Worldwide, food-allergy-related diseases are a significant health problem. While the food industry works on managing cross-contaminations and while clinicians deal with treatment, laboratories must develop efficient analytical methods to ensure detection of hidden allergens that can cause severe adverse reactions. Over the past few years, huge progress has been made in mass spectrometry for the analysis of allergens in incurred and processed foodstuffs, especially as regards sample preparation and enrichment (solid phase extraction, protein precipitation and ultrafiltration). These achievements make it possible to meet the Allergen Bureau's Voluntary Incidental Trace Allergen Labelling (VITAL) sensitivity criteria. The present chapter details the different steps in the development of mass spectrometry methods, from peptide selection to the validation of qualitative and quantitative methods. The chapter focuses mainly on studies performed with incurred and processed food samples to ensure the applicability of the methods to allergen detection in real food products.
- mass spectrometry
Food allergies have increased significantly, affecting between 3 and 4% of adults and at least 6% of children . According to the European Academy of Allergology and Clinical Immunology (EAACI), the prevalence of food allergy has doubled over the past 10 years . After an adverse reaction to a foodstuff, which may range from mild to severe (e.g. anaphylaxis) [3, 4], allergic patients have to exclude that foodstuff from their diet. Each year in the United States, some 100 deaths are caused by anaphylaxis due to food allergy , the main culprits being allergens from peanut, tree nuts, fish, shellfish and milk . Currently, there exist no treatments for food allergy, but clinical trials have been performed to test subcutaneous immunotherapy and oral immunotherapy used to desensitize patients . The high level of adverse reactions observed in these trials has led clinicians to find safer alternative therapies, such as sublingual and epicutaneous immunotherapy. These approaches consist, respectively, in placing allergens (drops or tablets) under the tongue or in using a skin patch to induce sustained protection against anaphylaxis . Although they do not treat allergic disease, they improve considerably the quality of life of highly allergic patients and constitute a real hope for them [9, 10]. The number of potentially allergenic ingredients that must appear on food labels differs in different parts of the world . In Europe, regulation (EU) 1169/2011 imposes indicating the following 14 ingredients: milk, peanut, egg, soybean, fish, crustaceans, cereals containing gluten, tree nuts, celery, lupin, mustard, sesame, molluscs and sulfur dioxide . This regulation fails to take into account the accidental introduction of allergens during production, transportation or storage, even though allergens introduced in this manner can trigger severe reactions [13–15]. To protect food consumers, the industry has widely used precautionary allergen labelling (PAL) (i.e. statements such as ‘may contain’, ‘may contain traces of…’) . Yet, the lack of correlation between the presence of allergens and precautionary labelling has led customers to lose trust in food labels [17–20]. In a study of food product recalls over a four-year period in the European Union, the United States, Canada, Hong Kong, Australia and New Zealand, 42–90% of the recalls, depending on the country, were justified by the presence of allergens not indicated on the label . Between 2007 and 2012, the Food and Drug Administration (FDA) recalled 732 products because of allergen contaminations  and allergic reactions are due to five foods: milk, egg, peanuts, wheat and soybean (Figure 1).
The distribution of these recalls in the European Union, reported in Figure 2, shows that the products most commonly involved in food recalls are cereals and bakery products.
The widespread use of PAL can be explained by the lack of regulatory thresholds and the complexity of food allergen management through the supply chain. To counter this lack, the Voluntary Incidental Trace Allergen Labelling (VITAL) system has been developed in Australia and New Zealand to assist food producers in managing cross-contaminations during food production . This system sets allergen thresholds, based on clinical studies, for the protection of 95–99% of the allergic population. Other referentials for allergen thresholds are the European Academy of Allergy and Clinical Immunology (EAACI) and the Netherlands Food and Consumer Product Safety Authority (NVWA)  (Table 1).
|Food||Reference dose VITAL (mg of proteins) [27, 34, 37]||Reference dose |EAACI (mg of proteins) ||Reference dose NVWA (mg of proteins) ||Reference dose VITAL (mg of proteins per kg) Portion size: 50 g|
While the systems just mentioned have no regulatory value, food laboratories use them in evaluating method sensitivity. To obtain a concentration expressed in ‘mg proteins per kilogram’, a food portion size must be considered in order to compare the analytical method with VITAL thresholds (e.g. a portion size of 50 g, Table 1). Yet while VITAL thresholds are expressed in ‘mg proteins’, laboratories express their results in ‘mg ingredients’ [29, 30] or may refer either to soluble proteins [31, 32] or total proteins  per kg. To compare method performances, a conversion factor must thus be applied (e.g. 25% proteins in whole peanuts ). Moreover, VITAL action levels have been determined from clinical studies, mostly on the basis of the allergenicity of raw ingredients, although studies have demonstrated a major decrease in allergenicity in baked products. For example, 50–85% of allergic children are able to tolerate baked egg  and a study published in 2015 found 63% to tolerate 3.8 g egg-white protein in baked-egg products .
Nevertheless, the prevalence of baked product recalls confirms that laboratories must develop sensitive methods for detecting allergens in processed foodstuffs. The most widely used methods are based on the recognition of allergen proteins by antibodies, notably lateral flow device methods and enzyme-linked immunosorbent assays (ELISAs) . DNA-based methods such as those exploiting the real-time polymerase chain reaction (PCR)  are also used to detect the presence of allergens. Currently, mass spectrometry is becoming an alternative to these methods, as heat-processing induces protein denaturation and structural modifications that might result in non-recognition of the target protein by conformational antibodies and thus in the case of ELISAs, lead to false negatives or at least major underestimation of allergen content [41–44]. Mass spectrometry has the advantage of permitting simultaneous analyses for several allergens in food, including processed food products, with high sensitivity and specificity.
This chapter highlights the important improvements made over the last 10 years in mass spectrometry applied to the development of allergen detection methods. It covers and discusses the mass spectrometry methods currently used to detect and quantify allergens in processed food products, including their validation.
2. Detecting food allergens
2.1. Selecting marker peptides
Food allergens (except sulfites) are proteins that need to be digested by enzymes (trypsin and chymotrypsin) so as to generate peptides suitable for routine mass spectrometry analysis. Identification and selection of robust peptides are generally done first on digested raw ingredients before analysis of digested processed ingredients in food matrices. This section summarizes two approaches commonly used to select marker peptides (the instrumental approach and the in silico approach) and the specificity and sensitivity critera used.
2.1.1. Peptide selection
184.108.40.206. Instrumental peptide selection
The first approach is to identify abundant marker peptides by high resolution mass spectrometry (HRMS). Downstream from allergen analysis by HRMS, the generated data are transferred into an algorithm for assigning peptides to MS/MS spectra (MASCOT, X!Tandem, SEQUEST) . For example, Sealey-Voyksner et al. analysed 12 tree nuts and peanut-raw and roasted (176.7°C, 30 min) by time of flight (q-TOF) (Agilent 6530) spectrometry and selected two abundant peptides per tree nut and four for peanut . In a previous study, ice cream spiked with peanuts was analysed by q-TOF (Waters Micromass II) to identify peptides of the Ara h1 allergen . In a 2012 study, Cucu et al. identified several soybean marker peptides by matrix-assisted laser desorption ionization (MALDI-TOF/MS) . The main advantage of this approach is that global peptide and protein profiles can be analysed for the different samples.
220.127.116.11. In silico peptide selection
Another strategy for selecting marker peptides is to retrieve target protein sequences from a database, e.g. Uniprot (
In silico digestion with multiple reaction monitoring (MRM) involves generating a list of criteria that must be applied or set by the user as regards peptides, transitions and MS/MS parameters (e.g. peptide length, charge states, fragmentation and enzyme). Then raw ingredients or incurred matrices can be analysed by ultra-high performance liquid chromatography coupled to tandem mass spectrometry (UHPLC-MS/MS). This approach allows identification of abundant peptides. It was used by Rogniaux et al. for the analysis of wheat varieties: several gluten peptides were identified with a linear ion trap quadrupole mass filter in tandem with an orbitrap (Thermo Fisher Scientific) .
An in silico approach also requires a complete database with available protein sequences. Uniprot inventories, however, can be too large (e.g. >145,000 proteins for the wheat speciesTriticum aestivum), making it necessary to first select target proteins from the literature. Use of a routine UHPLC-MS/MS instrument is the main advantage of the in silico approach for laboratories unwilling to invest in a high-resolution mass spectrometer.
BLAST: After this selection, blasting must be performed to guarantee the specificity of marker peptides. This step is mandatory but not always included in method development. In one study, for example, Hoofnagle et al. selected five peptides for the detection of β-casein in cookies: EMPFPK (6AA), VLPVPQK (7AA), AVPYPQR (7AA), GPFPIIV (7AA) and DMPIQAFLLYQEPVLGPVR (19AA) . Only one of these peptides could be blasted, and this peptide is 100% homologous to goat, zebu, buffalo, yak and sheep β-casein (Uniprot). In proteomics, peptide blasting should be systematic, even though the international trade frequently introduces new food products and although some proteins can still be missing in the different databases.
The specificity of selected fragments is also paramount. To improve specificity, the mass-to-charge ratio (m/z) of the precursor should be lower than the m/z of the fragments. Too-small fragments should be avoided. At least, fragments of 1 to 2 amino acids (b1, b2, y1, y2) should be excluded, which is not always the case in published methods [53, 54].
Blanks: Matrices without allergens must also be analysed to ensure the specificity of the selected transitions of the target peptides. As databases do not cover all possible proteins and as new food products enter the food chain regularly, this experimental testing is crucial to proving method specificity.
2.1.3. Identifying marker peptides in incurred foodstuffs
The advantage of using mass spectrometry is detection of allergens in industrial food products. For such applications, only target peptides and proteins that will be detected in incurred and processed matrices, such as those listed in Table 2, need to be retained in the analytical methods. Some peptides are common to the majority of published methods: FFVAPFPEVFGK and YLGYLEQLLR (Casein αS1), and GGLEPINFQTAADQAR (ovalbumin), among others. Target peptides detected after different extraction and purification steps in several types of matrices constitute potential marker peptides for the detection of allergens in a wide variety of foodstuffs.
|Heick et al. ||Bread (60 min, 200°C)||Milk||αS1-casein||YLGYLEQLLR||b2, y8|
|Peanut||Ara h1||DLAFPGSGEQVEK||a3, y9|
|Ara h3/4||RPFYSNAPQEIFIQQGR||y6, b7|
|Hazelnut||11S globulin||ADIYTEQVGR||y6, y7|
|Walnut||Jug r1||DLPNECGISSQR||y4, y10|
|Pilolli et al. 2016 ||Cookie (200°C, 12 min)||Milk||αS1-casein||FFVAPFPEVFGK||y8, y9, y10|
|YLGYLEQLLR||y5, y6, y8|
|Egg||Ovalbumin||GGLEPINFQTAADQAR||y7, y10, y12|
|YPILPEYLQCVK||b4, y8, y9|
|Peanut||Conarachin||VLLEENAGGEQEER||y7, y8, y12|
|EGEQEWGTPGSEVR||y6, y8, y9|
|Soy||Glycinin G1-G2||SQSDNFEYVSFK||y3, y10|
|FYLAGNQEQEFLK||y9, y10, y11|
|Hazelnut||11S globulin-like protein||ADIYTEQVGR||y6, y7|
|ALPDDVLANAFQISR||y7, y8, y13|
|Lamberti et al. ||Cookie (180°C, 10 min)||Milk||αS1-casein||YLGYLEQLLR||y8, y9, y10|
|FFVAPFPEVFGK||y8, y9, y10|
|Pedreschi et al. ||Cookie (180°C, 16 min)||Peanut||Ara h1||VLLEENAGGEQEER||y9, y8, y7, y6, y4, y2|
|DLAFPGSGEQVEK||y10, y9, y8, b4, b3, b2|
|Ara h2||CCNELNEFENNQR||y8, y6, y5, y4|
|NLPQQCGLR||y7, y6, y5, a2|
|CDLEVESGGR||y8, y6, y5, y4|
|CMCEALQQIMENQSDR||y14, y11, y10, y8, y7, y6, y5, b2|
|Ara h3||LNAQRPDNR||ymax, y8, y7, y5, b2|
|SPDIYNPQAGSLK||ymax, y12, y9, y8, y7, y5, b3|
|AHVQVVDSNGNR||b7, y6, b5|
|Huschek et al. ||Cookie (190°C, 13 min)||Soy||Gly m6||VFDGELQEGR||903.6/ 489.2/ 788.5|
|LSAEFGLR||432.3/ 779.4/ 579.3|
|LNALKPDNR||742.4/ 629.3/ 501.2|
|Sesame||Ses i6||ISGAQPSLR||472.3/ 728.4/ 671.4|
|AFYLAGGVPR||556.3/ 485.3/ 669.4|
|SPLAGYTSVIR||795.4/ 866.5/ 575.4|
|NPYHFSSQR||761.4/ 624.3/ 477.2|
|Planque et al. [33, 60]||Cookie (180°C—18 min), sauce (95°C, 45 min), ice cream and chocolate||Milk||αS1-Casein||FFVAPFPEVFGK||y6, y8, y9|
|HQGLPQEVLNENLLR||b4, y6, y7|
|YLGYLEQLLR||y5, y6, y7|
|αS2-casein||NAVPITPTLNR||b3, y8, y8|
|β-lactoglobulin||VYVEELKPTPEGDLEILLQK||y11, y14, y16|
|VLVLDTDYK||y5, y6, y7|
|LSFNPTQLEEQCHI||y7, y10, y10|
|Egg||Ovalbumin||GGLEPINFQTAADQAR||y10, y12, y12|
|LTEWTSSNVM EER||y7, y8, y9|
|ISQAVHAAHAEINEAGR||y9, y10, y11|
|Vitellogenin||EALQPIHDLADEAISR||y6, y7, y12|
|NIPFAEYPTYK||y4, y9, y9|
|NIGELGVEK||y5, y6, y7|
|YLLDLLPAAASHR||y7, y7, y11|
|Apovitellenin||NFLINETAR||y5, y6, y7|
|Peanut||Cupin||NTLEAAFNAEFNEIR||y7, y8, y9|
|RPFYSNAPQEIFIQQGR||b7, y6, y10|
|FNLAGNHEQEFLR||y5, y9, y10|
|TANELNLLILR||y6, y7, y8|
|Soy||Glycinin||ISTLNSLTLPALR||y7, y8, y9|
|EAFGVNMQIVR||y5, y6, y7|
|ELINLATMCR||y5, y6, y8|
|LITLAIPVNKPGR||y7, y9, y11|
|Parker et al. ||Muffin (177°C, 48 min)||Egg||Lysozyme||FESNFNTQATNR||Not provided|
|Gomaa et al. 2014 ||Cookie (177°C, 12 min)||Milk||αS1-casein||HQGLPQEVLNENLLR||best transitions not selected|
|Soy||Glycinin G1 precursor||HNIGQTSSPDIYNPQAGSVTTATSLDFPALSWLR|
|Glycinin G2 precursor||TNDRPSIGNLAGANSLLNALPEEVIQHTFNLK|
|Beta conglycinin alpha chain precursor||DLDIFLSIVDMNEGALLLPHFNSK|
|Wheat||Alpha amylase trypsin inhibitor||YFIALPVPSQPVDPR|
2.2. Developing mass spectrometry methods
After selection of marker peptides, the developed method must be able to detect traces of the allergen proteins in the ‘mg allergen proteins per kg food product’ range. The real challenge for laboratories is to achieve this sensitivity with processed foodstuffs. To reach this sensitivity, two factors must be considered: instrument sensitivity and optimization of sample preparation. The different strategies used to evaluate sensitivity are described below.
Instrument sensitivity: No comparison of the sensitivities of different instruments with the same peptide extract has yet been published for allergen analysis, although the sensitivity of the instrument is crucial to the sensitivity of the method, as in the case of other contaminants. One should bear in mind, however, that the most sensitive research-dedicated instrument might not be the best choice for routine analysis (automated injection and short analytical run).
Extraction and purification of proteins: The ideal sample preparation protocol should allow extraction of 100% of the target compounds, the final extract used for MS analysis being as pure as possible. Yet, the preparation of samples for food allergen analysis is difficult, because it should be applicable to a very broad range of food matrices and because the extractability of proteins might be altered in a processed food . In addition, several modifications can occur, e.g. asparagine deamination, the Maillard reaction and several reactions of lysine. Such modifications cause a mass shift of tryptic peptides, resulting in non-recognition of several peptides by mass spectrometry [64–66]. To improve protein extraction, different parameters can be optimized: the composition of extraction buffers, the temperature, the sample-to-buffer ratio and the presence of detergents. Furthermore, the purification step is as important as extraction in order to concentrate proteins in and eliminate interferences from the supernatant. Purification usually involves solid phase extraction (SPE), protein precipitation, ultrafiltration and size exclusion chromatography (SEC), among others. Optimizing extraction and purification is a key step in developing sensitive methods for the detection of allergens by mass spectrometry (Table 3).
|Heick et al. (2010) ||Milk, egg, soy, peanut, hazelnut, walnut, almond||Bread (200°C, 60 min)||2 g/20 ml||Ultrafiltration (Amilcon Ultra 15 mL, 5 kDa molecular weight cut-off) (Millipore)||Dilution: 1 mg of proteins by ml with NH4HCO3 (100 mM)||LC: 1200 HPLC (Agilent)||LOD (S/N>3)|
|Buffer: TRIS-HCl pH 8.2||Aliquot: 100 μl||Column: Xbridge C18 3.5 μm (2.1×150 mm) (Waters)||5 mg of soluble milk proteins by kg|
|Agitation: 60°C for 3h||Reduction: 50 μl DTT (200 mM), 45 min||MS: API 4000QTrap (MDS Sciex)||42 mg of soluble egg proteins by kg|
|Alkylation: 40 μl IA (1 M), 45 min in the dark||24 mg of soluble soy proteins by kg|
|20 μl DTT (200 mM) + 50 μl NH4HCO3 (100 mM)||11 mg of soluble peanut proteins by kg|
|Digestion: 10 μl trypsin (1 μg/μl) 12 h - 37°C||5 mg of soluble hazelnut proteins by kg|
|2 μl formic acid||70 mg of soluble walnut proteins by kg|
|3 mg of soluble almond proteins by kg|
|Pilolli et al. ||milk, egg,soy,peanut, hazelnut||Cookie (200°C, 12 min)||2.5 g/50 ml||1.2 μm acetate cellulose membrane, Size exclusion column (SEC) (G25 Sephadex column)||Elution SEC: 3.5 ml NH4HCO3 (50 mM)||LC: -||LOD (S/N>3)|
|Buffer: 20 mM TRIS-HCl pH 8.2||Aliquot: 300 μl||Column: Aeris Peptide XB-C18 (150 × 2.1 mm) (Phenomenex)||7 mg of milk by kg|
|Protein denaturation: 15 min at 95°C||MS: Dual pressure Linear Ion Trap Spectrometer Velos Pro (Thermo Fisher Scientic)||9 mg of egg by kg|
|Ultrasound: 30 min||Reduction: 15 μl of 50 mM DTT 30 min at 60°C||6 mg of soy by kg|
|Alkylation: 30 μl of 100 mM IAA 30 min in the dark at room temperature||13 mg of peanut by kg|
|Digestion:4 μl trypsin (1 μg/μl) 14h||7 mg of hazelnut by kg|
|Acidification: 1M HCl|
|Filtration: 0.2 μm|
|Lamberti et al. ||Milk||Cookie (180°C, 10 min)||10 mg /200 μl||Protein precipitation methanol/ chloroform||Pellet + 50 μl of 0.025M NH4HCO3 pH 8.0 RT||LC: HP 1100 HPLC (Agilent)||LOD: 1.3 mg of milk proteins per kg|
|Buffer: NH4HCO3 / (NH4)2CO3 +1 % SDS buffer, pH 8.2||3 μl of trypisn (75 ng/μl) 37°C, 90 min||Column: ACE C18 300A (250 mm × 1 mm)||LOQ: 4 mg of milk proteins per kg|
|Agitation: 20 min, 60°C||5 μl of 5% formic acid||MS: XCT-Plus Ion trap mass spectrometer (Agilent)|
|Pedreschi et al. ||Peanut||Cookie (180°C, 16 min)||Buffer: 20 mM TRIS - 150 mM NaCl, pH 7.4||GE Healthcare kit||50 μg of protein / 50 μl of Rapigest in a 50 mM ammonium bicarbonate buffer||LC: nano Acquity UPLC (Waters)||>10 mg of peanut per kg|
|Ultrasound: 4°C, 20 min||Cut-off filtration 3000 MWCO||2.5 μl of 50 mM DTT 30 min, 60°C||Column: nano Acquity BEH130 C18 1.7 μm (75 μm × 100 mm)|
|5 μl 100 mM IAA 30 min-dark||Column: nano Acquity UPLC Trap SymC18 5 μm (180 μm × 20 mm)|
|1 μl of 1μg/μm of trypsin 5h-37°C||MS: Q-Tof Ultima Global (Waters)|
|Huschek et al. ||Soy, sesame, lupin||Wheat, cookie (190°C, 13 min), bread (220°C, 30 min)||1 g||SPE cardridge (LiChrolut RP-18 Merck Millipore)||Alkylation: IAA 20 min at 50°C||LC: Nexera XR UHPLC (Shimadzu)||LOQ (S/N > 10)|
|Buffer: 100 mM NH4HCO3, 4M urea 5 mM DTT pH 8.2||Digestion: Trypsin||Column: Aeris Peptide XB-C18 (100 × 2.1 mm, 1.7 μm) (Phenomenex)||10–20 mg of soy per kg|
|formic acid 2%||10–50 mg of sesame per kg|
|Agitation 30 min RT||MS: Qtrap 5500 MS/MS (Sciex)||10–50 mg of lupine per kg|
|Planque et al. [33, 60]||Milk, egg, soy, peanut||Tomato sauce (95°C, 45 min), cookie (180°C, 18 min), ice cream, chocolate||2 g / 20 ml||Sep-Pack tC18 6cc (Waters)||10 ml extract + 10 ml NH4HCO3 (200 mM)||LC: UPLC Acquity (Waters)||LOQ (S/N > 10)|
|Buffer: 200 mM TRIS-HCl pH 9.2, 2M urea||Reduction: 1 ml DTT (400 mM), 45 min||Column: BEH130 (2.1 × 150 mm) (Waters)||0.5 mg of milk proteins by kg|
|Alkylation: 2 ml IAA (500 mM), 45 min in the dark||MS: Xevo TQS (Waters)||3.4 mg of egg proteins by kg|
|Agitation:30 min Ultrasound: 15 min||Digestion: Ratio protein:trypsin 1:20 16 h, 37°C||5 mg of soy proteins by kg|
|300 μl formic acid 20%||2.5 mg of peanut proteins by kg|
|Parker et al. ||Milk, egg, peanut||Muffin (177°C, 48 min)||Buffer: 2 M urea, 50 mM TRIS Ph 8.0, 25 mM DTT||Microcentrifuge tubes||Filter-aided sample preparation (FASP) sample cocentration and digestion protocol||LC: nano Acquity UPLC (Waters)||/|
|Amicon Ultra 0.5 ml||Reduction: 10 mM DTT||Column: nano Acquity BEH130 C18 1.7 μm (100 μm × 100 mm)|
|vortex: 5 min at 1400 rpm||Utracel-10 membrane||Alkylation:25 mM IAA||MS:6500 QTRAP (Sciex)|
|Ultrasound: 10 min at 4°C||Digestion: Ratio protein:trypsin 1:100 16 h -37°C|
|0.1% trifluoroacetic acid and 2% acetonitrile|
|Gomaa et al. ||Milk, soy, wheat||Cookie (177°C, 12 min)||1 g/9 ml||OMIX C18/ tip (Varian)||Protein extract: 100 μl at 2 mg/ml||LC: nano Acquity UPLC (Waters)||10 mg of casein per kg|
|Buffer: 50 mM NH4HCO3||Reduction: 1μl of DTT (0.5 M) 56°C, 20 min||Column: nano Acquity BEH130 C18 1.7 μm (75 μm × 100 mm)||10 mg of soy proteins per kg|
|Alkylation: 2.7 μl IAA (0.55 M, 15 min)||nano Acquity UPLC Trap SymC18 5 μm (180 μm × 20 mm)||100 mg of gluten per kg|
|Digestion: ratio enzyme:subtrate 1:20 3h-37°C||MS: Q-Tof Synapt MS (Waters)|
|5 μl trifluoroacetic acid (2.5%)|
Determining the sensitivity: The sensitivity of food allergen analysis can be determined on spiked samples (obtained by incorporating extracted proteins into a matrix after processing), fortified samples (obtained by incorporating raw ingredients into a matrix after processing) or processed samples (obtained by incorporating raw ingredients into a matrix before processing). For spiked and fortified samples (‘non-processed samples’), examples of the limit of quantification (LOQ) reached are 0.1 mg milk protein, 0.3 mg egg protein and 2 mg soy protein per kg cookies  and 0.1–1.3 mg tree nuts per kg biscuit . Although these studies demonstrate the sensitivity of mass spectrometry, the real challenge is to reach this sensitivity in thermally processed samples. Important improvements have been made over the last 5 years in the detection of allergens in processed samples. Recently, developed methods allow reaching an LOQ near or below the VITAL threshold (Table 1), e.g. 0.5 mg for milk protein, 3.4 mg egg protein, 5 mg soy protein and 2.5 mg peanut protein per kg incurred cookie (180°C, 18 min, with SPE purification) . In another study, the LOQs achieved were 30 mg egg (13.8 mg proteins), 20 mg milk (7.2 mg proteins), 19 mg soy (6.8 mg proteins), 20 mg hazelnut (3 mg proteins) and 40 mg peanut (10 mg proteins) per kg incurred cookie (200°C, 12 min, with SEC purification) .
As described above, the sensitivity reached for processed samples is lower than that obtained with spiked or fortified samples. The same applies to ELISAs, which can show up to 100-fold lower sensitivity when applied to processed food than when applied to raw food, as demonstrated by the poor performance of several ELISAs for egg detection in cookies after processing. In 2010, Dumont et al. showed that one ELISA kit was not even able to detect 1000 mg egg powder per kg baked cookie, and four others strongly underestimated the amount of egg in the samples (Figure 4 of Ref. ). While mass spectrometry and ELISAs show comparable sensitivities when applied to unbaked products, mass spectrometry seems to be the method of choice for the analysis of allergens in baked food products.
3. Quantifying food allergens
Detecting hidden allergens in food products is essential to protecting the food-allergic population. For full transparency of allergen labelling, laboratories should also be able to quantify allergens in order to help food manufacturers manage cross-contamination during food production . However, significant signal suppressions have been observed in various food matrices, and the level of suppression depends on the matrix considered. In one study, for example, high-protein-content food products showed greater suppression of the peptide signal than ones with a low protein content: the determined LOQ values were 20 mg skim milk powder per kg for high-protein foods and 5 mg skim milk powder per kg for low-protein foods . The food protein content is not the only parameter to be considered in relation to suppression of the peptide signal obtained by mass spectrometry: factors such as the type of process, the fat content and the presence of tannins also have an important influence on food allergen detection and must be taken into account.
While detecting allergens in various food products is difficult, quantifying them is even worse. In recent years, mass spectrometry techniques have been used for quantitation in proteomic analysis. Two approaches have emerged as the most relevant for food allergen quantification: label-free quantification and the use of stable-isotope-labelled peptides or proteins [70, 72, 73]. The two strategies are compared in Table 4 (target peptides, internal standards and calibration curves) and discussed in relation to the AOAC guideline 2016.002 method performance requirements for the quantification of allergens in food products, specifying a recovery between 60 and 120% and intra-day and inter-day coefficients of variation lower than 20 and 30%, respectively  (Table 5).
|Authors||Matrix||Allergen||Protein||Peptide||Mass spectrometer||Internal standard||Calibration curve|
|Careri et al. ||Rice crispy/chocolate snacks||Peanut||Ara h2||CCNELNEFENNQR||Q-TOF Micro (Waters)||No internal standard||Rice crispy/chocolate snacks were spiked with peanut proteins|
|Monaci et al. ||Fruit juices||Milk||α-lactalbumin||/||Ultima triple quadrupole mass spectrometer (Waters)||No internal standard||Fruit juices were spiked with milk proteins|
|Korte et al. ||Bread matrix, ice cream, chocolate, muesli with fruit and berry||Almond||Pru du 6.0101||GNLDFVQPPR||QTRAP 6500 (Sciex)||No internal standard||Matrices were spiked with allergen proteins|
|Pru du 6.0201||VQGQLDFVSPFSR|
|Hazelnut||Cor a 9||LNALEPTNR|
|Pistachio||Pis v 5||AMISPLAGSTSVLR|
|Walnut||Jug r 2||FFDQQEQR|
|Mattarozzi et al. ||Pasta, biscuit||Lupin||β-conglutin||IVEFQSKPNTLILPK||LTQ XL linear ion trap (Thermo)||No internal standard||Pasta and biscuits were fortified with lupin proteins|
|Zhang et al. ||Infant formulas and whey proteins||Milk||α-lactalbumin||VGINYWLAHK||Xevo TQ triple quadrupole (Waters)||KILDKVGINNYWLAHKALCSE||Matrices were spiked with synthetic peptide VGINYWLAHK|
|Posada-Ayala et al. ||Sauces and salty biscuit||Mustard||Sin a1||ACQQWLHK||6460 triple quadrupole (Agilent technologies)||Purified protein Sin a 1||Standard addition of mustard in sauces and salty biscuits|
|Authors||Matrix||Allergen||Protein||Peptide||Mass spectrometer||Internal standard||Calibration curve|
|Newsome and Scholl ||Cookie (180°C, 16 min)||Milk||α-s1 casein||HQGLPQEVLNENLLR||Hybrid triple-quadrupole 4000 QTRAP (AB Sciex)||HQGLPQEVLNENLLR[13C, 15N]||Cookies were spiked with isotope labelled peptides|
|Parker et al. ||Cereal bar (177°C, 30 min) Muffin (177°C, 48 min)||Egg||Lysozyme||NTDGSTDYGILQINSR||6500 Qtrap (AB Sciex)||Heavy isotope [13C, 15N] labelled peptides / labelled amino acid: R or K||Cereal and muffin were spiked with isotope labelled peptides|
|Huschek et al. ||Wheat, cookies (190°C, 13 min), soft bread (220°C-||Soy||Gly m6||VFDGELQEGR||QTRAP 5500 (Sciex)||VFDGELQEGR[13C6, 15N4]||Wheat and cookies were spiked with allergen proteins|
|Sesame||Ses i6||ISGAQPSLR||ISGAQPSLR[13C6, 15N4]|
|Lutter et al. (2011) ||Baby food soy-based formula, infant cereals, breakfast cereals, rince water||Milk||β-casein||AVPYPQR||6460 triple quadrupole (Agilent technologies)||AVPYPQR [13C6, 15N4]||0.1% formic acid were spiked with proteins|
|α-s2 casein||ALNEINQFYQK||ALNEINQFYQK[13C6, 15N2]|
|α-s2 casein||FALPQYLK||FALPQYLK[13C6, 15N2]|
|Monaci et al. ||White wine||Milk||α-s1 casein||FFVAPFPEVFGK||Extractive ESI Orbitrap (Thermo Electron)||FFV[15N]APFPEV[15N]FGK||White wine were spiked with milk and egg proteins|
|Yi-Shun et al. (2017) ||Beer, wine, chips, flour, cookies…||Gluten||α-glyadin||LWQIPEQSR||6490 triple quad (Agilent)||LWQIPEQSR[13C, 15N]||Matrices were spiked with gluten proteins|
|γ-Hordein||QQCCQQLANINEQSR||QQCCQQLANINEQSR [13C, 15N]|
|Ippoushi et al. ||Sweet cherry fruit||Cherry||Pru av2||TGCCAMSTDASGK||Xevo TQD Zspray ion source (Waters)||TGCCAMSTDASGK[13C6,15N2]||Sweet cherry fruit were spiekd with isotope labelled peptides|
|Rahman et al. (2012) ||/||Shrimp||Tropomyosine||SEEEVFGLQK||Micro mass Quattro Ultima (Waters)||SEEEV[D8]VFGLQK||Solvent were spiked with shrimp proteins|
|Chen et al. (2014) ||Baked food (170°C, 25 min)||Milk||β-casein||VLPVPQR||TOF-MS Synapt G2 HDMS (Waters)||VL[13C6, 15N]PV[13C5, 15N]PQK (IS1)||Solvent were spiked with milk proteins|
|QSVLSLSQSKVL[13C6, 15N]PV[13C5, 15N]PQKAVPYPQRQ (IS2)|
|Human β-casein (IS3)|
3.1. Label-free quantification
The label-free quantification strategy is based on comparing the peptide signal intensities of different samples (Table 4a). Three label-free quantification possibilities are described below.
External calibration: Monaci et al. used this approach to quantify milk proteins in fruit juice. Using a calibration curve obtained by spiking fruit juice with extracted milk proteins, they found recoveries between 68 and 79% . This strategy was also used to quantify peanut proteins in rice crispy/chocolate snacks . A significant suppression effect, ranging from 30 to 50%, was observed for the Ara h2 peptide signal, while suppression of the Ara h3/4 peptide signal was less than 10%. A more recent study by Mattarozzi et al. obtained recoveries between 95 and 118% for lupin β-conglutin peptide in spiked biscuits . Although less expensive than other approaches, this approach requires a calibration curve for each matrix.
Modified synthetic peptide approach: Zhang et al. introduced an internal standard peptide (KILDKVGINNYWLAHKALCSE) with an added asparagine residue (N) in the β-casein peptide VGINYWLAHK. They obtained recoveries between 98.8 and 100.6% . The use of an internal standard allows better recovery, but adding an amino acid can change the retention time and modify the ionization of target peptides.
Standard addition: This label-free quantification strategy consists in adding standards to the matrices. It was used by Posada-Ayala et al. for the quantification of commercial food products . This approach consists in adding different known quantities of extracted allergen proteins directly to the sample to be analysed before digestion and in quantifying the target allergens with the resulting calibration curve. The recovery was not specified, but this approach allows correcting at least for digestion and matrix effects. However, the theoretical level of contamination in the samples must be known in order to adapt the quantities of standards to be added.
3.2. Stable isotope labelling quantification
This strategy is based on the use of isotope-labelled (13C-, 15N-, D-labelled) peptides or proteins  (Table 4b). It is recommend to use a 6-Da mass difference with respect to the amino acid for doubly charged precursors and an 8–10-Da mass difference for triply charged precursors . Although more expensive than the strategies described above, this approach has the advantage that the unlabelled and isotope-labelled peptides show similar ionization and similar mass spectrometry response signals. For allergen quantification, three kinds of isotope-labelled standards exist : proteins , concatemers  (or long isotope-labelled peptides) and Aqua peptides [61, 71, 75, 84] (isotope-labelled peptides) (Figure 5).
Isotope-labelled proteins: The principle of this approach is to add a labelled protein to the sample before extraction. Newsome et al. studied the recovery of the milk allergen α-S1 casein in baked cookies using a labelled internal α-S1 casein, and obtained recoveries ranging from 60 to 80% . Use of an internal standard allows correcting for the matrix effect and for effects linked to different steps in the sample preparation protocol (protein extraction and enzymatic digestion). It thus allows accurate determination of the recovery and precision for processed samples. This ‘gold standard’ approach is really expensive, however, making its use unrealistic for the vast majority of routine laboratories.
Isotope-labelled peptides: The principle is to add labelled peptides to the sample after digestion and before the purification steps. This approach is less expensive than the use of isotope-labelled proteins, and synthetic labelled peptides can easily be commercially obtained. Huschek et al. used isotope-labelled peptides to quantify soy, lupin and sesame allergens . They determined the recovery of their method by spiking wheat, cookie and bread with the labelled peptides and obtained results between 69.4 and 112.9%. One could argue, however, that very similar matrices were used in this study (wheat-based products) and that this type of study should be extended to other matrices in order to validate the ability of the isotope-labelled peptide to correct for matrix effects.
Lutter et al. quantified milk proteins in baby food, infant cereals, breakfast cereals and rinsing water, using a calibration curve obtained by spiking 0.1% formic acid with milk protein. The estimated recovery rates were between 16 and 66%  Lutter et al. In this study, the isotope-labelled peptides were used to correct for effects related to different steps of the analysis. While using a single calibration curve can be useful in the routine laboratory context, the relatively low recoveries obtained in this study reveal the inability of an isotope-labelled peptide to correct for sample-preparation-related effects. We have compared the areas of milk, egg, peanut and soy peptide peaks for three matrices with and without isotope labelled peptides. Our results clearly show that an isotope-labelled peptide is able to correct for matrix effects but not for effects linked to the extraction and digestion steps  planque et al.
Isotope-labelled concatemers/long isotope-labelled peptides: The isotope-labelled concatemer used in this technique is a chimeric protein containing all the labelled target peptides. This internal standard is added to the sample before enzymatic digestion. The advantage of this method is that a single concatemer can contain peptides belonging to different proteins or allergens. This strategy has been used in proteomics, but it is not yet used for food allergen quantification . An emerging alternative to use of a concatemer is use of a so-called ‘long isotope-labelled peptide’. Chen et al. compared the use of three types of internal standard: human β-casein, isotope-labelled peptide VL [13C6, 15N] PV[13C5, 15N]PQK and a long isotope-labelled peptide QSVLSLSQSKVL[13C6,15N] PV[13C5,15N]PQKAVPYPQRQ . The long isotope-labelled peptide provided better recovery, due to correction for digestion-step-related effects. The recovery based on spiked materials was between 98.8 and 106.7%. In 2016, it was shown that long isotope-labelled peptides allow recoveries of 97.2–102.5% for α-lactalbumin and 99.5–100.3% for β-casein in the quantification of human milk . This strategy is a good compromise between isotope-labelled proteins and peptides. It allows correcting both for the matrix effect and for digestion-step effects, unlike the use of isotope-labelled peptides.
In conclusion, these studies show that using an isotope-labelled protein or a long isotope-labelled peptide provides better recovery than the isotope-labelled peptide approach. As explained below in the section devoted to result validation, the recovery must be determined with allergen-spiked samples and processed matrices in order to meet AOAC specifications. Published methods, however, do not always meet the AOAC requirements, even with spiked samples. For instance, Careri et al.  observed a suppression effect between 30 and 50% for the Ara h2 peptide signal, and Monaci et al.  obtained recoveries ranging from 68 to 79% for α-lactalbumin and β lactoglobulin. Altogether, these works show that internal standards are needed for the quantification of allergens in food matrices. Currently, furthermore, the use of a calibration curve for each type of sample is the best way to respect the AOAC guideline requiring a recovery between 60 and 120%.
Future studies should thus still be done to improve the quantification of allergens from a single calibration curve with a good recovery.
4. Validating food allergen methods
While mass spectrometry methods are increasingly sensitive, there remains room for improvement. Furthermore, there subsist obstacles to the harmonization of allergen detection methods in food laboratories . In April 2016, the AOAC SMPR 2016.002 guideline ‘Standard method requirements for the detection and quantification of selected food allergens’ was published. This guideline is the first to specify target limits for sensitivity and range of linearity, target matrices and reference materials for the analysis of allergens (egg, milk, peanut and hazelnut) in food matrices by mass spectrometry (Table 5).
|Analytical range, ppm||10–1000||10–1000||10–1000||10–1000|
To obtain comparable results among laboratories, it is crucial to adopt validation guidelines. The AOAC guideline, however, is not sufficiently detailed, and each laboratory tends to apply its own rules. In what follows, we compare this guideline with published methods in terms of sensitivity, range of linearity, recovery and precision.
Sensitivity: In the AOAC guideline, the method quantitation limit (MQL) is defined as MQL = average (blank) + 10 x s0 (blank). Laboratories, however, often use other strategies to determine the limit of quantification (LOQ), such as determining a signal-to-noise (S/N) ratio which should be higher than 10 [56, 60] or estimating an LOD and an LOQ as 3s/slope and 10s/slope, respectively, where s is the standard deviation of the blank signal (n = 10) . On the other hand, the sensitivity can differ from one matrix to another. For example, in a study where cookie, ice cream and sauce were spiked with 0.5 mg milk proteins per kg, the observed S/N ratio was 26 for the cookie matrix, 83 for ice cream and 228 for sauce . This also highlights the importance of a ‘fit-for-purpose’ description of an analytical method. Moreover, the sensitivities of developed methods should be determined on the same reference materials (MoniQa, LGC) to ensure (1) their capacity to reach the sensitivity set by the AOAC guideline and (2) an appropriate comparison of method performances.
Linearity: The range of linearity is set as 0.001 to 0.1% allergen contamination (10 mg to 1000 mg) of allergenic ingredients per kg) and thus does not always include the MQL (e. g. an MLQegg of 5 mg per kg). In the case of high-sensitivity methods, the coefficient of regression is determined using a lower range of concentrations [57, 71].
Recovery: Recovery must range from 60 to 120%. Such recovery values are hard to reach for the detection of allergens in processed samples, and recovery can only be determined by spiking food matrices with allergens. Focusing on egg, milk and peanut in spiked and incurred muffin and cereal bars, Parker et al. constructed calibration curves by spiking the matrices with allergen proteins . In the case of spiked muffin, the determined recovery was 98.6% for egg peptide (GGLEPINFQTAADQAR), 87.7% for milk peptide (YLGYLEQLLR) and 100.2% for peanut peptide (SPDIYNPQAGSLK). When the muffins were baked for 48 min at 177°C, the recoveries were dramatically lower: respectively 45.2%, 75.2% and 70.2%.
Inter- and intra-day coefficients of variation: According to AOAC SPMR, three unknown samples should be analyzed at least seven times to determine the reproducibility of the method. Lamberti et al. determined an intra-day coefficient of variation between 5 and 20% by performing three independent extractions at two different concentrations and three injections per extract .
Guidelines for the validation of mass-spectrometry-based methods for allergen analysis should be more precise, like the guidelines 2002/657/EC ‘Validation of residues in products of animal origin’ and SANCO/12574/2013 ‘Residues in products of animal origin method validation procedures for pesticide residues analysis in food and feed’. In SMPR 2016.002, several details are missing:
The number of target peptides that a method should include to confirm the presence of an allergen, as well as fragment ion number and/or type.
Criteria for the relative retention time, the ion ratio and the specificity of the method (blast, analysis of different blank and matrices), the level of spiking for determining precision and accuracy (for example, the LOQ, action limit and upper limit).
The major increase of the allergic population has prompted the development of numerous allergen detection methods. Over the past few years, improvements in the detection of allergens by mass spectrometry have been impressive, allowing detection of processed allergens with high sensitivity (a few mg of proteins per kg of food). Optimization of extraction and purification steps has notably played a key role in the improvement of analytical methods. Allergen quantification is performed mainly with labelled internal standards. The best approach involves the use of labelled proteins, allowing correction for effects occurring throughout the sample preparation protocol. The high cost of labelled proteins, however, has promoted the use of other strategies, such as methods based on long isotope-labelled peptides and standard addition of allergens.
The validation of qualitative and quantitative MS-based methods for routine detection of allergens is still very recent. The AOAC guideline is a first step towards the development of methods that will allow procedure harmonization, making it possible to compare results between laboratories. These methods should be both improved and extended to other allergens in order to demonstrate their validity and robustness.