Wheat is arguably the most important grain in the world and forms a staple part of the modern diet (Shewry and Tatham, 2016), being present in many processed foods including breads, noodles, pasta, biscuits, cakes and sauces (Kamal et al., 2009; Gao et al., 2016). Its versatility as a food ingredient results from the unique physicochemical properties of the gluten fraction of wheat seed protein. One of the earliest proteins to be studied, gluten was first described by Beccari in 1728 (Bailey, 1941), and is readily isolated from wheat flour as a viscoelastic mass by making dough and then washing it with dilute salt solutions. Gluten comprises the major storage proteins of wheat grain, which are traditionally divided into two groups based on their solubility called gliadins and glutenins (Osborne, 1907). The gliadins comprise monomeric subunits which are soluble in alcohol-water mixtures and are further classified, based on their mobility on electrophoresis at low pH, into α-,γ-, and ω-gliadins. The glutenins comprise two groups of subunits, called high molecular weight (HMW) and low molecular weight (LMW) glutenin subunits (Bietz and Wall, 1973, 1980), which form alcohol-insoluble polymers stabilized by inter-chain disulfide bonds. However, when the disulfide bonds are reduced, the glutenin subunits become soluble in aqueous alcohol and amino acid sequences show that gliadins and glutenin subunits are related. A characteristic of both groups is that they contain few arginine and lysine residues and are rich in the amino acids proline and glutamine, which result from the presence of large domains comprising repetitive short peptide sequence motifs dominated by proline and glutamine. It is therefore usual to define both as prolamins, the name originally applied only to wheat gliadins and related proteins from other species.

Wheat is very diverse, with thousands of different types cultivated across the world, and over 40 cultivars currently being recommended for growth in the UK alone. The bread making quality of wheat is associated particularly with allelic variation in the HMW subunits of glutenin but these are not the sole determinants of quality. Many other factors are involved such as variation in gluten proteins other than HMW subunits, especially those that may modify polymer formation. However, these are generally poorly understood and make bread making quality of difficult to predict (Groger et al., 1997; Liu et al., 2016). Gel-based methods have long been used to identify variations in gluten protein composition associated with bread making quality (Dupont et al., 2011) whilst more recently methods employing MS alone in bottom-up approaches has been used for profiling gluten proteins (Mamone et al., 2011; Prandi et al., 2014; Colgrave et al., 2015; Manfredi et al., 2015; Wang et al., 2015; van den Broeck et al., 2015; Barro et al., 2016; Martínez-Esteso et al., 2016). One study using data independent acquisition (DIA) identified only a few gluten proteins as a consequence of employing only tryptic digestion (Uvackova et al., 2013). Data dependent acquisition (DDA) of classical Osborne fractions prepared from flour has also allowed identification of several hundred gluten peptides (Fiedler et al., 2014), whilst another study identified more than 80 wheat-specific proteins, including gluten, albumin and globulin proteins using a gluten-enriched fraction (Colgrave et al., 2015). A more recent analysis of a gluten food ingredient, coupled with fractionation to enrich gluten proteins also allowed identification of several hundred gluten-specific peptides (Martínez-Esteso et al., 2016). Due to the complexity of gluten, a large proportion of the current MS analysis carried out on gluten has been focused to profile HMW glutenin subunits initially (Lagrain et al., 2013; Wang et al., 2014).

Wheat variety Hereward, which is classified as a group 1 winter wheat, shows consistent baking and milling properties, and is well suited to bread making (Nabim Wheat Guide, 1978). However, this does not explain its good bread making quality as Hereward contains a combination of HMW subunits (called subunits Bx7+By9 and Dx3+Dy12) which are associated with poor quality (Shewry et al., 2012; Bekderok et al., 2013). Using Hereward as a model wheat we aim to comprehensively characterize the gluten proteome using a linear ion trap (LTQ) instrument in a DDA mode and a quadrupole time of flight (QTOF) instrument incorporating ion mobility in a DIA mode. A curated database containing only full length gluten protein sequences was used for annotation in order to overcome shortcomings of uncurated public repositories.

The effectiveness of different types of buffer on the extraction and detection of gluten proteins was initially assessed using several gluten-specific antibody preparations in order to identify a simple single step extraction method that was effective and compatible with subsequent MS workflows. Buffer E1, previously used for extraction of peanut proteins (Johnson et al., 2016) which primarily extracts the water-soluble (albumin) and salt-soluble (globulin) proteins of wheat, recovered 24.9% protein from the grain (Figure 1A) and revealed a complex mixture of polypeptides of 6–116 kDa (Figure 1B). Although immunoblotting with the mAb IFRN 0610 indicated that some gliadin and glutenin polypeptides had been extracted, only weakly reactive polypeptides were observed with the G12 and R5 antibodies (Figure 1C). In order to improve the extractability of the gliadin and glutenin fraction 75% (v/v) ethanol was added to buffer E1 to give buffer E2. It slightly reduced the extent of protein extraction to 19.8% but SDS-PAGE analysis revealed a pattern of polypeptides much more consistent with the pattern observed for gliadins and glutenins with the HMW subunits of cv Hereward (7 + 9, 3 + 12) being clearly visible. The E2 extract was strongly reactive with all three antibodies, although more polypeptides were recognized by the Mab IFRN0610 than either the R5 or G12 antibodies. This might be expected since the IFRN 0610 epitopes are present in many different seed storage prolamins from wheat (Brett et al., 1999; Kahlenberg et al., 2006; Mokarizadeh et al., 2015). Thus, the inclusion of reducing agent and ethanol improved the solubilization of gliadins and glutenins.

The two-step extraction which involved extraction of the albumin and globulins prior to that of the gliadins and glutenins gave a slightly higher rate of overall protein extraction but the variation in extraction efficiency between samples was increased (Figure 1A). This is probably due to the difficulties in undertaking the two-step extraction rather than inherent seed-to-seed variation in protein content because seeds of a similar size and therefore protein content were used for all extractions. This two-step extraction approach yielded a pattern of polypeptides following SDS-PAGE analysis very similar to that of the buffer E2 extract and showed the same pattern of reactivity with the different antibodies (Figure 1C).

The same extracts were then subjected to LC-MS analysis using two different platform technologies operating in different acquisition modes. The LTQ is a linear trap quadrupole made up of two mass analyser the first being an ion trap that acts as a mass filter, followed by a quadrupole were the filtered ions are separated according to the m/z as they pass along the central axis of four parallel rods in an electrostatic field. Whereas the QTOF incorporated ion mobility separates the ions as described previously (Allen et al., 2016) followed by additional separation in the time of flight tube that accelerates ions by an electric field, and the time taken to travel a known distance is recorded. The ion mobility adds an additional level of separation of the ions in the gas phase based on the shape and charge of the ions, (Allen et al., 2016) The LTQ was operated in DDA mode of acquisition, where MS/MS scans are only acquired on a subset of precursors detected in an MS “survey” scan adhering to a minimum threshold, and is biased to pick peptides of the highest intensity (Egertson et al., 2015). The QTOF was operated using a DIA mode of acquisition whereby all peptides are fragmented and MS/MS scans collected regardless of signal intensity (Doerr, 2015). The resulting spectral libraries were analyzed using an in-house curated wheat gluten protein sequence database. This allowed a total of 2,736 gluten peptides to be identified across all the extracts analyzed, of which 1548 were identified by the QTOF and 1031 by the LTQ, with only 157 gluten peptides matched by both platforms. Subsequently these data were used to identify proteins on the basis that the same unique peptide(s) were observed in at least two replicate analyses of each extract from at least two seeds. Individual peptide scores ranged from 5 to 70.78 showing the confidence in the protein identifications (Supplementary Data Sheets 2, 3). A total of 127 protein accessions were identified on the basis of a single unique peptide, which was decreased to 63 if three unique peptides were required for a positive identification (Figures 2A, 3A), a list of the resulting 63 protein accessions with corresponding number of unique peptides and normalized relative abundance can be found (Supplementary Data Sheet 1). A phylogenetic tree of the gliadin (Figure 2A) and glutenin (Figure 2B) sequences from the database was used to visualize the protein identifications inferred from analysis of all the different extracts (E1–3) as a function of platform technology and, for the gliadins only, the number of unique peptides. Some protein accessions were identified by both mass spectrometry platforms and sometimes closely related, but distinct isoforms, were identified by different platforms. When three unique peptides were required to make an identification, a total of 19 gliadin species were identified which were distributed between the two main branches of the gliadins. Thus, four of the 12 ω-gliadin sequences were identified on one branch and 22 (14 α-gliadin and eight γ-gliadin species) out of a total of 340 α-gliadin and γ-gliadin protein sequences from the second. The LMW subunits form the largest class of glutenin sequences in the database, with 224 accessions, the HMW subunits being less abundant with only 57 sequences (Figure 2B). A total of 29 LMW subunit and eight HMW subunit sequences were identified, the majority of which were identified only by the QTOF platform. A total of 29 LMW glutenin and eight HMW glutenin sequences were identified, the majority of which were made only by the QTOF platform. Even though extract E1 contained little immunoreactive gluten (Figure 1) the majority (94%) of gluten proteins identified with at least three unique peptides were observed in all three extracts (E1–E3; Figures 2A,B, 3B; Data Sheets 1–4) with the majority of proteins identified in E1 having a lower relative normalized abundance compared to proteins identified in the two other extractions (Figure 3C). Differences in identifications made between extracts were limited to one additional LMW subunit of glutenin present in extracts E1 and E3 with a further LMW subunit of glutenin and an α-gliadin being identified in extracts E2 and E3 (Figure 3B). The QTOF and LTQ showed >75% and >82% commonality for all peptides identified across all biological and technical replicates with each extraction.

Since cv Hereward is known to contain only four HMW subunits (Dx3 + Dy12 and Bx7 + By9) (Belderok et al., 2000; Cunsolo et al., 2004) the identification of eight (three unique peptides criteria) and 12 (one unique peptide criteria) HMW subunit accessions was unexpected. These identifications were further investigated by inspecting the sequence identity for the available relevant HMW subunit accessions. Subunit Dx3, which is a rare allele and is yet to be sequenced. However, it usually considered to be closely related to HMW subunit Dx2, having a very similar molecular mass determined by ESI-MS (Bellil et al., 2012) and being associated with the Dy12 allele. Dx2 was therefore used as a surrogate for Dx3. A table of the eight identified HMW glutenin sequences are shown in Supplementary Material, Table S1. The eight identified sequences were compared to the two expected HMW subunit sequences (Uniprot accession P08489 and Uniprot accession Q42451) using sequence identity calculated using Clustal Omega (McWilliam et al., 2013). The sequences comprised two sequences closely related to Dx2 (Uniprot accession P08489) and five to By7 (Uniprot accession Q42451) with respective average percentage sequence identities of 81.5 and 95.3% (Supplementary Material, Table S1). The eight identified HMW glutenin subunits identified using a minimum of 3 unique peptide criteria were mapped onto a phylogenetic tree of the HMW glutenins to show the homologous relationship between the observed proteins (Supplementary Material, Figure S2). When the closely related sequences were further interrogated it was shown that these sequences belonged to subunits not expected to be present in wheat cv Hereward. An example of this is protein D0IQ05, a well annotated Dx5 sequence that shows a sequence identity to Dx2 subunit (Uniprot accession P08489) of 94%. Following in silico chymotryptic digestion of the two proteins corresponding to the Dx2 and Dx5 subunits (Uniprot accession P08489 and Uniprot accession D0IQ05) the 16 and 11 unique peptides were mapped onto the protein sequence (Supplementary Material, Figure S2). This shows unique peptides with either single amino acid mutations shown in blue or insert regions shown in purple for both proteins. The list of theoretical unique peptides for both proteins were searched through the global proteomics data, and of the 16 and 11 unique peptides for the Dx2 and Dx5 subunits nine and seven were identified in the MS data. The first unique peptides identified in both proteins after the start of the repetitive domain (as indicted by an arrow, Figure S3) varied by only one amino acid and were indicative of the different HMW subunits, with the serine present in the Dx2 protein being replaced by a cysteine in the Dx5 proteins. This change is a result of a single change in the second base of the genetic code. These two unique characteristic peptides were searched in the global proteomics data collected, and the partial MS/MS spectra for both are shown in Figures S4, S5 confirming the presence of both peptides and the identification of both HMW subunits. When the stringency requirements for identification were relaxed to a single unique peptide, the number of HMW subunits was increased to five homolog of Dx2 and seven homolog of Bx7, together with two homolog of Dy12 and three for By9. It seems likely that the identification of multiple homologs results from the close sequence similarity of the different isoforms and the fact that the HMW subunits of cv Hereward have yet to be sequenced and may have a combination of the single amino acid polymorphisms represented by the accessions identified.

Long peptides were identified by both platforms; the longest identified by the LTQ being 33 residues in length whilst the QTOF identified one peptide 53 residues in length, an exemplar partial MS/MS spectra for a 50 residue peptide identified using the QTOF is shown in Figure S7. None of the 33 residue peptides identified corresponded to coeliac toxic 33-mer (Mokarizadeh et al., 2015) to which the G12 antibody was raised. This was expected as the 33-mer is derived from simulated gastrointestinal digestion and is not flanked by chymotryptic cleavage sites. Instead five of the 33mer peptides were from γ-gliadins, four from LMW glutenin subunits and 16 were from HMW glutenins. The QTOF DIA approach did identify a greater number of longer peptides (79% of peptides comprising 10–24 residues compared to 63% with the LTQ; Figure 4A), many of which represented missed cleavages, with the shorter peptides identified on the LTQ often being derived from the longer peptides identified by the QTOF. The shorter peptides have a lower probability of spanning a variant amino acid position in a given protein sequence, reducing their capacity to support identification of individual isoforms of different gluten proteins compared to the longer peptides.

In the data gathered using the QTOF, a significant increase in the number of peptides identified was observed for peptides of 21 residues in length, with a peak of 106 peptides that does not fit the general distribution curve observed. The dynamic range of intensity for these peptide peaks was within the range observed for all the peptides identified suggesting they did not ionize any differently to the other peptides and may simply reflect the distribution of chymotryptic cleavage sites (Figure S6). A full list of the 21 residue peptides observed can be found in Supplementary Material, Data Sheet 4, with the corresponding protein accession, peptide sequence and modifications. Whilst the majority of peptides identified by both platforms were unmodified (Figures 4C1,C2) deamidation was observed in 40 and 37% of peptides identified respectively by the QTOF and LTQ platforms (Figures 4B,C). In general, deamidation was observed irrespective of peptide length apart from 21 residue peptides, 75% of which were deamidated (Figure 4B). In contrast, oxidation of methionine, which commonly occurs during sample preparation, was identified in ~3% of peptides identified by both platforms. In order to ascertain whether the deamidation was extensive, the relative normalized abundance of the unmodified and modified peptides were defined (Figure 5A). These data indicate that the deamidation was a widespread modification that spanned the range of abundancies observed.

The combined data from the QTOF and LTQ platforms enabled vastly improved sequence coverage to be obtained for many protein identifications made, as illustrated in Figure 6 for the LMW glutenin protein, B2Y2R5. The N-terminal sequences of the LMW glutenins are distinctive and have been used to classify this gluten sub-group (Shewry et al., 1986). B2Y2R5 should classically have an N-terminal sequence starting at residue 21 with the sequence ²¹MENSHIPGL²⁹, distinctive of a LMW-s glutenin subunit (Shewry et al., 1986). However, no peptide corresponding to this N-terminal sequence was identified; however due to the extensive sequence coverage and four peptides unique to that protein being observed on both platforms it can be confidentially identified. Deamidated residues (highlighted in black in Figure 6) were observed at various positions across the protein, and sometimes at multiple positions in the same peptide. However, in some instances, where there are runs of glutamines, the exact location of the deamidation is ambiguous, and the same m/z would be observed for the same peptide with the modification at multiple locations, this is highlighted in Figure 6 when black residues are underlined.

The highly complementary repertoire of peptides identified by the different platforms used in this study enabled a greater number of gluten specific peptides and protein accessions to be identified than have previously been reported using only MS methodology (Mamone et al., 2011; Uvackova et al., 2013; Fiedler et al., 2014; Prandi et al., 2014; Colgrave et al., 2015; Manfredi et al., 2015; Wang et al., 2015; van den Broeck et al., 2015; Barro et al., 2016; Martínez-Esteso et al., 2016) and is comparable with those identified using gel-based separation techniques (Dupont et al., 2011). This approach also increased the sequence coverage that could be achieved using data collected using only one platform, giving greater assurance of the “trueness” of protein identifications made. It has been previously shown that utilizing complementary instruments can dramatically improve the proteome coverage (Elias et al., 2005) and the low level of overlap observed at the peptide level can be explained by fundamental differences between the platforms such as ionization source. Additionally data processing for the QTOF excluded some shorter peptides for the purpose of sequence coverage comparison only. Furthermore, the QTOF platform employed ion mobility, a method that has previously been shown to enhance coverage of the serum proteome by 85% (Daly et al., 2014). Lastly, the different modes of data acquisition used also play a role since in DDA only peptides with signals that rise above the noise in a full-scan MS spectrum are selected for fragmentation (Doerr, 2015), which may represent as little as 16% of the total peptides in a sample (Michalski et al., 2011). In contrast in DIA the full spectrum is acquired and peptides are not fragmenting peptides based on predefined thresholds (Chapman et al., 2014).

More detailed comparison with previous untargeted MS profiling studies is difficult because they have either been undertaken with the aim of identifying peptide targets for detection of gluten (Colgrave et al., 2015; van den Broeck et al., 2015; Martínez-Esteso et al., 2016), or used only tryptic digestion and consequently identified only a few gluten protein accessions (Uvackova et al., 2013). A further complication is that putative identifications have also often been made using protein sequence accessions from other, closely related species such as Triticum diccocum and Aegilops squarrosa (Martínez-Esteso et al., 2016). Surprisingly, almost exactly the same proteins were profiled in extracts prepared using reducing agent and 75% (v/v) ethanol (E2 and E3) as were identified in an extract containing little immunoreactive gluten. Thus, unexpectedly, sample preparation was not the limiting factor in the capacity to comprehensively profile the gluten proteins. This highlights the highly sensitive nature of the mass spectrometry platforms able to detect proteins that are not optimized for extraction.

Widespread deamidation was observed in all the different types of gluten proteins identified, and observation also made by Martínez-Esteso et al., (2016) in their analysis of a gluten food ingredient. This may be a methodological artifact as, although glutamine is usually much less susceptible to deamidation than asparagine under neutral and alkaline conditions (Robinson, 2002) acidic conditions appear to favor the deamidation of glutamine residues (Joshi and Kirsch, 2002), even though the only acidic environment experienced by the proteins for a prolonged period of time is exposure to dilute formic acid solutions. It is also possible that deamidation occurs naturally in the plant, since trains of spots of the same Mr and different pI are frequently observed in 2D PAGE profiles of gluten. Although these are often attributed to sample preparation procedures (Johnson et al., 2016) they are present in protein from freshly isolated protein bodies which has not been exposed to extreme conditions before analysis (Field et al., 1982). Further studies modifying the extraction methods could be used to identify whether the deamidation of gluten is an artifact of the analysis or a genuine, previously unidentified, post-translational modification or an artifact of the extraction and analytical procedures. Anomalous N-terminal proteolytic processing of a LWM subunit was also observed, with a long peptide identified which spans the signal peptidase cleavage site to give the well-defined N-terminal consensus motif METSRV. Ragged N-terminal processing is not uncommon in plant seed proteins and has been described in purified plant proteins, such as the 2S albumins from Brazil nut (Moreno et al., 2004) and there is some variation in further proteolytic processing of gluten protein's by a putative vascular asparaginyl endoprotease (Egidi et al., 2014).

The low content of arginine and lysine residues in gluten proteins means that protease digestion must be carried out using less-commonly used and less predictable enzymes such as chymotrypsin. However, the distribution of chymotryptic cleavage sites in many proteins means this enzyme typically generates longer peptides than are generally encountered when digesting proteins with trypsin. These longer peptides may present a challenge for data acquisition using MS. The capacity to sequence longer peptides in the current study, especially by the QTOF platform (often resulting from missed cleavages) also enabled greater sequence coverage and supported greater specificity of protein identification and the individual isoform level. This was also observed when profiling peanut allergens (Johnson et al., 2016) and hence whilst there have been many concerns about efficiency of protease digestion steps, which maybe crucial for effective quantification, for identification purposes missed cleavages may enhance levels of identification when using platforms able to sequence longer peptides.

The HMW subunit Dx3 is usually considered to be a “mutant” form closely related to Dx2, due to the similar mobility observed on the SDS-PAGE. However, this has not been confirmed by direct analysis at the gene or protein level. The identification here of sequences corresponding to the established sequences of HMW subunits Dx2 and Dx5 therefore casts doubt on this assumption and suggests subunit Dx3 may in fact contain sequences similar to both the Dx2 and Dx5. This is supported by ESIMS analysis of the Dx2 and Dx3 subunits that showed signification differences in their Mr (Lagrain et al., 2013). Furthermore, when subunits Dx2, Dx3, and Dx5 are shown on SDS-PAGE the bands are not distinguishable (Lagrain et al., 2012), giving rise to the need to further clarify the Dx3 subunit through sequencing this protein. The presence of these sequences attributed to the unexpected HMW glutenin could go some way to explain why Hereward has unexpected good bread making quality.

Our data show that comprehensive proteomic profiling of plant proteins, such as gluten, is not limited by the proteomics methodology, but by access to appropriate genetic data in a form usable for proteomic informatics pipelines which can handle highly polymorphic proteins with regards amino acid substitutions, which have both repeating sequences and deletions (Vensel et al., 2011; Kasarda et al., 2013; Romero-Rodríguez et al., 2014). Such integrated approaches will be required to unravel why certain wheat cultivars, such as Hereward, have better than expected bread making quality and hence identify novel targets for crop improvement. These data will also facilitate the development of future approaches well-founded targeted MS methods of analysis of gluten in foods.

SB, LG, JL, PS, MJB, MB, and EM all gave substantial contributions to the concept and design of the work, and where involved in the acquisition, analysis and interpretation of data at varying points. All authors were involved in the drafting of the manuscript, and gave approval prior to submission. All authors acknowledge accountability for all aspects of the work.