Arabidopsis thaliana (ecotype Columbia 0) seeds are purchased from The Nottingham Arabidopsis Stock Centre (NASC) http://arabidopsis.info/. One hundred and thirty mg of Arabidopsis seeds were cultivated in Magenta boxes as previously described 14 . Magenta boxes were kept in the dark at 4°C during 48 h, and subsequently exposed to light for 4 h to synchronize germination. Finally, seedlings were grown in the dark at 23°C for 5 or 11 days. Seedling hypocotyls were cut just below the cotyledons and above the roots. Typically, 36 and 18 Magenta boxes were required for 5 and 11 day-old seedlings respectively.

Cell walls were prepared as previously described 14 . Proteins were extracted from the cell wall fraction in two successive steps, first with a CaCl₂ solution (5 mM sodium acetate buffer, pH 4.6, 0.2 M CaCl₂ and 10 μL protease inhibitor cocktail, Sigma), followed by two extractions with a LiCl solution (5 mM sodium acetate buffer, pH 4.6, 2 M LiCl and 10 μL protease inhibitor cocktail). Finally, proteins were desalted and lyophilized.

Lyophilized proteins were dissolved in a total volume of 2 mL of water. They were quantified with the Coomassie^® protein assay reagent kit (Pierce) using bovine serum albumin (BSA) as a standard 17 . One mg of proteins was used for chromatographic fractionation on a 1 mL HiTrap™ SP FF (Sepharose Fast Flow) column (Amersham Biosciences), equilibrated with 50 mM MES (pH 5.6) operated with an FPLC™ (Fast Protein Liquid Chromatography) System (Amersham Biosciences), controlled by the FPLCdirector™ version 1.0 (Amersham Biosciences). The protein solution was adjusted to 50 mM MES (pH 5.6), and 20 μL protease inhibitor cocktail (Sigma) were added before loading onto the column at a flow rate of 0.5 mL.min^-1. A 10 mL unfixed fraction was collected at the same rate. Three mL of first wash with 50 mM MES (pH 5.6) were collected at a flow rate of 1 mL.min^-1. Fixed proteins were eluted by a gradient from 0 M to 0.8 M NaCl in 50 mM MES (pH 5.6); 24 fractions (1 mL each) were collected at a flow rate of 1 mL.min^-1. Finally the column was successively washed with 3 mL of 1.2 M NaCl and 3 mL of 1.5 M NaCl in 50 mM MES (pH 5.6) at the same flow rate. These washes were also collected in 6 fractions (1 mL per tube). To prevent protein degradation, 2 μL of protease inhibitor cocktail (Sigma) were added to all the 1 mL fractions, 6 μL to the first wash and 20 μL to the unfixed fractions. The fractions were combined in twos and threes, depending on their protein concentration. Non-fixed proteins were concentrated by successive centrifugations at 4000 × g using the Centriprep^® centrifugal filter device (YM-10 kDa membrane for volumes greater than 6 mL or 5 kDa for smaller volumes, Millipore) at 4000 × g. All protein fractions were desalted prior to lyophilization using Econo-Pac^® 10DG columns (Bio-Rad) equilibrated with 0.2 M ammonium formate.

Each lyophilized fraction was dissolved in 200 μL water and separated by electrophoresis according to Laemmli 18 . Samples were loaded on 22 × 15 × 0.15 cm SDS-polyacrylamide gel with a concentration of 12.5% of acrylamide. The staining was carried out with a Coomassie Brilliant Blue (CBB)-based method 19 . Colored bands were excised from gels and digested with trypsin as described before 20 21 . Matrix-assisted laser desorption ionization – time of flight (MALDI-TOF) mass spectrometry (MS) analyses were performed as previously reported 20 21 . MALDI-TOF-TOF MS analysis was performed using a MALDI TOF-TOF Voyager 4700 (AppliedBiosystems/MDS Sciex, USA). N-terminal sequencing of the protein encoded by At5g14920 was performed at Plate-Forme d'Analyse et de Microséquençage des Protéines at the Institut Pasteur (Paris, France).

Peptide mass fingerprints were compared to the non-redundant database of Arabidopsis of NCBI http://prospector.ucsf.edu/html/instruct/allman.htm#database using ProteinProspector (MS-FIT: http://prospector.ucsf.edu/cgi-bin/msform.cgi?form=msfitstandard). A quantification index (QI) was calculated for each protein by adding the percentages of coverage, using peptide mass mapping (ratio between the number of amino acids in peptides detected by MS and the total number of amino acids of the protein) in all the bands of the FPLC profile in which the protein had been identified.

Sub-cellular localization, length of signal peptides, prediction of transmembrane domains, homologies to other proteins and protein functional domains were predicted as described before 15 . Glycoside hydrolases (GHs) and carbohydrate esterases (CEs) were respectively classified according to the CAZy database http://www.cazy.org 22 . Xyloglucan endotransglucosylase-hydrolases (XTHs) and expansins were respectively named according to http://labs.plantbio.cornell.edu/xth/ and http://www.bio.psu.edu/expansins/index.htm. Arabinogalactan proteins (AGPs) and fasciclin arabinogalactan proteins (FLAs) were named according to Schultz et al. 23 , Johnson et al. 24 and van Engels and Roberts 25 . Proteins homologous to COBRA, leucine-rich repeat extensins (LRXs) and Hyp/Pro-rich proteins were respectively annotated according to Roudier et al. 26 , Baumberger et al. 27 , and Fowler et al. 28 . Peroxidases were named according to the PeroxiBase http://peroxidase.isb-sib.ch/index.php 29 . Laccases were annotated as in Pourcel et al. 30 and McCaig et al. 31 . SKU-like proteins and phytocyanins were respectively named according to Jacobs and Roe 32 , and Nersissian and Shipp 33 . Proteases were annotated according to the MEROPS database http://merops.sanger.ac.uk/.

The aim of this proteomic study was to compare two developmental stages of Arabidopsis hypocotyls. This goal required about 40 g of homogenous material obtained in the following culture conditions: synchronized germination, high seedling density on the culture medium, and control of in vitro culture conditions. Five day-old elongating hypocotyls and 11 day-old fully-grown hypocotyls were compared 4 7 . First, the cell wall fraction was prepared to minimize intracellular contamination and loss of proteins 14 . To collect mg amounts of proteins for FPLC separation, two successive extractions were performed with 0.2 M CaCl₂ and 2 M LiCl 14 . Both extracts were combined and used for further analysis. Typically, about 1 mg of proteins was obtained from 1 g of dry cell wall fraction.

Finally, proteins were separated prior to their identification by peptide mass mapping using MALDI-TOF MS and bioinformatics. Since 2D-E is not appropriate for resolving CWPs 12 , these proteins were separated using mono-dimensional gel electrophoresis (1D-E). Approximately 60 bands were stained with Coomassie Brilliant Blue (CBB). Fifty two and 67 proteins were respectively identified in the extract from 5 and 11 day-old hypocotyls extracts (Additional data files 1, 2). Because of the limited resolution by 1D-E, many proteins, some of which have a low number of peptides, were identified in each band. In order to improve the separation and identification of proteins, we introduced an additional step prior to 1D-E. Since most CWPs are basic 16 , a cation exchange chromatography was performed using an FPLC device (Figures 1A, 2A). Fractions were collected and combined prior to separation by 1D-E (Figures 1B, 2B). At this point, approximately 500 bands were stained with CBB and were further analyzed by MALDI-TOF MS. The proportion of successful protein identification in stained bands was about 70%. Respectively 141 and 109 proteins were identified in 5 and 11 day-old hypocotyls. Many of the proteins were identified in several bands, thus reinforcing their identification. There was a great improvement in the quality of the analysis: (i) the number of identified proteins was doubled; (ii) the quality of the identifications was improved with higher numbers of peptides for identification of most proteins (Additional data files 3 and 4); (iii) the semi-quantification of proteins allowed the comparison of the two samples.

Combining results of 1D- and 2D-separation, 147 and 126 proteins were identified respectively in 5 and 11 day-old etiolated hypocotyls respectively (Table 1, Additional data file 5). On the one hand, bioinformatics prediction indicated that 120 (82%) and 101 (79%) proteins were secreted proteins in 5 and 11 day-old hypocotyls respectively. On the other hand, 27 and 25 proteins (in 5 and 11 day-old hypocotyls respectively) had no predicted signal peptide, and were considered as intracellular contaminants. Altogether, 173 proteins were identified in hypocotyls among which 137 (79%) were predicted to be secreted, indicating the good quality of cell wall preparations. In this article, these proteins will be called CWPs. Although many CWPs (84) were found in both samples, only 36 and 17 were identified only in 5 and 11 day-old hypocotyls respectively.

A second bioinformatics analysis of CWPs allowed their classification according to functional domains. For several protein families, experts' designation was used as described in Experimental procedures. The nine functional classes defined by Jamet et al. 12 were employed (Figure 3): proteins acting on carbohydrates, oxido-reductases, proteins with interaction domains, proteases, structural proteins, proteins involved in signaling, proteins related to lipid metabolism, proteins with miscellaneous functions, and proteins of unknown function. Some protein classes were more abundant at 5 days than at 11 days, e.g. proteins acting on carbohydrates (28 vs 21), proteins with interaction domains (25 vs 22), proteins related to lipid metabolism (7 vs 5), and proteases (14 vs 9) (Figure 3a). In each functional class, some CWPs were only identified at 5 or 11 days (Figure 3b). Differences appeared among proteins acting on carbohydrates (11 and 4 were found only at 5 or 11 days respectively), proteases (5 were only found at 5 days), proteins with domains of interactions with proteins or carbohydrates (3 were only found at 5 days), miscellaneous proteins and proteins of unknown function. In particular, the pattern of oxido-reductases appeared to be very different with 5 and 6 proteins being identified only at 5 and 11 days respectively. On the contrary, signaling and structural proteins showed minor changes. Due to their specific structural characteristics, they are not very present in either proteomes. Structural proteins are difficult to extract when they are cross-linked. They are also hard to identify because of numerous post-translational modifications (PTMs) 34 . Proteins involved in signaling, such as AGPs, may also have many PTMs 35 , and proteins with trans-membrane domains are not usually extracted with the protocol used 14 .

Finally, 51 proteins reported in this work were not identified in previous cell wall proteomic studies (Table 2). There are 11 CWPs that act on carbohydrates, 8 oxido-reductases, 5 proteases, 8 that carry interacting domains, 1 which is possibly involved in signaling, 1 structural protein, 4 proteins related to lipid metabolism, 5 proteins with diverse functions, and 7 without known function.

In the previous section, the comparison of the two physiological stages was based on the single criterion of presence/absence of a protein among proteins identified by MS. A more precise comparison would require the quantification of the proteins. However, for several reasons, CBB staining of the gels does not allow such quantification. Despite improvement in the separation of proteins by liquid chromatography followed by 1D-E, as compared to 1D-E alone, most proteins were found in several FPLC fractions and in several bands of the same FPLC fractions. This was probably due to PTMs, proteolytic maturation or degradation of proteins. Moreover, due to differences in ionization efficiency of diverse peptides, and variations related to competitive desorption of peptides at the time of ionization, MALDI-TOF MS analyses are not quantitative. We propose alternative ways to compare the proteins of both samples. In a first approach, FPLC fractions, in which proteins were identified, were counted. This was done in order to give a first criterion, based on the following rationale: an abundant protein is more difficult to resolve and will be distributed in more fractions than a rare protein. The calculations carried out on members of several gene families give an overview of the relative abundance of each protein (Figure 4). Two additional criteria were then considered with the aim of evaluating the relative amount of each protein at both stages of development: the number of bands per fraction, and the percentage of coverage of the amino acid sequence by peptide mass mapping. A correlation was observed between the abundance of a protein and the number of matching peptides expressed as a percentage of coverage (results not shown). For a given protein, adding up all values, a semi-quantitative index (QI) was obtained, which allows comparisons of the relative amount of each protein in the two samples (Additional data file 6).

Taking into account QIs, 63% of the proteins showed differences in the two developmental stages: 61 proteins (42%) are more abundant at 5 days and 26 (17%) at 11 days. Figure 3c presents the results ordered by functional classes. These results were fully consistent with those described above, which only take into account the presence of a protein in FPLC fractions (Figure 3b). However, the differences are now more important. For example, 17 out of 32 for proteins acting on carbohydrates are more abundant at 5 than at 11 days. The case is the same for proteins having domains of interactions, proteins related to lipid metabolism, and miscellaneous proteins.

For proteins acting on carbohydrates (Figure 4a), all the identified expansins were more represented at 5 than at 11 days. The situation was similar for PMEs with the exception of At3g43270 that was found in the same number of FPLC fractions at 5 and 11 days. On the contrary, XTHs and PGs presented a more complex pattern. Three XTHs were more abundant at 5 than at 11 days (AtXTH4, AtXTH5, and AtXTH31), whereas AtXTH33 was equally present at both stages, and AtXTH32 was found only at 11 days. Three PGs were found only at 5 days (At2g33160, At1g19170, and At4g18180), At3g15850 was equally represented at 5 and 11 days, and At3g06770 was only found at 11 days.

The distribution of oxido-reductases in the two stages looked as complicated as that of XTHs (Figure 4b). AtPrx36 and AtPrx57 were identified only at 5 days whereas AtPrx12, AtPrx43, and AtPrx72 were identified only at 11 days. AtPrx32 was more abundant at 5 days whereas AtPrx45 was more abundant at 11 days. Finally, AtPrx30, AtPrx34, and AtPrx69 were equally present at both stages. Among other proteins predicted to be oxido-reductases, berberine-bridge enzymes were found in the two stages of development.

Most proteins related to lipid metabolism were more abundant at 5 than at 11 days (Figure 4c). This is especially the case for proteins containing a GDSL-Lipase/Acylhydrolase domain (At1g29670, At1g54010, and At3g48460).

In the case of proteases (Figure 4d), the situation differed depending on the protease family. Three Asp proteases were found only at 5 days (At1g79720, At3g02740, and At3g52500), whereas 3 of them were more abundant at 11 than at 5 days (At3g54400, At5g10770, and At5g07030). At1g09750 was equally present at both stages. Three Cys proteases were more abundant at 11 than at 5 days (At1g47128, At5g43060, and At4g01610). Instead, a Ser protease was found only at 11 days (At4g30610). Finally, protease inhibitors (Figure 4d) were distributed in the two stages of development, with 4 of them equally present at both stages, 2 only found at 5 days, and 1 more represented at 11 days.