetSO) detected in the samples ( Figure 4 and Table 1). Other oxidation products were quantified using an alternative HPLC approach, with this yielding data (Figure 4(A) and Table 1) for the Phe oxidation product, m-Tyr, the Trp oxidation products Nformylkynurenine (NFK) and kynurenine (Kyn; quantified as a total, as NFK is hydrolysed to Kyn during sample preparation) and 3-hydroxykynurenine (3OHKyn), and the Tyr oxidation product, di-tyrosine (di-Tyr). Significant levels of these materials were detected in all samples with the highest levels detected, in each case, for sample D. No significant differences were detected between the other three IF samples in the levels of each of these products, with the exception of 3OHKyn, where sample C contained almost twice the concentration of samples A and B. In each case, a significant level of conversion of the parent amino acids to oxidation products was detected, with the highest levels detected for sample D in each case despite the much lower level of TCA-precipitated material (Phe $ 4%, Trp $ 48% and Tyr $ 23%; Supplementary Figure S3). Little loss of Phe ($0.6%) and low levels of loss of Tyr ($4%) and Trp ($12%) were detected for the other samples. These values are likely to be underestimates, as it is highly unlikely that all the products have been detected (cf. data for Trp and Tyr in [18]). These data suggest that the additional processing employed to manufacture the hydrolysed protein IF sample D induces a higher level of amino acid modification.  The proteins on which these products were present were examined further by Western blotting using an anti-N-formylkynurenine (anti-NFK) antibody. Representative blots for two different protein concentrations (Figure 4(B)) indicate antibody recognition of epitopes present on proteins with masses of $ 18-50 kDa, consistent with the presence of NFK on b-lactoglobulin and caseins, and possibly aggregates of these proteins. Sample B gave significantly greater staining than A and C, consistent with a higher concentration of NFK in this sample. No antibody-positive material was detected for sample D, consistent with the data in Figure 1.
The presence of total levels of di-Tyr in the IF samples was examined further by dot-blotting using an anti-di-Tyr antibody on nitrocellulose membranes. This approach only examines high mass material that remains on the membrane after vacuum-induced deposition, as low mass material is pulled through the membrane. Significant immunoreactivity (Figure 4(C)), was detected with all IF samples at the same protein concentration, with sample B showing the strongest immunoreaction, with (semi-) quantitative analysis (normalised to a positive control of photo-oxidised glucose-6-phosphate dehydrogenase [18]) given in Figure 4(D). These dot-blots confirm the presence and distribution of di-Tyr in proteins of IFs, but there are clear differences to the UPLC data, presumably arising from the different peptide and protein populations sampled by each method.

Quantification of dehydroalanine (DHA)
DHA, a non-genetically coded unsaturated amino acid formed by b-elimination reactions of Cys or Ser residues, readily reacts with nucleophiles via Michael Quantification of the spot intensities (obtained using ImageJ software, NIH Image J system, Bethesda, MD) from panel C, normalised to the G6PDH data at the same level of protein concentration with the control group set to 100%. Data are presented as means ± SD, from n ¼ 3 independent analyses. Statistical differences were detected via one-way ANOVA and Dunnett's multiple comparisons test. Ã p < 0.05 versus the control group.
reactions to give adducts. Thus, reaction of DHA with the thiol group of Cys gives the adduct species LAN, and reaction with the terminal amine group of Lys gives lysinoalanine (LAL) [13]. DHA is, therefore, a key intermediate in cross-link formation and has been detected previously in multiple proteins including human serum albumin [22]. DHA was quantified by determination of the extent of conversion of pyruvate (to which DHA is converted under acidic conditions) to lactate using an enzymatic assay [19]. The data obtained ( Figure 5(A) and Table 1) indicate that the DHA levels in the four IF samples were similar (0.43-0.58 nmol mg À1 protein), with sample C having the lowest levels.

Mass spectrometric analysis of lanthionine (LAN) and lysinoalanine (LAL)
The adduct species LAN and LAL were quantified in the IF samples by LC-MS. LAN concentrations in samples A-C ranged from 9.6 Â 10 À3 to 1.4 Â 10 À2 nmol mg À1 protein, and LAL concentrations 2.6-2.9 Â 10 À2 nmol mg À1 protein, with no significant differences detected between these samples ( Figure 5(B) and Table 1). Sample D contained much higher levels of both LAN (8.6 Â 10 À2 nmol mg À1 protein) and LAL (2.9 Â 10 À1 nmol mg À1 protein).

Detection and quantification of D-amino acids
Oxidation reactions, and high temperatures or pH, can induce racemisation (L-! D-conversion) of amino acids on proteins [8,23], with this potentially inducing major changes in protein structure, activity and properties, and increased toxicity [8]. UPLC with a chiral derivatisation method was therefore employed to examine the occurrence of such interconversion. Using this approach (Supplementary Figures S6 and S7), most amino acid enantiomeric pairs could be detected with the exception of Gly (achiral), L-/D-Asp (which eluted close to the void volume, and hence could not be accurately quantified), Pro (a secondary amine that does not react with OPA/IBLC reagent), and Cys and cystine (which are destroyed by the acid hydrolysis employed). These analyses showed that there were higher levels of some racemised (D-) amino acids in sample D than the other samples. It should be noted that these levels were relatively modest in terms of percentage values (compared to the total amino acid levels) in most cases, though due to the high levels of protein in the IF samples ($10% by mass, Supplementary Table S1), these levels might be of nutritional or toxicological significance. The extent of racemisation is dependent on the amino acid structure, and decreased in the order Arg $ Met > Val > His > Phe $ Lys > Tyr $ Ala $ Leu ( Table 1). The remaining amino acids had levels of the D-isomer below 1% compared to the L-isomer. As L-Thr and L-His coeluted under the conditions employed, while D-Thr and D-His could be separated, the levels of racemisation of these two amino acids was calculated assuming an equal abundance of parent L-Thr and L-His.

Discussion
This study aimed to quantify the levels of various protein oxidation products, and other modifications such as cross-links, aggregates and racemised amino acids, would might be present on the proteins and peptides present in commercial powdered IFs as a result of the processing and handling of these materials. Exact details of these conditions are unfortunately not available, as this is commercially sensitive information. This quantitative data is a prerequisite for the design of studies on the potential effects of these modified materials on the gastro-intestinal tract and physiology of infants exposed to IFs. The proteins present in IF may also be modified by glycation reactions [1,[24][25][26][27], or via the occurrence of lipid oxidation reactions and reaction with lipid oxidation products (e.g. unsaturated aldehydes and dialdehydes [7,28,29]), but these reactions and products have not been assessed in this study.
The data obtained here together with those from previous studies (e.g. [1,14,15,30]) indicate that significant levels of modifications are introduced into proteins during the manufacture of IFs ( Table 1). Each of the IF preparations (three conventional, samples A-C; and one hypoallergenic type, sample D), representing some of the major brands sold in Denmark, contained significant levels of modified proteins and amino acids.
Samples A-C all contained significant levels of reducible protein aggregates (high molecular mass species detected on non-reducing, but not reducing SDS-PAGE gels), as well as irreversible protein aggregates (smears at high molecular mass on gels run under reducing conditions). The proteins that contribute to these aggregates cannot be determined from the current data, as analysis of such highly heterogeneous material is challenging. The presence of significant levels of cross-links in these materials is confirmed by the detection of di-Tyr (by both UPLC and a specific anti-di-Tyr antibody), LAN and LAL, with the latter two detected by MS analysis. All of these species can give intra-or intermolecular cross-links. Furthermore, significant levels of DHA a precursor to both LAN and LAL [31] were detected in each sample. Interestingly, DHA, LAN and LAL were also detected with the hydrolysed IF sample D, with the levels of di-Tyr, LAN and LAL in sample D all higher than in samples A-C, even though little high molecular mass materials was detected on the protein gels. Overall, sample D appears to have (in general) higher levels of most of the examined modifications than the non-hydrolysed IF samples A-C. This suggests that these materials are formed in sample D to a higher level during processing, and are highly resistant to both acid and enzymatic hydrolysis, even though the remaining protein is converted to peptides or free amino acids. Whether each of these cross-links remain intact in the gastro-intestinal tract of humans after consumption remains to be established, but the detection of these species after treatment with strong acid during the analyses carried out here, suggests that these will also remain intact in vivo. The absence of high levels of DHA in sample D, when compared to samples A-C, is consistent with both increased formation and rapid subsequent removal of DHA by further reaction (i.e. significant cross-link formation). DHA is known to arise via multiple pathways involving b-elimination reactions of Ser and Cys residues, with the latter being more susceptible to degradation [22]. These reactions can occur via both non-radical and radical pathways, and therefore the formation of DHA (and hence LAN and LAL), cannot provide unequivocable evidence of radical-mediated oxidation reactions in these samples.
In contrast to the situation with the LAN and LAL cross-links, di-Tyr is known to arise via radical-radical termination reactions of Tyr-derived phenoxyl radicals, with the formation of new carbon-carbon or (to a lesser extent) carbon-oxygen bonds [32]. The detection of this species, by two independent methods, therefore, provides strong evidence for radical-mediated oxidation during sample processing during manufacturing, though at exactly which stage remains to be established. The differences in di-Tyr yields between the two techniques are likely to be due to the different protein populations sampled, with the UPLC data likely to be more quantitatively accurate.
The occurrence of radical-mediated oxidation is supported by the detection of protein carbonyls in all the IF samples, as these are known products of radical reactions involving carbon-centred-, alkoxyl-and peroxylradicals, and hydroperoxides [2]. The exact identity of the carbonyl species has not been established, but the data indicate that these are present on multiple proteins, and particularly those of higher molecular mass. The latter observation may reflect the presence of protein aggregates, and/or a greater extent of oxidation of proteins of high molecular mass as a result of the higher abundance of target species.
Trp, Tyr and Met residues are highly susceptible to oxidation, due to their low radical reduction potentials, and their capacity to act as "sinks" for (long range) electron-transfer processes [2]. It has been established that initial oxidation at multiple sites can give rise to ultimate damage at both Trp and Tyr residues, even if these are present in low abundance, and buried within protein structures [33]. Oxidation of Met to the corresponding sulphoxide, is a well-established process, and the detection of both Met sulphoxide, and significant levels of NFK, Kyn and 3OHKyn, are consistent with marked oxidation of both Met and Trp residues respectively in the IF samples. Whether these products arise via radical reactions, or via the action of two-electron oxidants (e.g. 1 O 2 or other photoexcited states) is unclear, as each process can generate these materials [2,34]. The detection of m-Tyr an oxidation product of Phe residues generated by oxygen-centred radicals, including the hydroxyl radical (HO . ) is consistent with radical-mediated events [2]. The Trp-derived materials are also known products of enzymatic reactions (e.g. the indoleamine-2,3-dioxygenase pathway), but there is no evidence for these enzymes in milk samples, so this pathway is unlikely to be significant.
These data, as well as the indirect evidence for the involvement of Cys residues (the presumed source of the reversible aggregates on the SDS-PAGE gels), are, therefore, consistent with the occurrence of radicalmediated reactions during the processing and storage of IF, in line with previous suggestions [14]. Modification of Trp and Met are of potential nutritional importance, as they are essential amino acids for growth and development, and might contribute to the immunogenicity of oxidised proteins arising from thermal processing [35].
The amino acid racemisation detected in the IF samples is of considerable potential importance, as six of the nine amino acids detected as D-isomers (His, Met, Val, Leu, Phe, Lys) in moderate to high levels are essential amino acids. The mechanism by which racemisation occurs (oxidative versus base-catalysed molecular reactions; Figure 6) cannot be determined from the current data, and we cannot exclude some formation during the analysis of these materials [8]. However, the observation that racemisation is selective to specific species, and occurs to a highly variable extent across all the amino acids, suggests that these changes are not solely an artefact.
Overall these data, together with previous studies [1,14,15], indicate that significant levels of modifications are introduced into proteins during the manufacture of IFs, with particularly high levels of some (but not all) species detected on the hydrolysed protein sample D. The levels of each of the modified materials detected are collected in Table 1, and expressed in nmol/mg protein for the oxidation products, and as a % D-isomers of the total amino acids for the racemisation data. The levels of most of these species are relatively modest, with the most abundant oxidative modification being methionine sulphoxide (formed from oxidation of Met residues) followed by protein carbonyls (a generic marker of damage to many aliphatic side-chains), dityrosine (from Tyr residues) and the Trp oxidation products kynurenine and 3-hydroxykynurenine. The concentration of protein carbonyls detected is of a similar magnitude to that reported previously ($6.5 nmol/mg protein) for some IFs [15] but lower than others ($19.4 nmol/mg protein [30]). The differences in these values may reflect the manufacturer and manufacturing process, as well as the length and conditions of storage.
The levels of the cross-link precursor DHA, and its products LAN and LAL were low compared to the levels of di-tyrosine, suggesting that di-tyrosine is the major species that gives rise to the higher mass materials detected on the reduced SDS-PAGE gels. The low levels of LAN and LAL detected in the intact protein samples (0.01-0.03 nmol/mg protein; Table 1) are consistent with earlier studies that have reported low levels in powdered IF (e.g. 0.078 nmol/mg protein [30]). Higher and more variable levels have been reported in liquid infant milk samples [36,37]. Values for di-tyrosine reported previously [30] are up to 78 pmol/mg protein. However, the marked differences in extent of cross-linking of the proteins detected between the non-reduced and reduced gels, with much higher levels of oligomerisation detected under the former conditions, suggest that disulphide cross-links (or related reducible species) may be quantitatively the most abundant source of cross-links in the IFs.
The extent of racemisation of the amino acids cannot be easily compared to these oxidative modifications as the original abundance of the proteins and their exact amino acid composition are not disclosed by the manufacturers on the product labels. However, the % conversion of some of the amino acids to the D-isomers is high (Table 1), suggesting that these species are quantitatively significant contributors to the total pool of modified species present in these IF samples.
Although the levels of some of these modifications are modest, it has been established that some of these, including both oxidation (e.g. di-Tyr, DOPA, o-Tyr and m-Tyr) and racemisation products (D-amino acids: see detailed discussion in [8]), can give rise to toxicity and other adverse effects on cells and tissues (e.g. kidney) both in vitro and in vivo [3,8,38] (Z. Chen et al, unpublished data). The observation of the highest levels of modification in the hypoallergenic (hydrolysed protein) IF samples is consistent with previous data [30], and of concern, as hydrolysed protein has been widely proposed and studied as a potential solution to reduce the immunogenicity of intact bovine proteins (e.g. bovine serum albumin [39] and insulin [40]) in infants fed IF. Data from a recent randomised clinical trial of IFs consisting of hydrolysed proteins against non-hydrolysed proteins has found no benefit with regard to a reduction in the risk of Type 1 diabetes [41], and hydrolysed proteins may even increase this risk [42]. The health gains from reduction of exposure to intact bovine proteins may therefore be counteracted by the increased levels of modifications and loss of parent amino acids in hydrolysed IFs. Variation in the levels of amino acid modification in IFs with and without hydrolysed proteins may account for the divergence of data with regard to the use of IFs containing hydrolysed proteins [41,42]. Furthermore, it is not clear whether the modification products detected in IFs all have similar effects with regard to impacts on health with both positive and negative effects reported for glycation products present in IFs (see, e.g. data on glycation products acting as antioxidants [43], and also proinflammatory agents [44]). Further studies would, therefore, seem to be warranted with both individual species, and the mixtures of these present in IFs.
Trp is required for the synthesis of serotonin and some endocrine hormones, with these molecules, together with kynurenines, playing a key role in regulating behaviour and metabolism [45]. Some studies have reported that IF feeding can lower serum Trp levels [46] while others have found no effect [47]. A recent study on infant rhesus monkeys has reported IF-induced differences in immunology and metabolism [48], whilst a study on 2-year infants has reported improved brain development in breast-versus IF-fed infants [49]. These data are supported by epidemiological studies [50]. The consequences of infant nutrition on signalling molecules and ultimately physiological status are still controversial, as are the mechanisms involved. Of potential relevance to the current study are reports of an interplay between Trp/Trp oxidation products, gut microbiota and the gut/brain axis via serotonin or endocrines [51]. A study on porcine colon has reported an IF-induced shift in Trp metabolism from serotonin towards tryptamine, resulting in lowered serum serotonin levels [52]; changes in the gut microbiota have also been reported in infants [53]. Thus, whilst Trp is often added as a supplement to IFs to compensate for potential losses during processing, the oxidation products formed from Trp degradation may contribute to the observed effects of IF-feeding.