An atlas of Arabidopsis protein S-Acylation reveals its widespread role in plant cell organisation and function

Abstract

S-acylation is the addition of a fatty acid to a cysteine residue of a protein. While this modification may profoundly alter protein behaviour, its effects on the function of plant proteins remains poorly characterised, largely as a result of the lack of basic information regarding which proteins are S-acylated and where in the proteins the modification occurs. In order to address this gap in our knowledge, we have used an optimised Acyl-RAC assay to perform a comprehensive analysis of plant protein S-acylation from 6 separate tissues. In our high and medium confidence groups, we identified 1849 cysteines modified by S-acylation, which were located in 1640 unique peptides from 1094 different proteins. This represents around 6% of the detectable Arabidopsis proteome and suggests an important role for S-acylation in many essential cellular functions including trafficking, signalling and metabolism. To illustrate the potential of this dataset, we focus on cellulose synthesis and confirm for the first time the S-acylation of a number of proteins known to be involved in cellulose synthesis and trafficking of the cellulose synthase complex. In the secondary cell walls, cellulose synthesis requires three different catalytic subunits (CESA4, CESA7 and CESA8) that all exhibit striking sequence similarity and are all predicted to possess a RING-type zinc finger at their N-terminus composed of 8 cysteines. For CESA8, we find evidence for S-acylation of these cysteines that is incompatible with any role in coordinating metal ions. We show that while CESA7 may possess a RING type domain, the same region of CESA8 appears to have evolved a very different structure. Together, the data suggests this study represents an atlas of S-acylation in Arabidopsis that will facilitate the broader study of this elusive post-translational modification in plants as well as demonstrating the importance of undertaking further work in this area.

Introduction

S-Acylation, also known as palmitoylation, is a reversible post-translational modification involving the transfer of a fatty acid group, frequently stearate or palmitate, to a cysteine residue in target proteins. The very hydrophobic nature of the acyl group can dramatically alter the protein properties. In proteins that lack transmembrane helices (TMHs), the addition of an acyl group will confer membrane localisation, while for proteins that already possess TMHs, addition of an acyl group has a variety of effects including altering their subcellular localisation, partitioning into membrane microdomains and enabling the formation of multi-protein complexes ^1,2. The extent of S-acylation in eukaryotic cells has been revealed by studies on yeast and mammalian cells. An estimate based on the proteins in SwissPalm, the recently established database of protein palmitoylation, suggests that 8% of mammalian proteins are likely to be modified by S-acylation and even this is likely to be an underestimate ³.

The acyl group is added by a family of protein acyl transferases (PATs) that are characterised by a DHHC motif within their catalytic domain. In Arabidopsis, 24 different PATs have been identified that are localised in the Golgi, ER and plasma membrane ⁴. The structure of a human PAT, DHHC17, has recently been solved ⁵, despite this and the identification of a large number of S-acylation sites, there are no recognisable motifs around the modified cysteines of the substrate proteins. A recent study that monitored S-acylation directly using mass spectrometry (MS) suggested that cysteines were modified independently of any sequence motif around the site, in a stochastic process that depends upon the accessibility of any given cysteine to the action of a PAT ⁶. Consequently, making accurate bioinformatics prediction of S-acylation sites remains very challenging and knowledge of the extent of S-acylation is dependent on empirical identification of S-acylation sites.

The plant “acylome” has been partially described by two proteomic studies from Arabidopsis seedlings ⁷ and poplar suspension cultures ⁸ that identified a total of 694 and 449 proteins, respectively, with various levels of confidence. Neither study identified the sites of the acyl group attachment and as a consequence acyl-sites have been identified for only a relatively small number of individual plant proteins. The lack of information about acyl-sites of plant proteins is particularly acute in comparison to the information available from mammalian systems ^9–12, however the importance of S-acylation has been demonstrated for a small number of proteins ^2,13–15. These proteins include heterotrimeric G proteins, small G proteins, proteins involved in calcium signalling, pathogenesis, transcription and cellulose synthesis ^2,13. Functional analysis on the role of S-acylation in the absence of information about the sites of modification is challenging. A study of the catalytic subunit of the plant cellulose synthase complex, AtCESA7, identified 6 S-acylated cysteines organised in two clusters ¹⁶. This work required the use of systematic mutagenesis, which for a protein with 26 cysteines, is a laborious approach and does not guarantee the identification of S-acylation sites. While the limited knowledge of acylation sites has hampered studies on the role of S-acylation in the function of plant proteins, analysis of the individual Arabidopsis PATs has implicated S-acylation in a variety of processes including root hair formation, cell death, ROS production and branching, cell expansion and division, gametogenesis and salt tolerance ^2,13–15,17. Consequently, while our understanding of both the extent and function of S-acylation of individual plant proteins remains limited, the available data indicates S-acylation is important for many aspects of plant cell function.

To meet the need for more information on both the true extent of S-acylation of plant proteins and the sites at which the modifications occur, using a series of rigorous controls, we have developed an Acyl-RAC assay that is optimised to identify sites of S-acylation in plants. Using this method, we have undertaken a comprehensive analysis of S-acylation from 6 Arabidopsis tissues. Within our high and medium confidence groups, we identify evidence for a total of 1849 cysteines modified by S-acylation which were located in 1640 unique peptides from 1094 different proteins. We verify the accuracy of the data and demonstrate its utility by focussing on cellulose biosynthesis and identify that S-acylation appears to be a common feature of proteins involved in cellulose biosynthesis. Furthermore, we demonstrate how functional analysis of this data provides new insights into protein structure. In particular, the function of the “RING finger” domain has diverged between different CESA isoforms, a finding that alters our understanding of how this domain functions during cellulose synthesis. Together, this study represents an important milestone in the study of S-acylation of plant proteins, demonstrates the extent of S-acylation of plant proteins and provides a unique dataset that may be used as a resource for the functional analysis of S-acylation of plant proteins.

Results and discussion

The identification of S-acylated proteins was performed using the Acyl-RAC assay ¹⁸ that relies on capture of S-acylated peptides on beads in a hydroxylamine dependent manner. We optimised the protocol by first performing a TCA Acetone precipitation on the frozen tissue powder to remove plant secondary metabolites ¹⁹ that otherwise have the potential to interfere with cysteine chemistry. Furthermore, we have previously shown that during membrane preparation and solubilisation with non-ionic detergents, a large fraction of CESA proteins remains with the cell wall debris ²⁰ suggesting that this approach may bias the composition of the starting material. Consequently, we solubilised the precipitated tissue pellet directly in a buffer containing 2.5% SDS.

Iodoacetamide and acrylamide were used to block free cysteines and the cysteines that were freed following hydroxylamine treatment, respectively, enabling us to overcome problems detecting cysteines residues modified by NEM. After blocking free cysteines, we captured S-acylated proteins by hydroxylamine (HA) dependent binding to thiol reactive beads. Following stringent washes, we performed on-bead trypsin digestion of the S-acylated proteins. The peptides containing acyl cysteines remained bound to the beads and were released in a DTT dependent manner (Figure S1). Previously acylated cysteines freed by the hydroxylamine treatment were alkylated with acrylamide prior to analysis by mass spectrometry.

Next, in order to calculate the rate at which false positives may arise, stem tissue was spiked with E coli extract. We obtained a measure of our detectable proteome by omitting the iodoacetamide blocking of free cysteines to allow the identification of all detectable cysteine peptides from the stem sample. We assayed this sample in triplicate both in the presence and absence of HA. A volcano plot of peptides from HA plus fraction versus the HA minus (Figure 1A) shows a symmetrical distribution indicating HA independent binding of these peptides to the beads. These sample were processed as part of our analysis (see below). We detected a total of 197 E coli proteins in the unblocked sample (Table S1), while only 1 protein was identified to be S-acylated in our high confidence group and a further 3 in the medium confidence group. As E coli lacks the biological machinery for S-acylation, these data can be used to calculate an estimate of the false positives being identified in the data. This translates to a false positive rate of 1 out of 121 (less than 1%) in our high confidence group, and a total of 4 (˜3%) if we also include the medium confidence group. It possible that even this low figure is an overestimate of the rate of false positives as the E. coli extraction was generated prior from a conventional cell lysis prior. The E coli extract was not treated with TCA until it was spiked into the plant material and so we cannot be sure that the cysteine chemistry was not altered prior to TCA addition, something that likely contributes to generating false positives. To our knowledge, this is the first study to report the use of such comprehensive controls to verify the methodology.

Figure 1. Overview of the Arabidopsis Acylome.

A) Fold enrichment (log2) and P values (log10) derived from peptide intensities of stem sample treated with HA compared to untreated controls. Charts are shown for unblocked control and iodoacetamide blocked test sample. The shaded areas of the plot for blocked sample indicate the way in which the peptides were classified into high (light green shading), medium (light blue shading) low (light yellow shading) confidence groups. Peptides in the unshaded area were not considered further (see text for details). Data was derived from analysis of three biological replicates. P values were calculated using 1-tailed T-test. B) PCA analysis of peptide data from extracts of hypocotyl (red circles), leaf (green circles), meristem (pale blue circle), seedling (yellow square), silique (pale green squares) and stem (dark blue squares). Extracts were treated with (solid) or without (open) hydroxylamine. (C) Distribution of the number of S-acylated cysteines per protein among those proteins from the high and medium confidence group.

S-acylated proteins from the blocked sample were placed into various confidence groups based on protein enrichment in the HA treated sample. A simplified flowchart illustrating various data processing steps is shown in Figure S2. We calculated the fold enrichment and P value for individual peptides by comparing the intensities in the HA treated samples with those of control samples. When the fold enrichment was plotted against P value, the bulk of the peptides formed a distinctive group that exhibited both a high fold enrichment and low P value (Figure 1A, Figure S3A). We defined this high confidence group as peptides that exhibited a fold enrichment of greater than 1,000 fold (log2 = 9.97) and a P value of less than 0.01 (Figure 1A, Table 1). We also defined a medium confidence group as those peptides that either exhibited a fold enrichment of more than 1000, but a P value between 0.1 and 0.01 or those a P value of less than 0.01, but a fold enrichment of between 10 and 1000. While the low confidence group comprised peptides with a fold enrichment of between 10 and 1000, but a P value of less than 0.10 (Figure 1A, Table 1, Figure S3A). We were able to plot a similar distribution for protein intensities (Figure S3B) and in most cases, the protein score matched to the score of the best peptide mapped to that protein. The final protein grouping was assigned by comparing the protein score with the score of the best peptide and taking the lower of the two (Figure S2). Therefore, a protein in the high confidence group must also possess a peptide from the high confidence group. The peptides and proteins in the high confidence group are detected in all three biological replicates of the HA treated sample but were undetectable in the untreated control. At this stage, we also filtered out proteins that are known to contain thio-ester bonds unrelated to S-acylation ⁷ (Figure S2). At the end of data processing of stem extracts, we identified 108 proteins in high confidence group and 186 proteins in medium confidence group for the blocked stem sample (Table 1).

Table 1. Summary of proteins, peptides and cysteines in within confidence groups.

Each ambiguous peptide is counted only once and hence reflects the actual number of peptides detected. For the protein and cysteine table, the numbers in ambiguous categories are likely to be overestimates as all proteins represented by the ambiguous peptides are counted. The data is shown for each of the 6 tissues analysed. The last column “best” indicates the best confidence for a peptide/protein/cysteine across all 6 tissues.

	Peptide Type	Confidence Group	Tissue
			Stem	Hypocotyl	Silique	Meristem	Seedling	Leaf	Best tissue
Proteins	Exclusive	High	108	204	277	347	98	152	601
	Exclusive	Medium	186	195	266	336	83	155	493
	Exclusive	Low	329	220	299	306	261	196	544
	Ambiguous	High	31	60	88	140	9	40	223
	Ambiguous	Medium	89	66	141	123	36	66	180
	Ambiguous	Low	198	109	130	124	112	96	185
Peptides	Exclusive	High	128	250	351	436	119	180	798
	Exclusive	Medium	262	290	392	557	98	217	842
	Exclusive	Low	582	364	510	569	410	289	1244
	Ambiguous	High	43	74	107	176	10	52	325
	Ambiguous	Medium	158	106	214	204	53	85	412
	Ambiguous	Low	429	208	284	278	206	158	592
cysteines	Exclusive	High	194	326	445	587	167	251	968
	Exclusive	Medium	331	310	415	586	120	220	881
	Exclusive	Low	697	384	500	549	458	293	1227
	Ambiguous	High	61	83	108	194	12	62	346
	Ambiguous	Medium	174	93	214	199	55	76	367
	Ambiguous	Low	412	191	243	233	197	145	480

In the unblocked sample, we detected a total of 300388 peptide spectral matches (PSMs) from 24793 peptides, out of which 21049 contained at least 1 propionamide modified cysteine and were mapped to a total of 7772 proteins. To enable comparison of this data with stem acylome, we focussed on the HA plus samples. We identified a total of 5001 proteins for which one or more exclusive peptides were detected in at least 2 replicates of HA treated unblocked samples. This set of 5001 proteins represents our detectable cysteine proteome in the stem. By comparing these 5001 proteins with 294 high and medium confidence S-acylated proteins, we estimate the extent of S-acylation in stem to be 5.9% (294 out of 5001).

Having determined the extent of the acylome in stem tissue and a measure of rate of false positives, we analysed 5 additional tissues. We processed 3 biological replicates each from Arabidopsis material collected from: 7-day old whole seedlings, and 4 tissues (in addition to stem tissue described above) collected from 5-week old plants: hypocotyl, siliques, rosette leaves and inflorescence meristem. The entire S-acylation data from all 6 tissues consisted of 76470 PSMs with associated intensity, peptide sequence and peptide modifications. This data was converted into normalised and log2 transformed intensity data for a total of 4804 peptide forms. A PCA plot using this data (Figure 1B) indicated a good clustering of the biological replicates and a good separation between samples treated with HA and controls that were untreated. Any peptides not containing a propionamide cysteine were removed and the remaining 3905 peptides were then matched to Araport11 proteins. 3325 peptides had a single unambiguous database match (exclusive peptides), representing 1916 proteins. A further 580 peptides matched to more than one protein (ambiguous peptides) that mapped up to additional 798 proteins. All proteins, peptides and cysteines were then classified into high, medium and low confidence group as described above (Table 1). The confidence scores were calculated for each tissue as well as an overall score by taking the best score across 6-tissues (Table 1).

Overview of S-acylated proteins

Unless otherwise stated, we have focussed on proteins and peptides belonging to the high and medium confidence group with a p-value of less than 0.01 and/or fold enrichment of >1000. The data for the low confidence group is included in Table S1 and may contain important information regarding acylation sites that is useful for data mining, but it is likely that the rate of false positives in the low confidence group may be higher. Within the high and medium confidence groups, a total of 1094 proteins were identified in one or more tissues (Table 1, Table 2). A comparison of the overlap between tissues shows that out of 1094 proteins, 71 were identified in all 6 tissues and 516 were identified in only one tissue. More than 1/3rd (201) of the proteins identified in only a single tissue were contributed by the meristem that likely reflects a unique acylome in this tissue (Table 2). In contrast only 8% of leaf proteins were uniquely found in that tissue, while more than 73% were found in 3 or more tissues. The ability to detect the same protein as being acylated in different tissues, that were processed and analysed independently, adds further confidence to the data. In order to get a crude estimate of the proportion of the proteome modified by S-acylation, we compared our data to a recent draft of the Arabidopsis proteome²¹. They identified a total of 18175 proteins from 30 different tissues. Analysis of this data suggests that most proteins are present in multiple tissues, so we compared our acylome from all tissues to the entire proteome. Our high and medium confidence group (1094) represents 6% of the 18175 proteins detected in this dataset. This is in good agreement with the figure calculated above by comparing our stem acylome to the detectable proteome.

Table 2. Distribution of acylated proteins across tissues.

Out of 1094 high/medium confidence proteins, 516 were detected in only one tissue while 71 were detected in all 6 tissues.

Tissue	Presence in number of tissues						Total
	1	2	3	4	5	6
Hyp	56	83	65	72	52	71	399
Leaf	25	57	48	56	50	71	307
Meris	201	184	89	83	55	71	683
SDL	26	19	17	24	24	71	181
Silique	133	134	79	74	52	71	543
Stem	75	31	32	43	42	71	294
Total	516	254	110	88	55	71	1094

We compared our acylome dataset with the only previous proteomics study of S-acylated proteins in Arabidopsis ⁷ that identified a total of 582 acylated proteins from root cell suspension culture that were classified as either high, (122) medium (94) or low confidence (336). We identified about 37% of their proteins in our study. The most comprehensive acylomes available in SwissPalm are for human and mouse ³ where more than 4000 S-acylated proteins are listed for both these species. We identified putative homologs of our 1094 S-acylated Arabidopsis proteins from both human and mouse and checked whether these homologous proteins were S-acylated. We identified human homologs for 539 proteins (using an e value of -10), of which 437 (81%) were known to be acylated in one of the 17 studies available ³. The corresponding figures for mouse were 539 and 455 (84%), respectively.

To understand how S-acylation might affect protein localisation, we analysed each protein for presence or absence of TMHs using Phobius ²² and predicted subcellular localisation using SUBA4 ²³ (Table 2). One of the striking features of the data was the very high proportion of proteins that lack TMHs. Of the 1094 proteins, 795 (73%) lack TMHs suggesting that S-acylation will have a large influence on the cellular distribution of the plant proteome. Of the 191 proteins that were predicted to be localised to the plasma membrane in SUBA4, 73 (38%) lack any TMHs and consequently S-acylation is likely to play an important, if not essential, role in anchoring many of them to the plasma membrane.

We next looked at functional classification of acylated proteins using Mapman bins ²⁴. We performed enrichment analysis in various protein lists including all S-acylated proteins (1094) and all S-acylated proteins within each tissue (Table S2, Table S3). Functional classes associated with metabolism of major biomolecules were among the significantly enriched, indicating the diversity of processes involving S-acylation.

One of the major advances of this study is to identify sites of S-acylation. To confirm the accuracy of our data, we compared our dataset to the relatively small number of S-acylation sites experimentally verified for plant proteins ²⁵. In addition to CESA proteins (see below), we also confirmed previously published S-acylation sites on several Rho GTPases. These proteins contain no TMHs, but association with membranes occurs via lipid modifications that are essential for signalling. In Arabidopsis there are 11 Rho of plants (ROPs) that fall in 2 classes ²⁶. Type-I ROPs terminate with a canonical C-terminal CaaX motifs that are attached to the membrane by prenylation. In addition, ROP6 has been shown to undergo transient S-acylation of C21 and C158 ^27,28. Type-II ROPs on the other hand terminate with plant specific GCxxxxxCG domains that are attached to the membrane by S-acylation ²⁹. Our data confirmed the S-acylation of the C terminal cysteines in the type-II ROPS (Figure S4A). RPM1 interacting protein 4 (RIN4) is a small protein which forms a central component of plant defence that associates with the membrane as a result of S-acylation of three C-terminal cysteines ³⁰. In our acylome, we identified that all 3 C-terminal cysteines are modified by S-acylation. Interestingly, we identified multiple peptides for RIN4 that contain various combinations of acylated (propionamide modified) and not-acylated (carbamidomethyl modified) cysteines, such peptides are analysed further in the next section. Remorins are a family of 16 proteins that are established markers of plasma membrane microdomains and known to be S-acylated. It has been shown that the S-acylation of C-terminal cysteines is responsible for membrane association of remorin proteins ³¹. Of the family of 16 remorins we identified that 6 are modified by S-acylation at their C-terminus (Figure S4B). We also detected S-acylation of a conserved N-terminal cysteine (C11) in POLTERGEIST (POL) and related protein PLL5. It is the same cysteine which has been shown to be modified by S-acylation previously³². Together these comparisons illustrate that our acylome dataset has captured most of the known S-acylation sites in plants proteins.

Overview of Acyl-Sites

To be classed into the high or medium confidence group, by definition a protein must have at least one peptide and by extension one cysteine with a p value of <0.01 and/or fold enrichment >1000. The 1094 proteins in these groups contained 1849 high/medium confidence cysteines and while many of the proteins (670) contained only a single high/medium confidence cysteine, the remainder (424) contained multiple cysteines (Figure 1C). There were 28 proteins for which we identified 4 or more high/medium confidence cysteines. These heavily acylated proteins included cellulose synthase proteins like CESA8 and CESA2 and a number of TETRASPANIN proteins that have been shown to be S-acylated in animal species³³ one of which contains 9 S-acylated cysteines. A member of the γ subunit of heterotrimeric GTP-binding proteins (AGG) family, that are known to be S-acylated ³⁴, possesses 47 cysteines of which 23 are S-acylated in our data.

The data allowed us to compare the association of S-acylation with other forms of protein lipid modification. Of our 1094 high/medium confidence proteins, 103 proteins were identified to be S-acylated within the 20 N-terminal amino acids of which, of which 60 were predicted to be modified by N-myristoylation. Similarly, of the 97 proteins that were identified as being S-acylated in the 10 C-terminal residues, 89 were predicted were also predicted to be S-farnesylated. These data suggest that S-acylation of proteins at their N or C terminus is frequently accompanied by other lipid modifications.

Several algorithms have been developed for prediction of S-acylation sites. We used CSS-PALM 4.0 ³⁵ to predict S-acylation sites in our high/medium confidence protein set (1094 proteins) and then we compared the overlap between the predicted sites and sites identified in this data. Out of 1849 high/medium confidence sites experimentally identified in our data, only 353 were predicted by CSS-PALM using the high threshold settings. A further 195 and 103 sites were predicted using medium and low settings respectively making a total of 651 sites out of 1849 (35%) being predicted by CSS-PALM (Table S4). This result demonstrates that while useful, the training sets used for these programs are not adequate, especially for plant proteins where the sites are known only for a small number of proteins. To analyse whether any consensus sequence may exist around S-acylation sites, we divided the proteins in two groups based on whether they contain a TMH or not. In neither of these classes did we identify any clear sequence consensus, but consistent with previous studies ^9,36 we did identify simple sequence motifs enriched in a limited subset of proteins (Table S5).

Using iodoacetamide (carbamidomethyl) and acrylamide (propionamide) to label free and S-acylated cysteines, respectively, allowed us to identify both types of cysteines in a single peptide when they existed in close proximity. The whole S-acylation dataset consisted of 3414 peptide forms (high/medium/low confidence, Table 1) and 3022 unique peptide sequences (Table S6). 2159 of these peptide sequences contained only a single cysteine and hence were capable of producing a single peptide form. Of the 863 peptide sequences that contained 2 or more cysteines, for the majority (750), we identified the version with all cysteines S-acylated, consistent with suggestions of weak cooperativity in S-acylation of closely spaced cysteines ⁶. Out of 863 multi-cysteine peptides capable of producing hybrid peptide form (containing cysteines labelled with both, carbamidomethyl and propionamide), we identified multiple forms for 222 of them (Table S6). It is unclear whether these partially S-acylated forms differ in their functionality.

S-acylation of protein kinases

A recent detailed study of S-acylation of LRR protein kinases analysed 7 well characterised proteins and found them all to be S-acylated, leading to the suggestion it may be a feature of all receptor kinases ³⁷. Hence, we performed a systematic search of LRRs and other protein kinases in our data. For this analysis, we used a previously annotated and categorised list of 942 protein kinases³⁸. There are several protein kinase families that are well represented in our data. We identified a total of 72 protein kinases to be S-acylated, 51 of which were present in our high/medium confidence group (Figures S5). We have classified the protein kinase S-acylation pattern based on the location of S-acylated cysteines in relation to the protein kinase domain and the TMH if present (Figure S5). Out of 51 kinases in high/medium confidence group 36 lacked a TMH, of these 28 were found to be S-acylated close to their N-terminus (Figure S5). 19 of those 28 are also predicted to be myristoylated. A close association between S-acylation and myristoylation at the amino terminus has been described before ³⁹ and the dual lipid modification is most likely responsible for their membrane anchoring. In contrast, MAP3K RAF kinases that lack a predicted TMH were S-acylated at their C-terminus. As observed in previous studies ^37,40, TMH containing kinases were frequently acylated at cysteines close to their TM helix. We also noted that kinases lacking TMHs were often acylated at cysteines that were conserved across the protein family.

Experimental verification of the acylome

We have previously demonstrated that CESA proteins, the catalytic subunits of the cellulose synthase complex, are extensively modified by S-acylation ¹⁶. The CESA proteins were well represented in our dataset (see below); however, we also identified several other proteins that are known to be involved in cellulose synthesis. This included the endoglucanase KORRIGAN1 (KOR1) and COMPANION OF CELLULOSE SYNTHASE 1 (CC1) both of which interact with the cellulose synthesis complex and/or microtubules^41,42. Furthermore within the acylome we also identified a number of proteins known to be involved in CSC trafficking (DET3), exocytosis (EXO84B, and SHOU4, SHOU4L, SHOUL2) and endocytosis (AP2M/µ2) (reviewed in ⁴³) (Table S7). In order to independently verify S-acylation of these proteins, we performed Acyl-RAC assays followed by immunoblotting using proteins transiently expressed in Tobacco and/or in stably transformed lines. These assays verified that KOR1, CSI1, CMU1, CMU2, SHOU4 and SHOU4L were clearly modified by S-acylation, while our negative controls were not (Figure 2). Since the data suggests that all cellulose synthesis related proteins might be modified by S-acylation, we hypothesised that other proteins that are known to be associated with CSC but not identified in our data, for example, CELLULOSE SYNTHASE INTERACTING (CSI) and CELLULOSE MICROTUBULE UNCOUPLING (CMU) would also be S-acylated. Acyl-RAC assays followed by immunoblots revealed that CSI, CMU1, CMU2 and CMU3 are indeed S-acylated to varying degrees (Figure 2B, C).

Figure 2. Validation of S-acylated proteins using Acyl-RAC assay and immunoblotting.

For each assay, the experimental sample (EX) was compared with the loading control (LC) with or without (+/-) hydroxylamine (HA) for hydroxylamine-dependent capture of S-acylated proteins. (A) Acyl-RAC assays were performed on stem sample that was spiked with protein extract from E coli culture expressing GFP fused to histidine-thioredoxin tag and probed with anti-His, anti-HSP70 and anti-laccase antibodies. Additional unblocked controls are included in these assays to demonstrate availability of thiol groups in the sample without iodoacetamide blocking. (B-C) Immunoblots of extracts containing GFP-fusion proteins expressed either transiently in tobacco using 35S promoter (B) or stably in Arabidopsis under the control of their native promoters (C) probed with an anti-GFP antibody.

CSI, CC and CMU all associate with microtubules as well as the cellulose synthase complex. In common with previous reports ⁴⁴, many tubulin subunits were found in our acylome dataset. However, the conserved nature of these proteins, especially alpha tubulins, makes it hard to distinguish individual family members by MS. We selected two alpha tubulins, TUA1 and TUA4, and two beta tubulins, TUB6 and TUB8, based on their expression in secondary wall forming tissues. We expressed GFP fusions of all 4 proteins transiently in tobacco leaves and/or in stable transgenic plants. Acyl-RAC assays confirmed that TUB6, TUA1 and TUA4 were all S-acylated while TUB8 was not (Figure 2B, C).

KOR1 is known to associate with the CESA proteins in the cellulose synthase complex ⁴². KOR1 was found in our acylome data in all 6 tissues, but only a single S-acylated site (C64) was detected. The presence of a single S-acylation site was examined using a variation of the Acyl-RAC assay, known as PEGylation ⁴⁵, in which the acyl groups is exchanged for PEG of sufficiently large size to cause a noticeable mobility shift on SDS-PAGE. Using this assay, the KOR1 antibody detected a dimer that is likely to represent the unmodified form and the modified form with a single PEG addition (Figure 3A) suggesting that KOR1 does only possess a single site for S-acylation. To verify the acyl site, we mutated the C64 to a serine and generated stably transformed lines. Acyl-RAC assays clearly demonstrated the mutation abolished S-acylation (Figure 3B). We analysed multiple phenotypes to assess the complementation of kor1 phenotype including root length with or without oryzalin (Figure 3C, Figure S6), etiolated hypocotyl length in presence or absence of low concentrations of the cellulose synthesis inhibitor isoxaben (Figure S7) and cellulose content of mature stems (Figure S8), but we were unable to identify a significant defect associated with the KOR1^C64S mutant suggesting that the phenotype is likely to be subtle or only apparent under certain conditions. C64 is located close to the TM domain of KOR and previous studies have also failed to identify a phenotype associated with mutations in these juxta-membrane sites ³⁷. Nevertheless, the correct identification of sites of S-acylation that together with the identification of previously described S-acylated proteins provides independent verification that the dataset presented here represents an accurate picture of S-acylation of Arabidopsis proteins.

Figure 3. S-Acylation of KOR1.

(A) Acyl-PEG exchange (APE) assays were performed on 8-days old seedlings tissue powder using 5 kD mPEG-maleimide and analysed on immunoblots probed with anti-KOR1 antibody. Position of PEGylated KOR1 is indicated with an asterisk (B) kor1-1 mutant was transformed with GFP fusions of either wild type KOR1 or the KOR1^C64S mutant expressed under control of the native promoter. Acyl-RAC assays were performed on 7 days old seedlings. For each assay, the experimental sample (EX) was compared with the loading control (LC) with or without (+/-) hydroxylamine (HA) for hydroxylamine-dependent capture of S-acylated proteins. Proteins were detected using an anti-GFP antibody. Approximate position of 100 kd size marker is shown with a black arrowhead in panels A and B. (C) The root length of wild type WS0, kor1-1 and independent T2 lines of kor1-1 transformed with either the wild type KOR or the KOR1^C64S mutant. 30 roots were measured for each line using ImageJ. Box boundaries represent the 25th and 75th percentile, centre line represents the median and whiskers extend to the minimum and maximum values. The dashed horizontal lines indicate mean root lengths of wild type WS0 and kor1-1.

Diversification of RING domain function between CESA classes

To demonstrate how these data may require us to re-think the structure and function of some protein domains, we looked more closely at how the CESA proteins are S-acylated. We previously identified 4 cysteines in the variable region 2 (VR2) and 2 cysteines in the C terminal region (CT) for AtCESA7 as major site of S-acylation ¹⁶. The acylome data presented here confirmed that the VR2 cysteines of the CESA3, CESA4, CESA5, CESA7 and CESA8 are all acylated. We also identified the S-acylation of CT cysteines for CESA2, CESA3 and CESA4. Furthermore, we identified two novel S-acylation sites for CESA proteins (Figure S4C). In particular, we identified a number of cysteines close to the N-terminus that are part of a putative RING-type zinc finger ⁴⁶ motif that is highly conserved in all CESA proteins and related CSLD group of glycosyltransferases. A structure for this region of CESA7 has been solved by NMR and demonstrates it to be a zinc binding RING domain ⁴⁷. Even though the 8 RING domain cysteines are conserved in CESA proteins and were previously considered to be orthologous domains ^46,48, we identified no acylation sites in the RING domain of CESA7 or CESA4 but we identified 3 cysteines of AtCESA8 in either the high or medium confidence dataset (Figure S4C). Although CESA7 cysteines in this region are clearly involved in coordinating Zn atoms⁴⁷ (Figure S9), S-acylation of these cysteines in CESA8 is not compatible with these regions forming a Zn-binding RING domain and suggests a very clear divergence in the structure, and probably the function, of this domain between CESA7 and CESA4/8. The surprising nature of this result made us consider whether the RING domains may exhibit some peculiar properties that would result in the cysteine residues being labelled in our study. The majority of RING domains fall within PFAM clan CL0229 and of the 486 members of CL0229 in Arabidopsis, 14 proteins, including 5 CESA proteins and CSLD2, were identified to be acylated in our high/medium confidence group. Of these, only 3, CESA8 and two other proteins, were acylated within the RING domain cysteines. Together, this data appears to support the idea that apart from CESA proteins putative RING domains are poorly represented in our data and for some CESA proteins it is likely that, under some conditions at least, this domain forms an alternative structure.

Since our data suggested a divergence in the structure of the N terminus among different CESA proteins, we aligned the RING domain sequences (46 amino acid long stretch) of all CESA proteins from 43 fully sequenced genomes and generated a consensus sequence for each CESA class (Figure S9B, C). The RING domain of CESA7 was most highly conserved; excluding the 8 conserved cysteine residues, 30 of the remaining 38 amino acids exhibited more than 90% conservation across species. In contrast, the consensus for CESA8 exhibited much greater variation with only 11 residues showing more than 90% conservation across the class, while the RING domain of CESA4 also appeared less well conserved than that of CESA7 with 16 invariant residues. We interpret this analysis as the higher conservation of CESA7 retaining its function as RING domain while the selective pressure on CESA4 and CESA8 leading functional divergence but retaining the conserved cysteines.

In order to validate the differential S-acylation of the RING domain cysteines of CESA 4, 7 and 8 directly, we analysed the VR2/CT cysteine mutants of these proteins. We have previously demonstrated that when all cysteines within the variable region 2 (4 cysteines) and C-terminal region (2 cysteines) of CESA7 are mutated, the level of CESA7 S-acylation was nearly abolished (Figure 4) ¹⁶. Similar results were observed when we mutated the cysteines in the same regions of CESA4 (8 cysteines). In contrast, mutation of VR2 and CT cysteines of CESA8 (5 cysteines) resulted in little or no reduction in the level of protein S-acylation (Figure 4). This result is consistent with CESA8 still being extensively S-acylated at sites other than the VR2 and CT sites, while CESA7 and CESA4 are not. These other sites are likely to be the N-terminal cysteines previously considered part of the metal binding RING domain.

Figure 4. Differential S-acylation of CESA isoforms.

(A) Diagrammatic representation of CESA protein structure indicating the position of cysteine residues (vertical bars). Cluster of mutated cysteines in the RING domain are coloured blue, while cysteines mutated in the variable regions 2 (VR2) and C-terminus (CT) are coloured red. The transmembrane (TM), conserved (CR1, CR2) and variable 1 (VR1) regions are also indicated. (B) WT or VR2/CT version of CESA7, CESA4 and CESA8 transformed into cesa7^irx3-7, cesa4^irx5-4 and cesa8^irx1-7 respectively. Acyl-RAC assays were performed on mature stem material. For each assay, the experimental sample (EX) was compared with the loading control (LC) with or without (+/-) hydroxylamine (HA) for hydroxylamine-dependent capture of S-acylated proteins. Detection was performed using protein specific antibodies. Three independent lines were analysed for each genotype, blots for one of which are shown.

If a sequence forms a RING domain, we can make some clear and testable hypotheses, principally that all 8 cysteines are required to coordinate metal ions and are essential for function. To test this hypothesis, we performed comprehensive mutagenesis of these cysteines in CESA4, CESA7 and CESA8. Initially, each individual cysteine was mutated to serine and the constructs transformed into their respective null mutants. Consistent with the known structure (Figure S9A) and the essential nature of the RING domain in CESA7 ¹⁶, mutating any single cysteine is sufficient to completely abolish any complementation of either cellulose content phenotype (Figure 5) or plant height (Figure S10). In contrast, when we mutate the corresponding cysteines of CESA4 or CESA8, we observe partial but significant complementation of both cellulose content and plant height (Figure 5 and S9). To confirm that our mutant proteins will no longer be able to function as a metal binding RING domain, we constructed a series of additional constructs in which we mutated pairs of cysteines from each putative Zn binding site (Figure S9A), or 4 cysteines, two from each putative Zn binding site or mutated all 8 cysteines. As expected, none of the CESA7 constructs were able to complement the cesa7^irx3-7 mutant. In contrast, the multiple cysteine mutant constructs of CESA4 and CESA8 were still able to complement the mutant in a manner comparable to the single cysteine mutants even though we designed the mutations to ensure the sequence is no longer able to function as a metal bind RING domain (Figure 5). In fact, for both CESA4 and CESA8 the construct in which all 8 cysteines were mutated gave a level of complementation comparable to complementation with the wild type construct (Figure 5). To test whether the residues of RING domain other than the conserved cysteines were important for protein function, we deleted the entire RING domain (all residues including and between the 1^st and 8^th cysteine). While deletion of the RING domain from CESA7 was essential for function, constructs in which this region was deleted from CESA4 exhibited significant complementation. Similarly, constructs with the same region of CESA8 deleted exhibited relatively poor, but significant complementation (Figure 5). We also performed complementary domain swaps of these residues with same region from other two CESAs (Figure 5, S10, S11). The region from either CESA4 or CESA8 swapped into CESA7 were not able to complement the cesa7^irx3-7 mutant. In contrast, domain swaps into CESA4 and CESA8 both exhibited significant complementation; however, the pattern of complementation exhibited the same trend as the deletion mutants with relatively good complementation of the cesa4^irx5-4 mutant and only marginal complementation of then cesa8^irx1-7 mutant. These deletions and domain swaps complementation studies reveal a consistent trend. For CESA7 they support formation of a RING domain that is essential for protein function. Consistent with the highly conserved nature of the CESA7, the same regions from CESA4 or CESA8 will not function in place of the sequence from CESA7. In contrast, for CESA8, deletion of this region still allows for some complementation, but much better complementation is obtained if the domain is retained even with all 8 cysteines mutated suggesting that while this domain is important for protein function, the cysteines residues appear to be not essential.

Figure 5. Analysis of RING domain cysteines in CESA7, CESA4 and CESA8.

Cellulose content of cesa7^irx3-7, cesa4^irx5-4 and cesa8^irx1-7 mutants transformed with the corresponding genes in which either one or more cysteines have been mutated to serines or the entire region was deleted or swapped with the same region from other CESAs. For each genotype at least 8 plants were analysed. The suffix VO indicates plants transformed with empty vector only. The number at the base of each bar refer to which cysteines are mutated and correspond to those shown in Figure S10. Horizontal dashed line indicates cellulose content of background mutants. Error bars shown are standard error of mean. Significance levels were generated using univariate ANOVA and are shown at 0.001 (***), 0.01 (**) and 0.05 (*) for comparison of cysteine mutants with their respective T-DNA mutants or as otherwise indicated. Part of the domain swap data has been published previously in Kumar et al, 2017.

Based largely on studies performed following expression in yeast, it was suggested that the RING domain cysteines of a CESA from cotton act to reversibly dimerise CESA proteins by forming disulphide bonds ⁴⁹. While our data do not support such models, they do suggest that in CESA4 and CESA8 this domain has diversified and is able to form structures other than a canonical zinc binding RING domain.

Concluding Remarks

We present a large-scale atlas of protein S-acylation across six tissues of Arabidopsis, thus enabling a broad study of this elusive modification and its biological roles in plants. Many of the proteins we identified in our acylome are predicted to be membrane localised even though they lack any TMHs. S-acylation is likely to have an important, if not essential, role in anchoring these proteins to the membrane and consequently is important to the overall organisation of plant cells. We have demonstrated the robustness of the data by identifying many of the plant proteins that have been identified as S-acylated in previous studies. While we have focused on proteins that fall into our high/medium confidence group, this is likely to be biased in favour of more abundant proteins that are preferentially detected by MS. Lower abundance proteins are likely to have lower confidence scores. However, many of these proteins would be truly S-acylated and should be taken into consideration when examining one’s favourite group of proteins. Our high/medium confidence group contains 1094 proteins that represent around 6% of the detectable Arabidopsis proteome. While this number initially appears high, it is comparable with the number of proteins identified in humans and mice. This suggests that S-acylation is a very important modification and likely to be as essential for the proper functioning of many plant proteins as it is in metazoans.

When a protein is identified as S-acylated, there is no measure of the relative amount of the modified vs unmodified form. Assays such as PEGylation provide a measure of the relative proportion of different protein forms but are currently low throughput. PEGylation of KOR suggests that a large proportion of the protein is acylated (Figure 3A), but it may be much lower for other proteins and may also vary dramatically in different tissues and/or under different environmental conditions. For proteins in which only a small proportion of the protein maybe S-acylated, it is unclear whether this is necessarily biologically relevant. Given the apparent ability of PATs to acylate any accessible cysteine ⁶, it is not hard to envisage that small proportion of proteins might become acylated as a result of chance encounters or protein misfolding. These types of interactions may just be serendipitous and of no biological relevance, however, they may represent a means of the cell identifying proteins that are either misfolded or mislocalised or represent some other kind of signal. In addition, while we have introduced stringent controls to demonstrate a low rate of false positives, it remains possible some putative S-acylation sites could be a result of unintended and unknown effects of hydroxylamine. We consider that first and foremost, the acylome data presented may be mined as a starting point for identifying how this modification affects protein function. We demonstrate the utility of the dataset by demonstrating the important differences in the structure and function of a conserved domain at the N-terminus of all CESA proteins that was previously assumed to perform similar function in various CESA classes.

Overall, our data is generated by a method that is optimised to be the best method currently available. The results indicates that S-acylation is widespread in plants proteins. Its role in cell wall biosynthesis and dynamics is likely to be even broader than previously shown. Novel functions in cellular traffic, signalling, defence and membrane-cytoskeleton interactions are very likely. We have shown how the identification of S-acylation sites in CESA proteins has forced a re-appraisal of structure-function relationships. Our acylome data, resolved to more than 4000 individual cysteine residues in all our confidence groups, makes it possible to address these and other important biological questions.

Materials and methods

Plant material

For proteomics, Arabidopsis Col0 seed was grown for 7 days on plates followed by a further 4 weeks on soil as previously described ⁵⁰. The material collected was composed of inflorescence stem (after removing flowers, siliques and leaves and the basal 1cm), siliques (at various developmental stages), meristem (apical 1 cm of stem tip including unopened flowers) and rosette leaves. Seedlings were grown on ½ strength MS plates for 7 days.

The loss of function mutants used in the study, cesa8^irx1-7, cesa7^{irx3-7 51}, cesa4^{irx5-4 52}, cc1, cc2, cc1 cc2 ⁴¹, cmu1, cmu2, cmu1 cmu2 ⁵³, csi1 ⁵⁴ and kor1-1 ⁵⁵ have been previously described. tub6-2 (SALKseq_126440) and tub8-1 (SALK_116231) were obtained from NASC and homozygous plants were isolated and confirmed by PCR based genotyping and crossed to create tub6 tub8 double mutant. All plant transformation and complementation analysis were as performed as described previously ^16,48,50. Cellulose content was determined using a medium throughput adaptation of Updegraff’s method ⁵¹. For KOR1 lines, complementation assay was performed using plate based assays.

Transient expression in tobacco has been described in detail previously ⁵⁶. Agrobacterium solutions were infiltrated into lower side of 4 weeks old Nicotiana benthamiana plants using a 1-ml syringe. The infiltrated leaves were harvested 4 days after infiltration, frozen in liquid nitrogen and stored at -80°C until further use.

E. coli protein expression

GFP protein tagged a 6x histidine and thioredoxin tag was expressed in BL21 cells. Cells were grown at 37°C for 3 hours, induced with IPTG and grown for further 20 hours. Cells were pelleted and resuspended in native extraction buffer containing 20 mM Tris pH 8.0, 10 mM Imidazole pH, 150 mM NaCl, 0.2% Triton X-100, 2 mM Mercaptoethanol, 1 mM PMSF, protease inhibitor, 1 mg/ml Lysozyme and 5 μg/ml DNAseI and sonicated for 2 minutes. Debris was removed by centrifugation and supernatant from 6 different preps was mixed. Protein content was measured using BCA method and known amount was used for spiking the plant samples.

Acyl-RAC assay, trypsin digestion and mass spectrometry

The optimised Acyl-RAC assay was based on previously described methods ^18,57. About 800 mg of frozen tissue powder was pre-cleared by adding 20 mL of 10% TCA in ice-cold acetone with 10 mM TCEP and incubating at -20°C for 1 hour. Tissue pellet was washed twice with ice cold acetone to remove TCA and TCEP and air dried. The pellet was then solubilised in 20 mL of lysis buffer (6M Urea, 100mM Tris pH 7.2, 150mM NaCl, 5mM EDTA, 2.5% SDS) with 50 mM iodoacetamide, 1 mM PMSF and Sigma plant protease inhibitor cocktail) and incubated at room temperature for 20 minutes on end-on-end rotation. Samples were then centrifuged at 4500g and supernatant filtered through miracloth and incubated at room temperature on a roller for 1 hour to block free cysteines. Free iodoacetamide was removed by chloroform methanol precipitation. Protein pellet was washed twice with 80% methanol and air dried followed by resuspension in 20 mL lysis buffer. Protein concentration was measured using Millipore BCA assay kit according to manufacturer’s instructions and typically revealed a protein concentration of 1-2 mg/mL. 100 µL of E coli extract (2 mg/mL) was added to the stem samples just prior to the TCA precipitation step. For unblocked sample, lysis buffer without iodoacetamide was used.

9 mL of blocked protein sample was combined with 4.5 mL of either 1M hydroxylamine (final concentration 333 mM) or water and 50 mg (dry weight equivalent) thiopropyl sepharose beads (GE Healthcare Catalog number 17042001) and incubated for 1 hour at room temperature in a column fitted with a porous disc. Samples were washed for 5 minutes in 3.5 ml: wash buffer (3x, 100 mM Tris HCl pH 7.4, 500 mM NaCl, 0.1 % SDS)and 50 mM ammonium bicarbonate (AmBic) (4x) and the beads dried completely by centrifugation at 300g for 2 minutes. The columns were capped at the bottom and 200 µl AmBic containing 4 ng/µl of Trypsin + LysC (Promega Catalog number V5073) was added to each sample and incubated in a 37°C shaker for 16 hours. After digestion, unbound peptides were collected by centrifugation at 300g for 2 minutes and the beads were then washed again with 8M urea (1x), 2M NaCl (1x), 80% acetonitrile + 0.1% TFA (1x) and 50 mM AmBic (3x). Bound peptides were then eluted by adding 200 µl AmBic containing 50 mM DTT and incubating in a 37°C shaker for 30 minutes followed by centrifugation at 300g for 2 minutes. Peptides were alkylated by adding acrylamide to a final concentration of 100 mM and incubating in dark at room temperature with gentle agitation for 2 hours. Further DTT was added to bring final concentration to 100 mM to quench the unreacted IAM and samples stored at -20°C until further use.

Reduced and alkylated peptides were acidified by adding equal volume of 0.1% formic acid. Acidified samples were bound to OLIGO R3 reverse phase resin (Thermo Scientific, catalogue number 1133903), washed twice with 0.1% formic acid and eluted with 200 µl of 30% acetonitrile + 0.1% formic acid. Peptides were then quantified using Pierce Quantitative Fluorometric Peptide Assay (Thermo Scientific, catalogue number 23290) according to manufacturer’s instructions. Known quantities (500 – 3000 ng) of peptides were then aliquoted into mass spec vials and dried completely using a speed vac.

Digested samples were analysed by LC-MS/MS using an UltiMate® 3000 Rapid Separation LC (RSLC, Dionex Corporation, Sunnyvale, CA) coupled to a Exploris 480 (Thermo Fisher Scientific, Waltham, MA) mass spectrometer. Peptide mixtures were separated using a multistep gradient from 95% A (0.1% FA in water) and 5% B (0.1% FA in acetonitrile) to 7% B at 1 min, 18% B at 35 min, 27% B in 43 min and 60% B at 44 min at 300 nL min-1, using a 75 mm x 250 μm i.d. 1.7 µM CSH C18, analytical column (Waters). Peptides were selected for fragmentation automatically by data dependant analysis.

Database searching

Database searching and PSM (peptide spectrum matches) processing was performed in Proteome Discoverer 2.5 (Thermo Fisher Scientific, UK). Two complementary search engines Sequest™ HT (Thermo Fisher Scientific, UK) and Mascot (Matrix Science, UK) were used for searching the data. These engines were set up to search the Arabidopsis thaliana Araport11 protein database (48359 entries) and uniport E coli proteome (for stem sample only) assuming the digestion enzyme trypsin. A Precursor Mass Tolerance of 10 ppm and Fragment Mass Tolerance of 0.02 Da was used for Sequest while Mascot was searched with a Precursor Mass Tolerance of 20 ppm and Fragment Mass Tolerance of 0.02 Da. Oxidation of methionine, carbamidomethyl of cysteine and propionamide of cysteine were specified as variable modifications for both search engines. All other program settings were kept as default. Modifications and precursor ion intensity were exported out of Proteome Discoverer and majority of further calculations and data processing was performed in Microsoft Excel 365.

Peptide and Protein confidence scores and classes

The data was first normalised using assay total signal normalisation. Each biological replicate constitutes one assay and includes 2 sample types, HA-plus and HA-minus. So there are 18 assays in the data, 3 for each tissue. Average assay total intensity across 18 assays was 1.4E10. For simplicity, the intensities were normalised for total signal intensity for each assay to be 1E10. The normalisation factor thus calculated was used to compute normalised intensity for each PSM.

The data was then reduced to peptide forms (defined as a combination of peptide sequence and modification) and intensities were computed for each peptide form by taking the sum of intensities of all PSMs matching to a particular peptide form. Three modifications were used in the searches: oxidation of methionine, carbamidomethyl of cysteine and propionamide of cysteine. At this stage, PSM intensities for all peptide forms were summed together leading to a total of 4803 peptide forms.

In order to get meaningful fold enrichment ratios, missing values were replaced with 2 (a small number relative to average peptide intensity of ˜590000) and data was log2 transformed. T-test P values and fold enrichments were then calculated on the transformed data.

The peptides sequences were then mapped to Araport11 proteome. When multiple splice variants matched to a peptide, only one splice variant (with lowest isoform number) was kept for each locus. For each protein, total intensity was computed by taking sum of peptides matching to that protein and computed for exclusive and ambiguous peptides separately. For proteins where an exclusive peptide was available, its protein score was based only on exclusive peptide intensities. Ambiguous peptides contributed to protein score calculation only when exclusive peptides were not present.

Next, a protein class was computed by comparing the protein score and the peptide score of the best peptide matching to that protein. The lower of those two scores was considered to be the protein class.

Further analysis of proteomics data

Over-representation analysis of MapMan ontology bins²⁴ was performed using online sever ORA (https://usadellab.github.io/MapManJS/test_oo11.html). Protein lists were compared against Araport11 proteome using a 2-sided Fisher’s exact test.

Motif discovery analysis around acyl-sites was performed using MoMo⁵⁸ with motif-x algorithm (https://meme-suite.org/meme/tools/momo). A 13 amino-acid long sequence (acyl cysteine with 5 amino acids on either side) was used for the analysis and the whole Araport11 proteome was used as background context sequences.

Predicted lipid modifications of protein, S-acylation and N-myristoylation were analysed with GPS-Lipid ⁵⁹. Proteins identified in this study were processed through GPS-lipid and all predicted sites with high, medium and low cut-off settings were recovered.

DNA Cloning

Gateway technology (ThermoFisher Scientific, UK) was used for construction of most of the plasmids. Cloning of wild type CESA8, 7 and 4 CDS fragments in a Gateway entry vector, pDONOR/pZEO, has been described previously ⁴⁸. All other CDS fragments including KOR, CC1, CC2, CMU1, CMU2, CMU3, CSI1, TUA1, TUA4, TUB6, TUB8, SHOU4 and SHOU4L were amplified using primers listed in Table S8 and entry clones were made using the same procedure. The vector for expressing CESA7 has been described previously. New destination vectors were constructed using pCambia1300 backbone. The component fragments of these vectors (promoters, tags, Gateway cassette and the terminator) were PCR amplified and cloned into a pJET vector using a CloneJET PCR Cloning Kit (ThermoFisher Scientific, UK). All primers (Table S8) included appropriate restriction sites to allow concatenation of the fragments. The inserts in the pJET vector were fully sequenced before assembling to create the final destination vectors. Cysteine mutants of CESA4, CESA7, CESA8 and KOR1 were made using PCR-based site directed mutagenesis ^16,60. Sequences of different GSF Primers used are listed in Table S8. Plasmids were sequenced fully to verify that no mutations had occurred.

Acyl-RAC immunoblots and Acyl PEG Exchange Assay

For stably expressed proteins, multiple independent lines were tested for protein expression levels. Tissue was ground to a fine powder in liquid nitrogen before being solubilized in Lamelli buffer. Roughly equal amounts of crude protein extract were separated on a 7.5% SDS-PAGE gel before being transferred to a PVDF membrane for western blotting. The blots were then probed with anti-GFP or protein specific antibodies and lines were selected from these analyses (Figure S12) for Acyl-RAC assays.

The Acyl-RAC assay protocol was the same as described above in the proteomics section. After washing the TPS beads with the wash buffer, instead of trypsin digestion, proteins were eluted with DTT. The proteins were separated on 7.5% SDS-PAGE gels, blotted onto PVDF membranes and probed with various antibodies. Primary antibodies used in this study included: Sheep anti CESA 7^48,61 (used at 1: 2000 dilution), Sheep anti CESA 8^48,61 (used at 1: 500 dilution), Sheep anti CESA 4^48,62 (used at 1: 100 dilution), Rabbit anti KOR1 ⁶³ (used at 1: 500 dilution), Mouse anti Histidine (Bio-Rad, Cat# MCA1396GA, used at 1: 2000 dilution), Mouse anti HSP73 (Enzo life sciences, Cat# ADI-SPA-818, used at 1: 20000 dilution), anti-Laccase (peptide antibody raised in rabbit against Apple laccase MDP0000242301; peptide sequence RDIHSEKWKGPRSTN, used at 1: 500 dilution) and Mouse anti GFP (Invitrogen, Cat# 14-6674-82, used at 1: 1000 dilution). The secondary antibody used were donkey anti-Sheep (Thermo, Catalog # A16041, used at 1:5000 dilution), goat anti-Rabbit HRP (Dako, Cat# 0448, used at 1:1000 dilution) and goat anti-Mouse HRP (Dako, Cat# 0448, used at 1:1000 dilution).

The acyl PEG exchange (APE) assay was performed using a modification of previously described protocol ⁴⁵. The initial part of the assay including tissue precipitation, crude protein extraction, and iodoacetamide blocking was the same as described above for the Acyl-RAC assay. After iodoacetamide blocking, the protein sample was divided into 2 portions. HA (final concentration 333 mM) was added to the HA+ sample while water was added to the HA-sample. After incubating the samples for 1 hour with gentle rotation, the protein samples were precipitated using chloroform methanol precipitation method. After washing the pellet with methanol, the pellet was dried and resuspended in 1 mL of 6M urea, 100 mM Tris HCl, pH7.4, 4 mM EDTA, 4 % SDS. mPEG maleimide 5 kD (Sigma) was then added to a final concentration of 1 mM and samples incubated at room temperature with gentle agitation. The samples were then precipitated again and resuspended in 100 µl of 2x Laemmli buffer without DTT and analysed on non-reducing SDS-PAGE followed by botting and probing with anti-KOR antibody.

Table 3. Prediction of sub-cellular localisation of high confidence proteins.

Proteins from high and medium confidence class were analysed for subcellular localisation using SUBA4 and presence of trans-membrane helices (TMHs) using Phobius.

SubaCon localisation prediction	All	TMH	No_TMH	No_TMH (%)
cytosol	208	12	196	94
plasma membrane	191	118	73	38
plastid	176	29	147	84
extracellular	150	28	122	81
nucleus	117	4	113	97
golgi	57	42	15	26
mitochondrion	52	8	44	85
vacuole	36	23	13	36
endoplasmic reticulum	35	21	14	40
peroxisome	29	1	28	97
nucleus,cytosol	15	0	15	100
Other	28	13	15	54
Total	1094	299	795	73

Data availability

All processed data is included with the manuscript as supplementary tables. Unprocessed mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE ⁶⁴ partner repository with the dataset identifier PXD031348 and 10.6019/PXD031348. All figures have associated raw data, which will be provided upon request. Araport11 (https://www.arabidopsis.org/index.jsp) and E coli UP000000625 (https://www.uniprot.org/proteomes/UP000000625) were used as the protein databases for Arabidopsis and E. coli for performing mass spectrometry database searches.