Cell wall proteins (CWPs) are important players in plant cell walls, otherwise mainly constituted of polysaccharides, and eventually of phenolic compounds around differentiated lignified cells (Carpita and Gibeaut, 1993). They have been involved in the remodeling of cell wall polymers networks by hydrolysing covalent bounds, inserting newly synthesized polysaccharides, cross-linking together structural proteins, proteins and polysaccharides or polysaccharides and phenolic compounds (Franková and Fry, 2013; Cosgrove, 2015). Together with extracellular peptides, some CWPs have also been involved in signaling, thus allowing cell-to-cell communication (Matsubayashi and Sakagami, 2006; Lamport and Várnai, 2013; Tavormina et al., 2015). Altogether, CWPs and peptides contribute to both developmental processes and response to environmental cues (Tenhaken, 2015; Borassi et al., 2016). Most of them undergo post-translational modifications (PTMs) during their transport through the secretion pathway (Faye et al., 2005; Kim and Brandizzi, 2016) which can modify their conformation, their biological activity and/or their ability to interact with cell wall components (Lannoo and Van Damme, 2015; Baer and Millar, 2016; Strasser, 2016). As an example, the site-directed mutagenesis of the N-glycosylation motifs of a class III peroxidase was shown to reduce its thermal stability, its catalytic activity and to modify its conformation (Lige et al., 2001).

During recent years, proteomics has facilitated a better knowledge of the plant cell wall proteome by increasing its coverage thanks to the design of specific strategies able to recover protein extracts enriched in extracellular proteins from organs of several model plants and crops (Lee et al., 2004; Albenne et al., 2013; Komatsu and Yanagawa, 2013; Rodríguez-Celma et al., 2016). Beyond the identification of proteins, technological advances have also permitted description of their PTMs. Various methods have been developed to address this particular question. In particular, immobilized affinity chromatography (IMAC) and lectin-affinity chromatography have allowed studying protein phosphorylation and glycosylation, respectively (Ytterberg and Jensen, 2010; Nakagami et al., 2012; Ruiz-May et al., 2014; Canut et al., 2016).

Glycosylation is one of the major PTMs of CWPs and refers to N-glycosylation, O-glycosylation and addition of glycophosphatidylinositol (GPI)-anchors, also named glypiation (Faye et al., 2005). Each of these PTMs occurs on specific amino acid sequences. N-glycosylation is the best described. It occurs on Asn residues in Asn-X-Ser/Thr motifs, where X cannot be a Pro residue. In these motifs, the hydroxyl functional group of Ser/Thr residues was shown to be required in the transglycosylation reaction on the Asn residue (Bause and Legler, 1981), whereas the presence of a Pro residue modifies the local conformation of the protein, thus preventing its N-glycosylation (Bause, 1983). The different structures of N-glycans are well-known, thus allowing systematic search in mass spectrometry (MS) data obtained in conditions preserving glycan-peptide bonds (Ruiz-May et al., 2012). GPI-anchors are transferred by a transamidase to a carboxy-terminal GPI-attachment signal peptide which can be predicted by bioinformatics (Eisenhaber et al., 2003). Several targeted proteomic studies have contributed to the identification of GPI-anchored proteins and some of them could be released from plasma membrane fractions by a phospholipases C or D which cleave GPI-anchors (Borner et al., 2003; Elortza et al., 2003; Elortza et al., 2005). O-glycosylation is the most complex type of glycosylation. In plant CWPs, it can occur on Ser and hydroxyproline (Hyp) residues (Faye et al., 2005). Galactose can be linked to Ser and Hyp residues whereas arabinose can only be linked to Hyp residues (Canut et al., 2016). According to the so-called Hyp contiguity hypothesis initially proposed for hydroxyproline-rich proteins (HRGPs), contiguous Hyp residues are arabinosylated and clustered non-contiguous Hyp residues are galactosylated (Shpak et al., 2001). Then, glycosyltransferases can extend the O-glycans in different ways depending on the initial pattern of Pro hydroxylation. The correct O-glycosylation of HRGPs was shown to be required for their conformation or their biological activity (Stafstrom and Staehelin, 1986; Velasquez et al., 2011).

Pro hydroxylation is a major step for O-glycosylation, but it is still difficult to predict in which amino acid context it occurs. In a previous review, we have proposed an extended Pro hydroxylation code (Canut et al., 2016), based on both (i) the initial Pro hydroxylation code (Kieliszewski and Lamport, 1994) and (ii) additional experimental LC-MS/MS and Edman sequencing data. The extended code has taken into account more protein and peptide families than the former one, including structural proteins like HRGPs and Pro/Hyp-rich proteins, solanaceous lectins, allergens, systemins and CLE peptides. Briefly, Pro residues could be hydroxylated when they are located after Ala, Gln, Hyp, Pro, Ser, Thr, and Val residues, whereas the first Pro residue following the other amino acids could not be hydroxylated (Figure 1). Only little information is available regarding Trp and Met residues. In a large mutagenesis screen performed on the amino acids surrounding the only Pro residue of sporamin shown to be hydroxylated, Trp and Met were not shown to favor the hydroxylation of the following Pro residue (Shimizu et al., 2005).

In this article, our aim was to test the extended Pro hydroxylation code on a new set of CWPs including non-structural CWPs. We have thus selected five CWPs with various contents in Pro residues. Three of them were rich in Pro residues among which AtPRP4 which is a structural CWP (Fowler et al., 1999) and two of them were poor in Pro residues. We have performed a deep data mining on two recent cell wall proteomic studies performed on rosettes and stems of Arabidopsis thaliana (Hervé et al., 2016; Duruflé et al., 2017). From the fine analysis of MS data, we have compared the observed patterns of Pro/Hyp location to the predicted ones according to the Pro hydroxylation extended code. The limits of the existing extended Pro hydroxylation code are discussed and new motifs are described.

This work has allowed mapping the Pro/Hyp residues in five CWPs and comparing the prediction of Pro/Hyp location according to a previously proposed extended Pro hydroxylation code (Kieliszewski and Lamport, 1994; Canut et al., 2016). Among these five proteins, only AtPRP4 was assumed to contain Hyp residues because it is known as a structural CWP with a high content of Pro residues and canonical amino acid motifs such as KKPCPP (7 occurrences) and PPV (14 occurrences) (Showalter et al., 2010). However, to our knowledge, its pattern of Pro hydroxylation has not yet been described. Regarding the other four proteins, our results have allowed enlarging the number of protein families possibly modified at the post-translational level by the hydroxylation of Pro residues. They have also shown that the Pro hydroxylation patterns can be variable at a given amino acid position.

The prediction of Hyp residue location could be done with a high level of confidence in one of the so-called HRGPs using the extended Pro hydroxylation code probably because this code was designed from the analysis of such protein sequences (Kieliszewski and Lamport, 1994; Canut et al., 2016). They include proteins rich in Pro, Ala, Ser and Thr residues such as (i) extensins with repetitive S(P)_{n ≥ 2} and YXY motifs, (ii) arabinogalactan proteins (AGPs) with AP/PA/SP/TP repeats and Pro-rich proteins (PRPs) with PPVX[KT], KKPCPP and PPV motifs (Showalter et al., 2010) or chimeric proteins containing a Pro-rich domain with XP_nY motifs (Hijazi et al., 2012; Canut et al., 2016). However, although prediction of the location of Pro/Hyp residues in the AtPRP4 sequence was very efficient (94.9% of successful predictions), some variability could be observed at a low frequency, particularly in the KPPPK motif, with 19 canonical Pro hydroxylation patterns (KPOOK) out of the 29 observed patterns and 10 variants (KPPOK, KOPPK, and KPPPK).

For the other proteins rich in Pro residues, many exceptions to the extended Pro hydroxylation code could be observed. In particular, in At1g31580, only about one third of the three amino acid-motifs had Hyp at the predicted location in RPI/T/R and VPG/I motifs and the FPPR motifs was systematically found with two Hyp residues (55/55). Besides, only one predicted VOOV motif could be recorded in At2g10940 out of the 140 observed peptides. The major variant was VPOV (133/140). A similar situation was found for the predicted VOK and VOV motifs (5/119 and 0/22 observations, respectively). For proteins having a low content in Pro residues, the prediction of Pro/Hyp location was also inefficient. Only a few Hyp residues could be observed and at a low frequency. Unexpected Hyp residues were also found in a previous study focused on class III peroxidases (Nguyen-Kim et al., 2016). For these proteins, a few Hyp residues were observed in CPN/Q/R, DPA, GPS/N, HPD, IPD, and LPA/Q/S motifs whereas some Pro residues were observed in APF/A, VPT, SPT/D, and TPG/L motifs. These results suggest that the extended Pro hydroxylation code cannot be used for such proteins. They also show that the rule established for the hydroxylation of the Pro residue within the EPA motif of sporamin cannot be applied (Shimizu et al., 2005). Based on mutagenesis of the surrounding amino acids, it was shown that the hydroxylation of the Pro residue required the following environment in tobacco BY-2 cells: [AVSTG]-P-[AVSTGA]-[GAVPSTC]-[APSDE].

Our results raise the question of the specificity of prolyl-4 hydroxylases (P4Hs) which have to recognize some features on the target protein at the level of its primary amino acid sequence or the secondary/tertiary structure. The specificity of three out of the 13 P4Hs of A. thaliana was characterized. P4H1 was shown to preferentially hydroxylate the second Pro residue in PPG motifs (Hieta and Myllyharju, 2002). All the peptides hydroxylated by P4H2 have at least three consecutive Pro residues and the third of them is preferentially hydroxylated (Tiainen et al., 2005). Finally, P4H5 was shown to hydroxylate Pro residues in SP₄ motifs in a sequential way, but never on the fourth Pro residue (Velasquez et al., 2015). Besides, P4H2 and P4H13 were assumed to complement the Pro hydroxylation pattern of SP₄ motifs in extensins (Velasquez et al., 2015). The characterization of additional P4Hs will give clues to understand this process which is probably tightly regulated because of its importance for biological activity. Indeed, this PTM is the first step prior to O-glycosylation: (i) poly-arabinosylation in extensins; or (ii) complex O-glycans like type II arabinogalactans (AGs), type III AGs or peanut agglutinin (PNA) AGs in AGPs, allergens or AtAGP31, respectively (Hijazi et al., 2014). None of the five CWPs analyzed in this work is known to be O-glycosylated. However, a previous proteomic study based on affinity chromatography with PNA, a lectin specific for galactose residues, has allowed identifying a protein of the same family as At3g08030 (Zhang et al., 2011). The next step will consist in correlating the presence of Hyp residues to O-glycosylation. Finally, the presence of Pro/Hyp peptide variants raises the question of the role of Hyp residues as previously discussed for class III peroxidases (Nguyen-Kim et al., 2016). This variability could be incidental or contribute to the regulation of the biological activity of CWPs.

Altogether, Pro hydroxylation events are probably more abundant in CWPs than initially thought, but the precise rules of this PTM need additional experiments to be fully described. Some clues can be proposed from our results. For example, a Hyp residue is found in the VPX motifs of AtPRP4 (77 Pro hydroxylations among the 82 observed VPX motifs) and At1g31580 (85 out of 110), whereas it is mainly a Pro residue in the three other proteins (only 15 Pro hydroxylations out of the 406 observed VPX motifs). The Pro residues in the 97 observed SPX motifs of At3g08030 were never hydroxylated whereas only a few lack of Pro hydroxylation have been described in cell wall Pro-rich proteins (Canut et al., 2016). A systematic mining of MS data is now required to permit the identification of other CWPs or secreted peptides containing Hyp residues and to map them. This task is challenging because, as mentioned above, 17.5% of the CWPs identified in one of our rosette experiments had at least one peptide carrying a Hyp residue. However, the amount of MS data corresponding to all these proteins was not sufficient to perform a relevant statistical analysis and to propose yet a further expanded Pro hydroxylation code. Finally, such a code should probably take into account tissue-specific patterns as for O-glycosylation (Estevez et al., 2006) and protein families. This work paves the way for a better description of Pro hydroxylation patterns in CWPs.

HD and VH performed the protein extractions from purified cell walls and contributed to the analyses of results. TB and MZ did the MS/MS analyses. CD and EJ initiated the research, designed the study and discussed the results. EJ coordinated the analysis of the results and the writing of the manuscript. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.