Advertisement

Reduction of multiple reaction monitoring protein target list using correlation analysis

Open AccessPublished:July 22, 2022DOI:https://doi.org/10.3168/jds.2021-21647

      ABSTRACT

      High mass resolution mass spectrometry provides hundreds to thousands of protein identifications per sample, and quantification is typically performed using label-free quantification. However, the gold standard of quantitative proteomics is multiple reaction monitoring (MRM) using triple quadrupole mass spectrometers and stable isotope reference peptides. This raises the question how to reduce a large data set to a small one without losing essential information. Here we present the reduction of such a data set using correlation analysis of bovine dairy ingredients and derived products. We were able to explain the variance in the proteomics data set using only 9 proteins across all major dairy protein classes: caseins, whey, and milk fat globule membrane proteins. We term this method Trinity-MRM. The reproducibility of the protein extraction and Trinity-MRM methods was shown to be below 5% in independent experiments (multi-day single-user and single-day multi-user) using double cream. Further application of this reductionist approach might include screening of large sample cohorts for biologically interesting samples before analysis by high-resolution mass spectrometry or other omics methodologies.

      Key words

      INTRODUCTION

      Milk is an opaque oil-in-water emulsion comprising various particles ranging in size from very small whey protein particles over casein micelles to very large milk fat globule membrane (MFGM) particles (
      • Christian P.
      • Smith E.R.
      • Lee S.E.
      • Vargas A.J.
      • Bremer A.A.
      • Raiten D.J.
      The need to study human milk as a biological system.
      ). Milk fat globule membrane particles comprise a triglyceride core surrounded by a phospholipid monolayer that is enclosed by a small cytoplasmic space and packaged in a lipid bilayer (
      • Arranz E.
      • Corredig M.
      Invited review: Milk phospholipid vesicles, their colloidal properties, and potential as delivery vehicles for bioactive molecules.
      ;
      • Manoni M.
      • Di Lorenzo C.
      • Ottoboni M.
      • Tretola M.
      • Pinotti L.
      Comparative proteomics of milk fat globule membrane (MFGM) proteome across species and lactation stages and the potentials of MFGM fractions in infant formula preparation.
      ). This triple-layer assembly contains not only bioactive lipids but also numerous proteins, some of which are anchored within the lipid bilayer (
      • Zheng H.
      • Jiménez-Flores R.
      • Everett D.W.
      Bovine milk fat globule membrane proteins are affected by centrifugal washing processes.
      ).
      In multiple pre-clinical and clinical studies, infant formula (IF) products enriched with MFGM showed beneficial effects. In pre-clinical studies, MFGM-enriched IF resulted in an increase in infant neurodevelopmental scores compared with IF alone, closely matching scores of breast-fed infants (
      • Brink L.R.
      • Lonnerdal B.
      Milk fat globule membrane: The role of its various components in infant health and development.
      ). Gong and coworkers investigated the effects of MFGM supplementation on rat pups and impressively demonstrated that supplementation with MFGM allows for near normal (breast-fed) intestinal development (
      • Gong H.
      • Yuan Q.
      • Pang J.
      • Li T.
      • Li J.
      • Zhan B.
      • Chang R.
      • Mao X.
      Dietary milk fat globule membrane restores decreased intestinal mucosal barrier development and alterations of intestinal flora in infant-formula-fed rat pups.
      ). In clinical studies involving 582 human infant subjects, the safety of MFGM-enriched IF was successfully established (
      • Jaramillo-Ospina A.M.
      • Toro-Campos R.
      • Murguía-Peniche T.
      • Wampler J.L.
      • Wu S.S.
      • Berseth C.L.
      • Uauy R.
      Added bovine milk fat globule membrane in formula: Growth, body composition, and safety through age 2: An RCT.
      ). Timby and coworkers demonstrated a higher cognitive score at 12 mo of age in infants fed with MFGM-enriched IF compared with standard IF (
      • Timby N.
      • Domellöf E.
      • Hernell O.
      • Lönnerdal B.
      • Domellöf M.
      Neurodevelopment, nutrition, and growth until 12 mo of age in infants fed a low-energy, low-protein formula supplemented with bovine milk fat globule membranes: A randomized controlled trial.
      ). Another example was published by Schneider and coworkers, quantifying significant differences in myelin structure in infants after 6 mo of treatment, using magnetic resonance imaging and analyses (
      • Schneider N.
      • Bruchhage M.M.K.
      • O'Neill B.V.
      • Hartweg M.
      • Tanguy J.
      • Steiner P.
      • Mutungi G.
      • O'Regan J.
      • McSweeney S.
      • D'Sa V.
      • Deoni S.C.L.
      A nutrient formulation affects developmental myelination in term infants: A randomized clinical trial.
      ). Based on these studies and others, interest in quantifying MFGM particles has been increasing (
      • Singh H.
      The milk fat globule membrane—A biophysical system for food applications.
      ).
      Analytically, casein micelles and MFGM particles overlap in size, ranging from 50 to 500 nm and 100 to 5,000 nm, respectively (
      • Astaire J.C.
      • Ward R.
      • German J.B.
      • Jiménez-Flores R.
      Concentration of polar MFGM lipids from buttermilk by microfiltration and supercritical fluid extraction.
      ;
      • Cheong L.Z.
      • Jiang C.
      • He X.
      • Song S.
      • Lai O.M.
      Lipid profiling, particle size determination, and in vitro simulated gastrointestinal lipolysis of mature human milk and infant formula.
      ). Therefore, simple filtration is insufficient to enrich for MFGM particles, and multi-step separation and enrichment processes might result in fragmentation of MFGM particles with partial recovery of all MFGM proteins (
      • Costa M.R.
      • Elias-Argote X.E.
      • Jiménez-Flores R.
      • Gigante M.L.
      Use of ultrafiltration and supercritical fluid extraction to obtain a whey buttermilk powder enriched in milk fat globule membrane phospholipids.
      ;
      • de Toledo Guimarães J.
      • Silva E.K.
      • de Freitas M.Q.
      • de Almeida Meireles M.A.
      • da Cruz A.G.
      Non-thermal emerging technologies and their effects on the functional properties of dairy products.
      ;
      • Huang Z.
      • Zheng H.
      • Brennan C.S.
      • Mohan M.S.
      • Stipkovits L.
      • Li L.
      • Kulasiri D.
      Production of milk phospholipid-enriched dairy ingredients.
      ). Hence, an analytical method quantifying whey, casein, and multiple MFGM proteins is required to estimate the degree of MFGM enrichment following the enrichment process. As the abundance of these 3 protein ranges multiple orders of magnitude, a bottom-up proteomics approach is preferred over top-down (
      • Fuerer C.
      • Jenni R.
      • Cardinaux L.
      • Andetsion F.
      • Wagniere S.
      • Moulin J.
      • Affolter M.
      Protein fingerprinting and quantification of β-casein variants by ultra-performance liquid chromatography-high-resolution mass spectrometry.
      ). Another analytical challenge is the compatibility of these lipid-rich fractions with reverse-phase chromatography and proper ionization of peptide analytes during electrospray ionization before entering the mass spectrometer (
      • Ebhardt H.A.
      • Root A.
      • Liu Y.
      • Gauthier N.P.
      • Sander C.
      • Aebersold R.
      Systems pharmacology using mass spectrometry identifies critical response nodes in prostate cancer.
      ).
      Numerous research groups have investigated various aspects of the MFGM proteome (
      • Li M.
      • Zheng K.
      • Song W.
      • Yu H.
      • Zhang X.
      • Yue X.
      • Li Q.
      Quantitative analysis of differentially expressed milk fat globule membrane proteins between donkey and bovine colostrum based on high-performance liquid chromatography with tandem mass spectrometry proteomics.
      ). Initial findings identified about a dozen proteins associated with MFGM (e.g., see Table 2 in
      • Fontecha J.
      • Brink L.
      • Wu S.
      • Pouliot Y.
      • Visioli F.
      • Jiménez-Flores R.
      Sources, production, and clinical treatments of milk fat globule membrane for infant nutrition and well-being.
      ), but advancements in analytical techniques, especially mass spectrometry, have revealed hundreds of proteins associated with MFGM particles (
      • Affolter M.
      • Grass L.
      • Vanrobaeys F.
      • Casado B.
      • Kussmann M.
      Qualitative and quantitative profiling of the bovine milk fat globule membrane proteome.
      ). The most recent data-mining approach across 35 proteomic data sets, by Delosière and colleagues, identified 246 proteins associated with MFGM in at least one lactation stage, spanning from colostrum to the beginning of the dry period (
      • Delosière M.
      • Pires J.A.A.
      • Bernard L.
      • Cassar-Malek I.
      • Bonnet M.
      Dataset reporting 4654 cow milk proteins listed according to lactation stages and milk fractions.
      ).
      In addition to identification, quantification of proteins is important for establishing the degree of enrichment during the MFGM enrichment process and quality control. High-resolution mass spectrometry often uses label-free quantification to estimate amounts of proteins per sample (
      • Zhao L.
      • Du M.
      • Gao J.
      • Zhan B.
      • Mao X.
      Label-free quantitative proteomic analysis of milk fat globule membrane proteins of yak and cow and identification of proteins associated with glucose and lipid metabolism.
      ). Including stable isotope-labeled reference peptides increases the accuracy of quantification, especially when using triple quadrupole (QQQ) mass spectrometers (
      • Kiel C.
      • Ebhardt H.A.
      • Burnier J.
      • Portugal C.
      • Sabido E.
      • Zimmermann T.
      • Aebersold R.
      • Serrano L.
      Quantification of ErbB network proteins in three cell types using complementary approaches identifies cell-general and cell-type-specific signaling proteins.
      ). Ideally, one reference peptide per protein is chosen and quantified using QQQ mass spectrometers with robust ion optics and HPLC, especially in standard liquid chromatography (LC) flowrates, for precise quantification with less than 20% coefficient of variation (CV) in cross-laboratory comparisons (
      • Addona T.A.
      • Abbatiello S.E.
      • Schilling B.
      • Skates S.J.
      • Mani D.R.
      • Bunk D.M.
      • Spiegelman C.H.
      • Zimmerman L.J.
      • Ham A.J.
      • Keshishian H.
      • Hall S.C.
      • Allen S.
      • Blackman R.K.
      • Borchers C.H.
      • Buck C.
      • Cardasis H.L.
      • Cusack M.P.
      • Dodder N.G.
      • Gibson B.W.
      • Held J.M.
      • Hiltke T.
      • Jackson A.
      • Johansen E.B.
      • Kinsinger C.R.
      • Li J.
      • Mesri M.
      • Neubert T.A.
      • Niles R.K.
      • Pulsipher T.C.
      • Ransohoff D.
      • Rodriguez H.
      • Rudnick P.A.
      • Smith D.
      • Tabb D.L.
      • Tegeler T.J.
      • Variyath A.M.
      • Vega-Montoto L.J.
      • Wahlander A.
      • Waldemarson S.
      • Wang M.
      • Whiteaker J.R.
      • Zhao L.
      • Anderson N.L.
      • Fisher S.J.
      • Liebler D.C.
      • Paulovich A.G.
      • Regnier F.E.
      • Tempst P.
      • Carr S.A.
      Multi-site assessment of the precision and reproducibility of multiple reaction monitoring-based measurements of proteins in plasma.
      ;
      • Bringans S.
      • Ito J.
      • Casey T.
      • Thomas S.
      • Peters K.
      • Crossett B.
      • Coleman O.
      • Ebhardt H.A.
      • Pennington S.R.
      • Lipscombe R.
      A robust multiplex immunoaffinity mass spectrometry assay (PromarkerD) for clinical prediction of diabetic kidney disease.
      ). Using QQQ mass spectrometers for quantification of peptides, and therefore proteins, requires method development of predefined transitions, also known as multiple reaction monitoring (MRM;
      • Ebhardt H.A.
      • Root A.
      • Sander C.
      • Aebersold R.
      Applications of targeted proteomics in systems biology and translational medicine.
      ). For complex backgrounds, 4 transitions are quantified at a specific retention time to rule out contamination from other possible product ions (
      • Röst H.
      • Malmström L.
      • Aebersold R.
      A computational tool to detect and avoid redundancy in selected reaction monitoring.
      ). When QQQ mass spectrometers are not available, hybrid quadrupole Orbitrap mass spectrometers might be used for targeted proteomics, in a methodology termed parallel reaction monitoring (PRM;
      • Ramachandran B.
      • Yang C.T.
      • Downs M.L.
      Parallel reaction monitoring mass spectrometry method for detection of both casein and whey milk allergens from a baked food matrix.
      ). The development of PRM and MRM methods reduces the number of proteins quantified (
      • Bär C.
      • Mathis D.
      • Neuhaus P.
      • Dürr D.
      • Bisig W.
      • Egger L.
      • Portmann R.
      Protein profile of dairy products: Simultaneous quantification of twenty bovine milk proteins.
      ), and the question arises which proteins to choose for comprehensive quantification without losing essential information.
      Here we present a reductionist approach, conserving essential information for developing a targeted proteomics MRM method using prior knowledge, hybrid quadrupole Orbitrap, and QQQ mass spectrometers. The process included a quantitative proteomics analysis of whey extracts, MFGM-enriched and skim milk products quantified by an Orbitrap mass spectrometer in PRM mode. Subsequent data analysis revealed which proteins are correlated in a positive or negative manner and the strength of the relationship across the tested data set (
      • Friendly M.
      Corrgrams: Exploratory displays for correlation matrices.
      ;
      • Chaulk S.G.
      • Ebhardt H.A.
      • Fahlman R.P.
      Correlations of microRNA: MicroRNA expression patterns reveal insights into microRNA clusters and global microRNA expression patterns.
      ). From the reduced number of proteins, an MRM method termed Trinity-MRM was developed, quantifying all 3 dairy protein groups of casein, whey, and MFGM proteins. The workflow comprised protein extraction from cream, trypsin digestion, and subsequent quantification of peptides using triple quadrupole mass spectrometer in conjunction with the Trinity-MRM method was shown to be robust in an inter-day single-user and intra-day multi-user reproducibility studies.

      MATERIALS AND METHODS

      Dairy Samples Quantified

      All dairy samples are of bovine origin (Taxonomy 9913 Bos taurus). Commercially available whey protein concentrate (WPC) products of different protein and fat contents were sourced from the EU, USA, and New Zealand, prepared in all cases from sweet whey: 5 batches of a high-fat WPC enriched at 30% α-lactalbumin (WPC_alac), 3 batches of whey protein enriched at 41% α-lactalbumin, and 2 regular WPC containing 70% and 35% protein, respectively. Because many new MFGM products were available on the market, 4 different products were sourced with up to 5 batches per product (MFGM_P1 through MFGM_P4). For comparative purposes, 3 batches of skim milk powder (SMP) were added.

      PRM Sample Preparation

      Unless otherwise stated, chemicals were purchased from Sigma-Aldrich, a division of Merck. Sample preparation was carried out at room temperature. Samples were dispersed in 18.2 MΩ·cm water (Millipore water purification system) to reach a protein content of 3.5% (wt/vol), mixed for 30 min, denatured, and reduced by addition of 4 volumes of denaturing buffer (guanidine HCl 7.5 M, trisodium citrate 6.25 mM, 1,4-dithiothreitol 23 mM), incubated on a rotating wheel (Rotator SB3, Stuart Equipment, Cole-Parmer) for another 30 min, and centrifuged at 16,000 × g (Centrifuge 5415R, Eppendorf) for 10 min to pellet debris (
      • Fuerer C.
      • Jenni R.
      • Cardinaux L.
      • Andetsion F.
      • Wagniere S.
      • Moulin J.
      • Affolter M.
      Protein fingerprinting and quantification of β-casein variants by ultra-performance liquid chromatography-high-resolution mass spectrometry.
      ). A total of 200 μL of the aqueous phases were mixed with 100 µL of iodoacetamide 100 mM (in 100 mM ammonium bicarbonate, ABC) and incubated for 30 min at room temperature in the dark. The alkylated samples were diluted 10 times in ABC 100 mM (20 µL of alkylated sample + 180 µL of ABC) and digested overnight with 6 µL of trypsin (sequencing-grade modified Trypsin V5111, Promega), 0.2 µg/µL. Six microliters of formic acid 10% was added to stop digestion. Digested samples were centrifuged at 14,000 × g for 10 min, and the aqueous phases were transferred into vials for injection.

      PRM LC-MS/MS

      Parallel reaction monitoring (PRM) was carried out using a Thermo Scientific Vanquish Flex ultra-high-performance liquid chromatograph coupled with a Thermo Scientific Q Exactive High Field hybrid quadrupole Orbitrap mass spectrometer. Peptides were separated on a C18 column (Waters Acquity UPLC BEH C18, 130 Å, 1.7 µm, 1 mm × 150 mm, 186002347) using a C18 pre-column (Waters Acquity UPLC BEH C18 VanGuard, 130 Å, 1.7 µm, 2.1 mm × 5 mm, 186003975). A predefined inclusion list (see Supplemental Table S1; https://panoramaweb.org/CorrelationAnalysis.url) was set up with LC tandem mass spectrometry (LC-MS/MS) parameters.

      PRM Data Analysis

      Mass spectrometry data files were imported into Skyline version 20.1 (
      • Pino L.K.
      • Searle B.C.
      • Bollinger J.G.
      • Nunn B.
      • MacLean B.
      • MacCoss M.J.
      The Skyline ecosystem: Informatics for quantitative mass spectrometry proteomics.
      ) using a predefined transition list. Quantitative values were obtained using Skyline's report function and total area value. The heuristic clustering of PRM data (Figure 1) was generated in RStudio version 1.2.1335 using R version 4.0.2 and library (pheatmap) version 1.0.12, which is an implementation of traditional heatmaps with more user control of dimensions and appearance. Figure 2 shows correlation analysis of data generated with library (corrplot) version 0.84 using order = FPC (first principal component order).
      Figure thumbnail gr1
      Figure 1Heuristic clustering of parallel reaction monitoring data. A total of 32 samples were analyzed and fall into 5 groups, 3 of which have 5 or more members and are labeled as clades 1 through 3 on the left-hand side dendrogram. The 24 protein analytes listed across the bottom are divided into 3 groups: caseins on the left, MFGM and minor whey proteins in the middle, and major whey proteins on the right-hand side. For full list of abbreviations, see .
      Figure thumbnail gr2
      Figure 2Correlation analysis. Blue ellipses tilted to the right-hand side show positive correlation of proteins across the data set, whereas red ellipses tilted to the left-hand side are anti-correlated. For correlation scale from −1 (red) to 1 (blue), see color gradient below the correlation matrix. Variance across all sample types (average for multiple batches of the same product). Clearly visible are groups of proteins with similar correlation (e.g., BT1A1 and MFGM). For groups with similar correlation, one member is quantified to reduce redundancy. Examples of unique correlation are PLIN2 and LALBA.

      MRM Sample Preparation (Liquid)

      Unless otherwise stated, chemicals were purchased from Sigma-Aldrich, a division of Merck. The 18.2 MΩ·cm water was generated on site using a Veolia Purelab Flex 2 purification system. For Trinity-MRM version 1 development, double cream (DC) was purchased from the local grocery store in Ireland and contained 48% fat and 2.1% protein as per label declaration. The DC was kept in a 4°C fridge for storage, and aliquots were taken periodically to perform all subsequent steps at room temperature (20–23°C). The DC sample was weighed 0.2 g ± 10% directly into a microcentrifuge tube using a micro-balance (Mettler Toledo NewClassic MF, model no. MS204S) and values recorded for subsequent data analysis. Subsequently, 0.4 mL of chloroform and 0.4 mL of methanol (both purchased from VWR, a division of Avantor Sciences) were added to the microcentrifuge tube containing DC. Following vigorous vortexing for 30 s, the emulsion was separated by centrifugation at 3,000 × g at 15°C for 5 min. After centrifugation, the protein material was localized between the 2 liquid phases as a white disk (
      • Braakman R.B.H.
      • Bezstarosti K.
      • Sieuwerts A.M.
      • de Weerd V.
      • van Galen A.M.
      • Stingl C.
      • Luider T.M.
      • Timmermans M.A.M.
      • Smid M.
      • Martens J.W.M.
      • Foekens J.A.
      • Demmers J.A.A.
      • Umar A.
      Integrative analysis of genomics and proteomics data on clinical breast cancer tissue specimens extracted with acid guanidinium thiocyanate–phenol–chloroform.
      ). Both methanol and chloroform phases were carefully removed using a micropipette, leaving the white disk containing proteins in the microcentrifuge tube. This white disk is resuspended in 0.18 mL of urea 1 M/ABC 50 mM buffer and incubated at 60°C for denaturation for 30 min at 500 rpm using a thermal shaker (Eppendorf ThermoMixer C). For reduction, 5 µL of 1,4-dithiothreitol 0.2 M were added and incubated at 58°C for 15 min (
      • Suttapitugsakul S.
      • Xiao H.
      • Smeekens J.
      • Wu R.
      Evaluation and optimization of reduction and alkylation methods to maximize peptide identification with MS-based proteomics.
      ). Both incubations were carried out in a thermal shaker at 500 rpm. Alkylation was carried out by adding 5 µL of iodoacetamide 0.4 M at room temperature (approximately 20–24°C) in the dark for 15 min. Then 10 µL of trypsin 10 µg/µL (Trypsin recombinant, proteomics grade, Roche 03708969001) were added and incubated overnight at 37°C in a thermal shaker at 500 rpm. After overnight incubation, 20 µL of acetonitrile/formic acid (100 mL/2 mL) plus 5 µL of the spike-in peptide mix were added. After briefly vortexing, the samples were centrifuged at 12,000 × g at 4°C for 5 min. After centrifugation, 0.175 mL of the supernatant containing tryptic peptides was transferred into an MS vial, and 5 µL were injected per LC-MS/MS run.

      MRM QQQ LC-MS/MS Method

      For details on Trinity-MRM, see Supplemental Table S2 (https://panoramaweb.org/CorrelationAnalysis.url), which contains the acquisition method report of the Agilent QQQ 6495A mass spectrometer coupled to an Agilent 1290 series HPLC. The C18 reverse-phase columns were Agilent Zorbax analytical column SB-C18 2.1 × 5 mm 1.8 µm (P.N. 821725-902) with pre-column SB-C18 2.1 × 100 mm 1.8 µm (P.N. 858700-902). Stable isotope-labeled peptides were synthesized by Sigma-Aldrich and Thermo Scientific, containing C-terminal [13C6,15N4]Arg or [13C6,15N2]Lys to >95% purity.

      MRM Data Analysis

      Result files were analyzed using Skyline software (64-bit, version 20.2.0.343) to determine the area under the curve (total area). A report generated by Skyline was analyzed using Microsoft Excel (Microsoft 365 for Enterprise) and plotted using RStudio (version 1.2.1335), R (version 4.0.2), and pheatmap library (version 1.0.12). Coefficient of variation calculations of repeatability study were carried out as previously described (
      • Bringans S.
      • Ito J.
      • Casey T.
      • Thomas S.
      • Peters K.
      • Crossett B.
      • Coleman O.
      • Ebhardt H.A.
      • Pennington S.R.
      • Lipscombe R.
      A robust multiplex immunoaffinity mass spectrometry assay (PromarkerD) for clinical prediction of diabetic kidney disease.
      ).

      RESULTS AND DISCUSSION

      Product Range Protein Quantification Using PRM

      Extensive analysis of bovine dairy, enriched dairy fractions, and derived products analyzed by high-resolution mass spectrometry in data-dependent shotgun mode generated an extensive catalog of protein identifications in these samples. Based on prior experience, a PRM method targeting all 3 major protein classes (caseins, whey and MFGM proteins) was established (
      • Affolter M.
      • Grass L.
      • Vanrobaeys F.
      • Casado B.
      • Kussmann M.
      Qualitative and quantitative profiling of the bovine milk fat globule membrane proteome.
      ;
      • Lutter P.
      • Parisod V.
      • Weymuth H.
      Development and validation of a method for the quantification of milk proteins in food products based on liquid chromatography with mass spectrometric detection.
      ). The PRM method targets 24 proteins using 45 peptides and 45 precursors measuring 379 transitions (Supplemental Table S1). On average, 2 peptides per protein were measured with a single charge state and roughly 8 transitions per precursor monitored. The LC gradient is half an hour in length, with peptides eluting between 7 and 25 min. The PRM method was applied on a range of dairy products and quantifies 24 proteins using label-free quantification. Commercially available whey, MFGM, and skim milk products were sourced within the EU, USA, and New Zealand (see Materials and Methods, Supplemental Table S3; https://panoramaweb.org/CorrelationAnalysis.url). In total, 9 distinct products with up to 5 batches per product resulted in 32 samples for quantitative analysis using the previously described PRM method. Quantitative proteomics data for 24 proteins (see Table 1) were normalized for equal sample loading across the data set and proteins grouped using Euclidean-distance clustering for visualization as heatmap in Figure 1.
      Table 1Summary of protein acronyms, Uniprot identifiers (
      UniProt Consortium
      UniProt: The universal protein knowledgebase.
      ), and protein groups based on correlation analysis
      Asterisks indicate proteins quantified in the final Trinity-MRM (multiple reaction monitoring) assay. CorrAnalysis = correlation analysis.
      SymbolAcronymProtein nameUniprot numberProtein group (CorrAnalysis)
      IgHG2IgHG2.partialIgG2 heavy chain, partialA0A3Q1N3I9Group 1
      PIGRPlgRPolymeric immunoglobulin receptorP81265Group 1
      TRFLLactoferrinLactotransferrinP24627Group 1*
      ALBUBSAAlbuminP02769Group 1
      IgHG1IgHG1.partialIgG1 heavy chain, partialA0A3Q1M3L6Group 1
      FABPHFABPFatty acid-binding protein, heartP10790Group 2
      MUC15MUC-15Mucin-15Q8MI01Group 2
      MUC1MUC-1Mucin-1Q8WML4Group 2*
      IgHMIgHM.partialIgM heavy chain, partialG5E513Group 3
      XDHXDHXanthine dehydrogenase/oxidaseP80457Group 3*
      GLCM1Glycam1Glycosylation-dependent cell adhesion molecule 1P80195Group 4
      PERLLactoperoxidaseLactoperoxidaseP80025Group 4
      OSTPOPNOsteopontinP31096Group 4*
      IgHG3IgHG3.partialIgG3 heavy chainA0A3Q1LPG0Group 4
      BT1A1BTNButyrophilin subfamily 1 member A1P18892Group 5*
      MFGMLactadherinLactadherinQ95114Group 5
      IgHGAIgHA.partialIgA heavy chain, partialA0A3Q1LRW4Group 5
      LACBβ-LacBeta-lactoglobulinP02754Unique*
      PLIN2ADPHPerilipin-2, adipophilinQ9TUM6Unique*
      LALBAα-LacAlpha-lactalbuminP00711Unique*
      CASKκ-CNKappa-caseinP02668Group 6
      CASA1αS1-CNAlpha-S1-caseinP02662Group 6
      CASBβ-CNBeta-caseinP02666Group 6*
      CASA2αS2-CNAlpha-S2-caseinP02663Group 6
      1 Asterisks indicate proteins quantified in the final Trinity-MRM (multiple reaction monitoring) assay. CorrAnalysis = correlation analysis.
      The separation of samples resulted in 5 groups: the largest clade (clade 1) was the reference standard of WPC with 35% and 70% protein content, a group of products that also included the whey protein enriched at 41% α-lactalbumin, albeit with higher amounts of LALBA (α-lactalbumin or α-lac), consistent with the label declaration of the whey α-lac enriched product. The WPC_alac formed its own clade (clade 2), with higher amounts of minor whey proteins such as PIGR (plgR) and TRFL (Lf) compared with clade 1. Interestingly, both α-lac products were in 2 different clades, reflecting different manufacturing processes to obtain an α-lac enriched product from whey: one process used different membranes in cascades with α-lac enriched in the permeate fraction, whereas the other process used membrane filtration to remove components of low molecular weight before enriching for α-lac using pH, ionic strength, and temperature shifts (
      • Barone G.
      • O'Regan J.
      • O'Mahony J.A.
      Influence of composition and microstructure on bulk handling and rehydration properties of whey protein concentrate powder ingredients enriched in α-lactalbumin.
      ). Using targeted proteomics on α-lac products clearly differentiated these 2 processing streams based on proteins quantified.
      Most MFGM products formed clade 3, which was comparable to the casein amounts found in clades 1 and 2. Clade 3 showed an enrichment in MFGM-related proteins such as XDH or MUC1 compared with clade 1. Most quantified proteins were similar between clades 2 and 3, suggesting that most tested MFGM products are similar to α-lac enriched whey. Also, considering the relatively lower amount of casein proteins of the first 3 clades compared with SMP, this suggested a sweet whey process stream for processing of most MFGM products. Surprisingly, MFGM_P4 did show a distinct protein profile compared with all other MFGM products. The MFGM_P4 products had relatively high amounts of caseins and relatively low amounts of LACB (β-lac). Most proteins of MFGM_P4 were similar to an SMP, with the exception of PLIN2, BT1A1, and MFGM proteins, which were collectively more abundant in MFGM_P4 compared with SMP products. These 3 proteins are classified as MFGM proteins, but so are MUC1, FABPH, and XDH, none of which were found in relatively low amounts in MFGM_P4 compared with all other tested MFGM products. This unique protein profile suggested a unique processing stream for MFGM_P4 compared with the other 3 MFGM products.
      The overall protein signature of dairy products tested allowed for classification and grouping of products. The quantification of proteins of different batches of the same product allowed for assessing variability and consistency of input products and processing. Of the MFGM products, P1 was the most consistent, with an average CV of 6.43% ± 2.92 across 5 batches for all proteins quantified, except immunoglobulins, which fell below the limit of detection in some samples. Two other products were similar in consistency with WPC_alac and WPC70, with average CV of 8.24% ± 6.48 and 8.99% ± 5.91, respectively. Greater variability in products was found in MFGM_P2 and WPC35, with CV of 17.71% ± 9.83 and 23.08% ± 19.27. Of the protein groups, caseins (CASA1, CASA2, CASB) were inconsistent, with CV of 7.42%, 13.97%, 20.78%, 31.13%, and 31.79% for MFGM_P1, WPC70, WPC_alac, MFGM_P2, and WPC35, respectively. Consistent with sweet whey products, the major whey protein LACB was consistent across all 5 products, with CV of 1.47%, 3.17%, 5.50%, 5.88% to 6.54% for WPC70, WPC35, MFGM_P1, WPC_alac, and MFGM_P2. Of the MFGM proteins, PIGR was most consistent for MFGM_P1 (CV 2.51%) and MFGM_P2 (CV 8.43%).
      This label-free quantification and data analysis across 24 proteins and 32 samples, spanning a range of dairy products from skim milk to high-fat samples, using high mass resolution mass spectrometry, provides valuable insights into composition and consistency of ingredients important for product formulation. The question arises: how many of those 24 proteins must be quantified each time for routine testing of batches using a low-resolution mass spectrometer?

      Reduction of Protein Target List Using Correlation Analysis

      For routine batch testing, it is less important to cover the entire range of 24 proteins, but rather to optimize for speed and cost efficiency of analytics without sacrificing essential information. Many MRM panels have been developed in the past, starting with highly fractionated protein extract spectral libraries and reducing the target MRM list by detecting the proteins in unfractionated samples across a sample set in a reproducible manner (low CV;
      • Ebhardt H.A.
      • Root A.
      • Liu Y.
      • Gauthier N.P.
      • Sander C.
      • Aebersold R.
      Systems pharmacology using mass spectrometry identifies critical response nodes in prostate cancer.
      ). A reductionist approach based on limit of quantification results in several proteins quantified with redundant information values across a given sample set (
      • Ebhardt H.A.
      • Root A.
      • Liu Y.
      • Gauthier N.P.
      • Sander C.
      • Aebersold R.
      Systems pharmacology using mass spectrometry identifies critical response nodes in prostate cancer.
      ). Quantifying proteins with little or no additional information value will increase the number of proteins quantified per MRM method, increasing the number of stable isotope-labeled ([13C6,15N4]Arg or [13C6,15N2]Lys) reference peptides to be synthesized, adds data analysis time after acquisition, and increases costs.
      To rationally reduce the protein target list for MRM method development, correlation analysis was applied to the acquired data set (
      • Friendly M.
      Corrgrams: Exploratory displays for correlation matrices.
      ). The influence of batch testing was reduced by averaging protein values per product before correlation analysis. The output of the correlation analysis is shown in Figure 2. The correlation plot depicts blue right-tilted ellipses for proteins with a high correlation across the data set, whereas red left-tilted ellipses indicate proteins with anti-correlation across the data set. The color code for degree of correlation is shown in the legend. High correlations are further emphasized by narrow ellipses. Supplemental Figure S1 (https://panoramaweb.org/CorrelationAnalysis.url) shows additional numerical correlation values in the top right triangle, which may be used for objective reduction of redundant protein quantification. The color coding and numerical values allow for grouping of proteins, described subsequently, with proteins showing similar behavior across the tested data set.
      Not surprisingly, all quantified casein proteins correlate well, meaning they are correlated or anti-correlated compared with other proteins. A simple plausible explanation is that caseins are typically organized in casein micelles (
      • Tuinier R.
      • de Kruif C.G.
      Stability of casein micelles in milk.
      ) and therefore are expected to co-localize during processing of dairy products. For MRM method development, of the 4 caseins quantified in PRM mode only one is required for quantification in MRM mode without losing vital information. Also consistent with prior knowledge, ALBU and IgG1/IgG2 (heavy chain fragments) correlate well (
      • Levieux D.
      • Ollier A.
      Bovine immunoglobulin G, β-lactoglobulin, α-lactalbumin and serum albumin in colostrum and milk during the early post partum period.
      ;
      • El-Hatmi H.
      • Levieux A.
      • Levieux D.
      Camel (Camelus dromedarius) immunoglobulin G, α-lactalbumin, serum albumin and lactoferrin in colostrum and milk during the early post partum period.
      ). Correlation analysis of these 7 proteins strongly suggests that a protein from each group will be sufficient to explain the variance in the data. As immunoglobulins undergo frequent genomic changes (
      • Sen D.
      • Gilbert W.
      Formation of parallel four-stranded complexes by guanine-rich motifs in DNA and its implications for meiosis.
      ;
      • Saini S.S.
      • Farrugia W.
      • Muthusamy N.
      • Ramsland P.A.
      • Kaushik A.K.
      Structural evidence for a new IgG1 antibody sequence allele of cattle.
      ) and sometimes fall below the limit of quantification, the authors avoided immunoglobulin analytes.
      Based on our data, reduction per protein family, such as minor whey or MFGM proteins, is not advisable. Minor whey proteins such as OSTP and TRFL, which are known to form a protein complex (
      • Liu L.
      • Jiang R.
      • Lönnerdal B.
      Assessment of bioactivities of the human milk lactoferrin–osteopontin complex in vitro.
      ), are not well correlated, as implied by the distance of these 2 proteins within the correlation plot: OSTP is found in the middle of the plot and TRFL near the top. Similarly, MFGM protein pairs BT1A1:MFGM and FABPH:XDH correlate well with each other, with 0.98 and 0.99, respectively (Supplemental Figure S1), allowing only one protein to be quantified per pair. Although all 4 proteins are MFGM-associated, only one protein per pair requires quantification to reflect the variance within the quantified sample set (FABPH:MFGM correlation = 0.83). Consistent with prior knowledge, α-lac, a major whey protein, and PLIN2 (ADPH, MFGM protein) show distinct correlation patterns and should both be quantified in the final MRM assay. This series of reductionist decisions results in 3 individual proteins to be quantified, as their correlation patterns are distinct. Further, there are 6 protein groups of which only one member requires quantification (see Table 1 for details). Together, 9 proteins require quantification for a routine batch-checking MRM method without losing any vital information. This is a near 3-fold reduction in proteins requiring quantification, method development, and subsequent data analysis time. Using these 9 proteins, dairy products were re-clustered using data from Figure 1, resulting in the same clustering of protein analytes and dairy grouping of products, which form 5 clades (Figure 3).
      Figure thumbnail gr3
      Figure 3Reduced number of protein analytes explaining variance in the data set. Using the reductionist approach of the correlation analysis, the remaining 9 proteins were used to cluster the original data displayed in using the same hierarchical Euclidian clustering algorithm. The result shown here groups protein analytes into the same 3 classes, and all dairy products cluster into the same clades as seen in .

      Trinity-MRM Method Development

      Based on the best-performing peptides of the PRM data and the public MRM data depository Panorama (
      • Sharma V.
      • Eckels J.
      • Taylor G.K.
      • Shulman N.J.
      • Stergachis A.B.
      • Joyner S.A.
      • Yan P.
      • Whiteaker J.R.
      • Halusa G.N.
      • Schilling B.
      • Gibson B.W.
      • Colangelo C.M.
      • Paulovich A.G.
      • Carr S.A.
      • Jaffe J.D.
      • MacCoss M.J.
      • MacLean B.
      Panorama: A targeted proteomics knowledge base.
      ), 9 peptides were selected and chemically synthesized to include a COOH-terminal stable isotope-labeled [13C6,15N4]Arg or [13C6,15N2]Lys residue, which are also referred to as heavy reference standards. The COOH-terminal AA was stable-isotope labeled to ensure numerical values for m/z selection in quadrupoles 1 and 3 of the QQQ mass spectrometer were distinct from the unlabeled endogenous peptide. The synthetic peptides initially served to develop QQQ mass spectrometer-specific MRM assays, including optimized transitions and retention times. After method development and optimization, synthetic peptides serve as heavy reference standards.
      For MRM assay development, each peptide was separately analyzed on the reverse-phase C18 column in MS1 scanning mode, based on theoretical m/z values for doubly and triply charged precursors. The total ion current was extracted to obtain a retention time as shown in Figure 4A for peptide TPLPLAGPPR (BT1A1). Using MS2 scanning capabilities of the Agilent QQQ 6495A mass spectrometer, a low-resolution MS2 spectrum was obtained (Figure 4B) and product ions matched to the theoretical values to identify the most intense product ions using Skyline software. When the most abundant ion was identified as the precursor ion, collision energy optimization was carried out to obtain an even fragment ion ladder (
      • Ebhardt H.A.
      • Nan J.
      • Chaulk S.G.
      • Fahlman R.P.
      • Aebersold R.
      Enzymatic generation of peptides flanked by basic amino acids to obtain MS/MS spectra with 2× sequence coverage.
      ), as shown in Figure 4C. The final list of most intense transitions was confirmed in MRM mode, as shown in Figure 4D. As MS2 spectra, collision energy optimization, and MRM assays were carried out on the same mass spectrometer, a high correlation between obtained MS2 spectra and MRM trace was observed. For example, y7 (m/z 717) is the most intense ion in MRM mode (Figure 4D) and optimized MS2 (Figure 4C). Optimization was repeated for all synthetic peptides.
      Figure thumbnail gr4
      Figure 4Trinity-MRM (multiple reaction monitoring) development, MS component. (A) Peptide NH2-TPLPLAGPPR-COOH was chemically synthesized using Arg[13C6, 15N4] and analyzed using HPLC coupled to a triple quadrupole (QQQ) mass spectrometer in MS1 scanning mode for MS2 fragmentation. The total ion current (TIC) shows the peptide's precise RT. (B) The initial MS2 fragmentation pattern was extracted from TIC. The main peak (m/z 514) corresponds to the [M+2H+]2+ precursor, suggesting insufficient fragmentation at a CE of 17 eV. (C) Result of CE optimization: precursor fragment is not the dominant peak anymore at a CE of 23 eV. (D) Most abundant product ions from panel C were quantified in MRM mode, and ion chromatogram plus ion legend is shown. Note: as both the MS2 low-resolution spectrum and MRM measurements were carried out on the same QQQ mass spectrometer, the MRM ion chromatogram intensities correspond very well with the MS2 spectrum (e.g., y7+ ion is the most intense product ion). RT = retention time; ESI = electrospray ionization; Frag = fragment; TIC = total ion current; CID = collision induced dissociation.
      To obtain the complete Trinity-MRM assay, all peptides were simultaneously analyzed in MRM mode. Following LC gradient optimization, peptides elute across the entire 20-min gradient, as shown in Figure 5A. Empirically determined retention times of the final LC method were compared with theoretical data calculated using SSRCalc 3.0 calculator and demonstrates a good linear correlation between theoretical and empirical values, as shown in Figure 5B, with r = 0.97. Having established the Trinity-MRM quantifying casein, whey and MFGM proteins in a single MRM method using synthetic peptides, the focus shifted to method reproducibility of MFGM-enriched samples.
      Figure thumbnail gr5
      Figure 5Trinity-MRM (multiple reaction monitoring) optimization, liquid chromatography (LC) component. (A) Optimized LC gradient where peptides elute off an C18 reverse-phase chromatography column over a 15-min time frame. (B) Correlation plot between theoretically calculated RT (SSRCalc 3.0) based on the AA sequence of the peptide versus empirical measurements (r = 0.97).

      Protein Extraction Method of High-Fat Dairy Products

      Trinity-MRM method development was performed with synthetic peptides in absence of a sample matrix. As peptide retention times may vary with sample matrix (
      • Rosenberger G.
      • Koh C.C.
      • Guo T.
      • Röst H.L.
      • Kouvonen P.
      • Collins B.C.
      • Heusel M.
      • Liu Y.
      • Caron E.
      • Vichalkovski A.
      • Faini M.
      • Schubert O.T.
      • Faridi P.
      • Ebhardt H.A.
      • Matondo M.
      • Lam H.
      • Bader S.L.
      • Campbell D.S.
      • Deutsch E.W.
      • Moritz R.L.
      • Tate S.
      • Aebersold R.
      A repository of assays to quantify 10,000 human proteins by SWATH-MS.
      ), an MFGM-enriched sample was sought to establish retention times in a complex sample. Double cream (DC), with 48% fat content, represents a readily available dairy product naturally enriched in MFGM proteins and challenging for proteomics workflows. For protein purification from DC, a rapid chloroform:water:methanol extraction method was developed (see Methods and Materials for full details). Liquid extraction was chosen over acetone precipitation or solid extraction methods due to its speed and consistency during the development process using DC.
      To establish CV values of the Trinity-MRM method in combination with the protein extraction method, a DC sample was quantified on 3 consecutive days, with 3 samples each day, resulting in 9 peptide samples. Each peptide sample was injected 3 times into the LC-MS/MS system, yielding 27 result files. Primary ion chromatograms are shown in Figure 6A for XDH peptide VSLSTTGFYR whereby the lines in Figure 6A.1 represent the sum of 3 transitions quantified: red for the sum of endogenous signal and blue for the corresponding spike-in stable isotope-labeled peptide. Panels 6A.2 and 6A.3 show co-eluting transitions of endogenous and spike-in reference peptides. Figure 6B shows peak intensities of the entire data set for TRFL peptide SFQLFGSPPGQR and MUC1 peptide VSLYFLSFR. The first 3 injections of the same sample each day show the greatest variability: for example, on d 3, technical variability within sample 1 had an average CV of 10.85% across 3 injections of the same sample, whereas samples 2 and 3 had average CV of 2.45% and 2.13%, respectively. This variability was either an increasing trend in the case of the TRFL peptide, or a decreasing trend in the case of MUC1 peptide. As the trends are opposing for different peptides, this does not seem to be a C18 column-conditioning phenomenon at the start-up of the LC-MS/MS system. Subsequent worklists had a pooled sample injected repeatedly before quantitative measurements to overcome this variance issue. The majority of technical replicas across 9 samples and 9 peptides had a CV below 5%, with the low-abundance MUC1 protein contributing the highest variability, with up to 19.8% in the data set on d 1 sample 1 (Figure 6C, 1TR). A daily average CV was computed to be 4.76% (Figure 6C, 2DV), which exceeds current MRM inter-laboratory standards.
      Figure thumbnail gr6
      Figure 6Reproducibility data set. (A) Shown in A.1 is the ion chromatograms of XDH peptide VSLSTTGFYR from endogenous (red) and spike-in stable isotope labeled reference peptide (blue; sum of transitions). A.2 and A.3 show individual transitions for endogenous and spike-in reference peptide. (B) Total peak area of endogenous and spike-in peptide across the entire data set of 3 samples, 3 replicas per sample over 3 d. Note that the first 3 injections per day have a greater variance than samples 2 and 3 of the same day. (C) CV of sample reproducibility: 1TR = technical reproducibility per sample, 2DV = daily variability, 3IL = inter-laboratory study showing CV per user, 4All = overall CV of inter-laboratory study. (D) Example of peak area bar graphs from inter-laboratory study. Note: before the quantification samples, 3 pooled injections were quantified resulting in an even quantification series in comparison with panel B.
      Having demonstrated high repeatability of the protein extraction and Trinity-MRM methods, an intra-laboratory reproducibility study was carried out. Three laboratory members extracted the same DC sample on the same day in quadruplicate, injecting each sample twice. This setup resulted in 4 extractions per laboratory member and a total of 24 injections for quantification. With the knowledge gained from the repeatability study, 3 pooled samples were injected before quantification runs to allow for signal stabilization. Comparing Figure 6B with 6D clearly shows a more consistent peak area at start-up of each data set. The average CV of the ratio between DC-derived peptide and spike-in stable isotope-labeled reference standard per user were 2.04%, 4.85%, and 2.73%, respectively (Figure 6C, 3IL). Overall, the intra-laboratory study achieved a CV of 3.21% (Figure 6C, 4All). The intra-laboratory study presented here was carried out by starting from the unprocessed DC sample, performing protein extraction protocol and digestion protocol independently, which demonstrates the robustness of the entire workflow. In comparison, a recent inter-laboratory study using the same biological samples and independently prepared peptide samples demonstrated a CV of 11% for biomarker analytes (
      • Bringans S.
      • Ito J.
      • Casey T.
      • Thomas S.
      • Peters K.
      • Crossett B.
      • Coleman O.
      • Ebhardt H.A.
      • Pennington S.R.
      • Lipscombe R.
      A robust multiplex immunoaffinity mass spectrometry assay (PromarkerD) for clinical prediction of diabetic kidney disease.
      ).

      Data Availability

      Mass spectrometry data are available on PanoramaWeb Public.

      PRM Data

      MRM Data

      CONCLUSIONS

      To our knowledge, this is the first time correlation analysis has been applied to reduce an omics data set for routine quantification of proteins using MRM without losing vital information. Based on whey, MFGM, and skim milk products, a quantitative proteomics data matrix was established. Three MFGM products were correlated with whey protein extracts, whereas one MFGM product showed a distinct protein profile more closely related to skim milk. Correlation analysis of the analyzed data matrix revealed that a large portion of protein quantification data were redundant and could be omitted for faster analysis. The resulting Trinity-MRM method contains peptides for 9 proteins, down from the original 24 proteins, which represents a 63% reduction in analytes without loss of information. This reduction will lead to a decrease in LC-MS/MS analysis time, LC method shortening, and increase in dwell time per transition, which increases accuracy and sensitivity. This approach could be applicable for other dairy products and systems pharmacology approaches.

      ACKNOWLEDGMENTS

      This study received no external funding. The authors have not stated any conflicts of interest.

      REFERENCES

        • Addona T.A.
        • Abbatiello S.E.
        • Schilling B.
        • Skates S.J.
        • Mani D.R.
        • Bunk D.M.
        • Spiegelman C.H.
        • Zimmerman L.J.
        • Ham A.J.
        • Keshishian H.
        • Hall S.C.
        • Allen S.
        • Blackman R.K.
        • Borchers C.H.
        • Buck C.
        • Cardasis H.L.
        • Cusack M.P.
        • Dodder N.G.
        • Gibson B.W.
        • Held J.M.
        • Hiltke T.
        • Jackson A.
        • Johansen E.B.
        • Kinsinger C.R.
        • Li J.
        • Mesri M.
        • Neubert T.A.
        • Niles R.K.
        • Pulsipher T.C.
        • Ransohoff D.
        • Rodriguez H.
        • Rudnick P.A.
        • Smith D.
        • Tabb D.L.
        • Tegeler T.J.
        • Variyath A.M.
        • Vega-Montoto L.J.
        • Wahlander A.
        • Waldemarson S.
        • Wang M.
        • Whiteaker J.R.
        • Zhao L.
        • Anderson N.L.
        • Fisher S.J.
        • Liebler D.C.
        • Paulovich A.G.
        • Regnier F.E.
        • Tempst P.
        • Carr S.A.
        Multi-site assessment of the precision and reproducibility of multiple reaction monitoring-based measurements of proteins in plasma.
        Nat. Biotechnol. 2009; 27 (19561596): 633-641
        • Affolter M.
        • Grass L.
        • Vanrobaeys F.
        • Casado B.
        • Kussmann M.
        Qualitative and quantitative profiling of the bovine milk fat globule membrane proteome.
        J. Proteomics. 2010; 73 (19944786): 1079-1088
        • Arranz E.
        • Corredig M.
        Invited review: Milk phospholipid vesicles, their colloidal properties, and potential as delivery vehicles for bioactive molecules.
        J. Dairy Sci. 2017; 100 (28343627): 4213-4222
        • Astaire J.C.
        • Ward R.
        • German J.B.
        • Jiménez-Flores R.
        Concentration of polar MFGM lipids from buttermilk by microfiltration and supercritical fluid extraction.
        J. Dairy Sci. 2003; 86 (12906046): 2297-2307
        • Bär C.
        • Mathis D.
        • Neuhaus P.
        • Dürr D.
        • Bisig W.
        • Egger L.
        • Portmann R.
        Protein profile of dairy products: Simultaneous quantification of twenty bovine milk proteins.
        Int. Dairy J. 2019; 97: 167-175
        • Barone G.
        • O'Regan J.
        • O'Mahony J.A.
        Influence of composition and microstructure on bulk handling and rehydration properties of whey protein concentrate powder ingredients enriched in α-lactalbumin.
        J. Food Eng. 2019; 255: 41-49
        • Braakman R.B.H.
        • Bezstarosti K.
        • Sieuwerts A.M.
        • de Weerd V.
        • van Galen A.M.
        • Stingl C.
        • Luider T.M.
        • Timmermans M.A.M.
        • Smid M.
        • Martens J.W.M.
        • Foekens J.A.
        • Demmers J.A.A.
        • Umar A.
        Integrative analysis of genomics and proteomics data on clinical breast cancer tissue specimens extracted with acid guanidinium thiocyanate–phenol–chloroform.
        J. Proteome Res. 2015; 14 (25611981): 1627-1636
        • Bringans S.
        • Ito J.
        • Casey T.
        • Thomas S.
        • Peters K.
        • Crossett B.
        • Coleman O.
        • Ebhardt H.A.
        • Pennington S.R.
        • Lipscombe R.
        A robust multiplex immunoaffinity mass spectrometry assay (PromarkerD) for clinical prediction of diabetic kidney disease.
        Clin. Proteomics. 2020; 17 (33093819): 37
        • Brink L.R.
        • Lonnerdal B.
        Milk fat globule membrane: The role of its various components in infant health and development.
        J. Nutr. Biochem. 2020; 85 (32758540)108465
        • Chaulk S.G.
        • Ebhardt H.A.
        • Fahlman R.P.
        Correlations of microRNA: MicroRNA expression patterns reveal insights into microRNA clusters and global microRNA expression patterns.
        Mol. Biosyst. 2016; 12 (26563430): 110-119
        • Cheong L.Z.
        • Jiang C.
        • He X.
        • Song S.
        • Lai O.M.
        Lipid profiling, particle size determination, and in vitro simulated gastrointestinal lipolysis of mature human milk and infant formula.
        J. Agric. Food Chem. 2018; 66 (30362342): 12042-12050
        • Christian P.
        • Smith E.R.
        • Lee S.E.
        • Vargas A.J.
        • Bremer A.A.
        • Raiten D.J.
        The need to study human milk as a biological system.
        Am. J. Clin. Nutr. 2021; 113 (33831952): 1063-1072
        • Costa M.R.
        • Elias-Argote X.E.
        • Jiménez-Flores R.
        • Gigante M.L.
        Use of ultrafiltration and supercritical fluid extraction to obtain a whey buttermilk powder enriched in milk fat globule membrane phospholipids.
        Int. Dairy J. 2010; 20: 598-602
        • de Toledo Guimarães J.
        • Silva E.K.
        • de Freitas M.Q.
        • de Almeida Meireles M.A.
        • da Cruz A.G.
        Non-thermal emerging technologies and their effects on the functional properties of dairy products.
        Curr. Opin. Food Sci. 2018; 22: 62-66
        • Delosière M.
        • Pires J.A.A.
        • Bernard L.
        • Cassar-Malek I.
        • Bonnet M.
        Dataset reporting 4654 cow milk proteins listed according to lactation stages and milk fractions.
        Data Brief. 2020; 29 (32016138)105105
        • Ebhardt H.A.
        • Nan J.
        • Chaulk S.G.
        • Fahlman R.P.
        • Aebersold R.
        Enzymatic generation of peptides flanked by basic amino acids to obtain MS/MS spectra with 2× sequence coverage.
        Rapid Commun. Mass Spectrom. 2014; 28 (25380496): 2735-2743
        • Ebhardt H.A.
        • Root A.
        • Liu Y.
        • Gauthier N.P.
        • Sander C.
        • Aebersold R.
        Systems pharmacology using mass spectrometry identifies critical response nodes in prostate cancer.
        NPJ Syst. Biol. Appl. 2018; 4 (29977602): 26
        • Ebhardt H.A.
        • Root A.
        • Sander C.
        • Aebersold R.
        Applications of targeted proteomics in systems biology and translational medicine.
        Proteomics. 2015; 15 (26097198): 3193-3208
        • El-Hatmi H.
        • Levieux A.
        • Levieux D.
        Camel (Camelus dromedarius) immunoglobulin G, α-lactalbumin, serum albumin and lactoferrin in colostrum and milk during the early post partum period.
        J. Dairy Res. 2006; 73 (16566855): 288-293
        • Fontecha J.
        • Brink L.
        • Wu S.
        • Pouliot Y.
        • Visioli F.
        • Jiménez-Flores R.
        Sources, production, and clinical treatments of milk fat globule membrane for infant nutrition and well-being.
        Nutrients. 2020; 12 (32486129)1607
        • Friendly M.
        Corrgrams: Exploratory displays for correlation matrices.
        Am. Stat. 2002; 56: 316-324
        • Fuerer C.
        • Jenni R.
        • Cardinaux L.
        • Andetsion F.
        • Wagniere S.
        • Moulin J.
        • Affolter M.
        Protein fingerprinting and quantification of β-casein variants by ultra-performance liquid chromatography-high-resolution mass spectrometry.
        J. Dairy Sci. 2020; 103 (31759609): 1193-1207
        • Gong H.
        • Yuan Q.
        • Pang J.
        • Li T.
        • Li J.
        • Zhan B.
        • Chang R.
        • Mao X.
        Dietary milk fat globule membrane restores decreased intestinal mucosal barrier development and alterations of intestinal flora in infant-formula-fed rat pups.
        Mol. Nutr. Food Res. 2020; 64 (32918844)e2000232
        • Huang Z.
        • Zheng H.
        • Brennan C.S.
        • Mohan M.S.
        • Stipkovits L.
        • Li L.
        • Kulasiri D.
        Production of milk phospholipid-enriched dairy ingredients.
        Foods. 2020; 9 (32121655): 263
        • Jaramillo-Ospina A.M.
        • Toro-Campos R.
        • Murguía-Peniche T.
        • Wampler J.L.
        • Wu S.S.
        • Berseth C.L.
        • Uauy R.
        Added bovine milk fat globule membrane in formula: Growth, body composition, and safety through age 2: An RCT.
        Nutrition. 2022; 97 (35193081)111599
        • Kiel C.
        • Ebhardt H.A.
        • Burnier J.
        • Portugal C.
        • Sabido E.
        • Zimmermann T.
        • Aebersold R.
        • Serrano L.
        Quantification of ErbB network proteins in three cell types using complementary approaches identifies cell-general and cell-type-specific signaling proteins.
        J. Proteome Res. 2014; 13 (24313378): 300-313
        • Levieux D.
        • Ollier A.
        Bovine immunoglobulin G, β-lactoglobulin, α-lactalbumin and serum albumin in colostrum and milk during the early post partum period.
        J. Dairy Res. 1999; 66 (10480080): 421-430
        • Li M.
        • Zheng K.
        • Song W.
        • Yu H.
        • Zhang X.
        • Yue X.
        • Li Q.
        Quantitative analysis of differentially expressed milk fat globule membrane proteins between donkey and bovine colostrum based on high-performance liquid chromatography with tandem mass spectrometry proteomics.
        J. Dairy Sci. 2021; 104 (34531055): 12207-12215
        • Liu L.
        • Jiang R.
        • Lönnerdal B.
        Assessment of bioactivities of the human milk lactoferrin–osteopontin complex in vitro.
        J. Nutr. Biochem. 2019; 69 (31048205): 10-18
        • Lutter P.
        • Parisod V.
        • Weymuth H.
        Development and validation of a method for the quantification of milk proteins in food products based on liquid chromatography with mass spectrometric detection.
        J. AOAC Int. 2011; 94 (21919337): 1043-1059
        • Manoni M.
        • Di Lorenzo C.
        • Ottoboni M.
        • Tretola M.
        • Pinotti L.
        Comparative proteomics of milk fat globule membrane (MFGM) proteome across species and lactation stages and the potentials of MFGM fractions in infant formula preparation.
        Foods. 2020; 9 (32906730)1251
        • Pino L.K.
        • Searle B.C.
        • Bollinger J.G.
        • Nunn B.
        • MacLean B.
        • MacCoss M.J.
        The Skyline ecosystem: Informatics for quantitative mass spectrometry proteomics.
        Mass Spectrom. Rev. 2020; 39 (28691345): 229-244
        • Ramachandran B.
        • Yang C.T.
        • Downs M.L.
        Parallel reaction monitoring mass spectrometry method for detection of both casein and whey milk allergens from a baked food matrix.
        J. Proteome Res. 2020; 19 (32483969): 2964-2976
        • Rosenberger G.
        • Koh C.C.
        • Guo T.
        • Röst H.L.
        • Kouvonen P.
        • Collins B.C.
        • Heusel M.
        • Liu Y.
        • Caron E.
        • Vichalkovski A.
        • Faini M.
        • Schubert O.T.
        • Faridi P.
        • Ebhardt H.A.
        • Matondo M.
        • Lam H.
        • Bader S.L.
        • Campbell D.S.
        • Deutsch E.W.
        • Moritz R.L.
        • Tate S.
        • Aebersold R.
        A repository of assays to quantify 10,000 human proteins by SWATH-MS.
        Sci. Data. 2014; 1 (25977788)140031
        • Röst H.
        • Malmström L.
        • Aebersold R.
        A computational tool to detect and avoid redundancy in selected reaction monitoring.
        Mol. Cell. Proteomics. 2012; 11 (22535207): 540-549
        • Saini S.S.
        • Farrugia W.
        • Muthusamy N.
        • Ramsland P.A.
        • Kaushik A.K.
        Structural evidence for a new IgG1 antibody sequence allele of cattle.
        Scand. J. Immunol. 2007; 65 (17212764): 32-38
        • Schneider N.
        • Bruchhage M.M.K.
        • O'Neill B.V.
        • Hartweg M.
        • Tanguy J.
        • Steiner P.
        • Mutungi G.
        • O'Regan J.
        • McSweeney S.
        • D'Sa V.
        • Deoni S.C.L.
        A nutrient formulation affects developmental myelination in term infants: A randomized clinical trial.
        Front. Nutr. 2022; 9 (35242798)823893
        • Sen D.
        • Gilbert W.
        Formation of parallel four-stranded complexes by guanine-rich motifs in DNA and its implications for meiosis.
        Nature. 1988; 334 (3393228): 364-366
        • Sharma V.
        • Eckels J.
        • Taylor G.K.
        • Shulman N.J.
        • Stergachis A.B.
        • Joyner S.A.
        • Yan P.
        • Whiteaker J.R.
        • Halusa G.N.
        • Schilling B.
        • Gibson B.W.
        • Colangelo C.M.
        • Paulovich A.G.
        • Carr S.A.
        • Jaffe J.D.
        • MacCoss M.J.
        • MacLean B.
        Panorama: A targeted proteomics knowledge base.
        J. Proteome Res. 2014; 13 (25102069): 4205-4210
        • Singh H.
        The milk fat globule membrane—A biophysical system for food applications.
        Curr. Opin. Colloid Interface Sci. 2006; 11: 154-163
        • Suttapitugsakul S.
        • Xiao H.
        • Smeekens J.
        • Wu R.
        Evaluation and optimization of reduction and alkylation methods to maximize peptide identification with MS-based proteomics.
        Mol. Biosyst. 2017; 13 (29019370): 2574-2582
        • Timby N.
        • Domellöf E.
        • Hernell O.
        • Lönnerdal B.
        • Domellöf M.
        Neurodevelopment, nutrition, and growth until 12 mo of age in infants fed a low-energy, low-protein formula supplemented with bovine milk fat globule membranes: A randomized controlled trial.
        Am. J. Clin. Nutr. 2014; 99 (24500150): 860-868
        • Tuinier R.
        • de Kruif C.G.
        Stability of casein micelles in milk.
        J. Chem. Phys. 2002; 117: 1290-1295
        • UniProt Consortium
        UniProt: The universal protein knowledgebase.
        Nucleic Acids Res. 2017; 45 (27899622): D158-D169
        • Zhao L.
        • Du M.
        • Gao J.
        • Zhan B.
        • Mao X.
        Label-free quantitative proteomic analysis of milk fat globule membrane proteins of yak and cow and identification of proteins associated with glucose and lipid metabolism.
        Food Chem. 2019; 275 (30724237): 59-68
        • Zheng H.
        • Jiménez-Flores R.
        • Everett D.W.
        Bovine milk fat globule membrane proteins are affected by centrifugal washing processes.
        J. Agric. Food Chem. 2013; 61 (23931678): 8403-8411