a, Pairwise global Pearson’s expression correlation analysis of all 30 tissues (n = 1 measurement per tissue) on the transcriptome level (bottom triangle) and proteome level (top triangle) using all identified gene loci. Proteins correlate more strongly between tissues than transcripts. Turquoise squares mark examples for morphologically highly similar tissues. Tissues are coloured as in Fig.1.b, Scatter plots showing highly reproducible abundance measurements for transcript (top) and protein (bottom) in morphologically similar tissues that were marked ina; namely, node (ND) versus internode (IND), leaf distal (LFD) versus leaf proximal (LFP) and root (RT) versus root upper zone (RTUZ).r denotes the Pearson’s correlation coefficient;n denotes the number of transcripts or proteins.c, Percentage of genes encoded by a specific chromosome that were identified at the transcriptome, proteome or phosphoproteome level.d, Percentage of Swiss-Prot and TrEMBL protein database entries as well as protein evidence categories from UniProt that were identified at the transcriptome, proteome or phosphoproteome level. Evidence level: (1) protein evidence; (2) transcript evidence; (3) homology; (4) predicted; and (5) uncertain.e, Comparison of protein identifications between an earlierArabidopsis proteome study⁷ based on 12 tissues, this study (30 tissues) and the number of protein-coding genes in Araport11.f, iBAQ intensity distribution of proteins identified in this study. Proteins also identified in a previous study⁷ are projected into the same plot.g, Left, proportion of identified P-sites on S, T or Y residues with highly confident localization of the phosphorylation site within the identified peptide sequence (termed class I P-sites if the localization score is greater than 0.75). Right, distribution of proteins for which phosphorylated S, T or Y residues were identified.h, Left, Venn diagram comparing phosphoprotein datasets from a previous publication⁸, PhosPhAT4.0 and this study. Right, Venn diagram comparing P-site localization confidence between class I sites identified in this study and the low and high confidence datasets reported in a previous publication⁸.

Source data

a, Number of identified N-terminal (NT) or C-terminal (CT) peptides of proteins in either unmodified or phosphorylated form.b, Frequency of amino acids following the initiator methionine in N-terminal peptides with (−X) or without (M−X) cleavage of the initiator methionine. X denotes the amino acid after the start codon.c, Frequency of protein N-terminal acetylation for amino acids inb. Because trypsin was used for protein digestion, the frequencies for Arg and Lys residues could not be determined (n.d.).d, Distribution of peptide-based sequence coverage of proteins in individual tissues and for the combined dataset (tissue abbreviations as in Fig.1). Boxes contain 50% of the data and show the median as a black line. The top and bottom quartile ranges are shown as whiskers. The number of proteins is indicated for each tissue.e, Pie charts showing the percentage of proteins identified by <3, 3–10 or >10 peptides either allowing shared (razor) peptides or restricting to unique peptides only.f, Left, number of protein isoforms detected at the transcript and protein level compared with the number of all annotated isoforms in Araport11. Right, number of multiple isoforms of the same gene distinguished at the peptide level.g, Validation of protein isoform and sORF identification by comparing the tandem mass spectra from the tissue atlas to those of synthetic peptide reference standards. The normalized spectral contrast angle (SA) was used as a similarity metric (Methods). Candidate isoforms and sORFs were considered valid if the spectral contrast angle of the spectra was >0.7. These data are reported in Supplementary Data 3.h, Amino acid sequence and mirror plots of tandem mass spectra for two peptides of the sORF BIP138_4. The spectra pointing upwards were collected from tissue digests; those pointing downwards were collected from synthetic peptides. The normalized spectral contrast angle and Pearson’s correlation coefficient (r) were used as similarity metrics (Methods) and indicate that both high-scoring spectra (n = 1 acquired spectra) are near identical, thus validating the identification of this sORF as an expressed protein.i, Dynamic range of transcript abundance (grey) and proportion of transcripts that were also identified at the protein level projected into this plot (blue). OM, orders of magnitude. Note that for lower abundance transcripts, fewer proteins were detected.j, Dynamic range of protein abundance and proportion of proteins with phosphorylation evidence. Protein abundance spans six orders of magnitude, whereas transcript abundance only spans four (i). In addition, note that phosphorylation was detected across the entire protein abundance range.k, Percentage of all annotated kinases (K), phosphatases (P), transcription factors (TF) and transcription regulators (TR) detected at the transcript, protein or phosphoprotein levels. Numbers below thex axis denote the number of genes for these protein classes in theA. thaliana genome.

Source data

a, Distribution of expression specificity categories for protein and transcript identifications. See Methods for the definition of these categories. In brief, there are very few transcripts and proteins that are only expressed in a single tissue. The quantities of the shared transcripts or proteins can differ vastly between tissues (b).b, Left, protein identifications shared between flower (FL) and flower organs showing an almost complete qualitative overlap of proteins. Sepal (SP), petal (PT), stamen (ST), carpel (CP). Right, clustering ofz-scored protein intensities showing distinct quantitative expression differences between flower organs.c, Expression analysis of flower organ identity marker at the protein and transcript level. PISTILLATA (PI, green), APETALA3 (AP3, red), APETALA1 (AP1, orange), AGAMOUS (AG, blue). The expression of these markers is in line with the model of flower organ identity (AP1 expression marking sepal, AP1, AP3, PI marking petal, AG, AP3, PI marking stamen and AG marking carpel).d, Total number of transcripts plotted against the total number of proteins detected in each individual tissue (n = 30 tissues) showing that the more genes are expressed as mRNAs, the more proteins can be detected in a tissue (Pearson’s correlationr = 0.79). Tissues are coloured according to tissue groups as in Fig.1.e, Cumulative abundance plots of intensity-ranked identifications of transcripts and proteins for five representative tissues. The five most abundant transcripts and proteins are listed in descending order for each tissue. These are generally not the same. In addition, note that the characteristics of the plots are not the same for all tissues. In flower, the protein line rises more quickly than the transcript line. The opposite is true for pollen and a more even characteristic is observed in seed.f, Distribution of shared and unique identifications among the 100 most abundant transcripts and proteins in each tissue. Relatively few proteins and transcripts are found together on the list of the 100 most abundant transcripts and proteins. This demonstrates that the quantitative differences in transcript and protein expression are more important in defining a tissue than the qualitative expression of transcripts or proteins.g, List of 11 proteins that were found as the most abundant protein (in at least one tissue) and their proportion of the total iBAQ intensity in each tissue. Individual proteins can represent up to 9% of the total protein in a given tissue.h, Principal component analysis (PCA) of the core tissue proteomes and transcriptomes (that is, the proteins and transcripts that were identified in every tissue) usingz-scored abundances. Only about 30% of all protein and 20% of all mRNAs were detected in every of the 30 tissues despite the fact that all tissues were deeply profiled at both protein and transcript level. This shows that strong qualitative and quantitative expression differences exist between tissues. The PCA separates tissues into photosynthetically active versus inactive tissues (component 1) and separates pollen from all other tissues (component 2), indicating that the molecular composition of pollen is particularly different from all other tissues.i, Proportion of the total summed protein intensity for genes with specific subcellular compartment annotation (from SUBA⁷⁷;Methods) in the different tissue groups. The comparison of photosynthetically active and inactive tissues shows that most of the protein content in photosynthetically active tissues is contained in the plastids, whereas most protein is found in the cytosol for photosynthetically inactive tissues. Proteins with only one single subcellular compartment annotation were selected for the plot and the proportion of their iBAQ intensities were averaged for each tissue group. Nucleus (n = 1,393), endoplasmatic reticulum (n = 58), Golgi (n = 68), peroxisome (n = 67), plastid (n = 525), mitochondrion (n = 317), vacuole (n = 71), cytosol (n = 385), cytoskeleton (n = 1), plasma membrane (n = 268), extracellular (n = 351).

Source data

a, Pearson’s correlation (r) of transcriptome and proteome expression (core datasets;n = 5,043) for each tissue.b, Pearson’s correlation between measured and predicted protein abundance levels in all tissues. Predicted protein abundance levels were obtained from the best fitting feature selection model for each tissue (Methods). The number of genes used for the correlation analysis is indicated for each tissue.c, Violin plots showing the spread in relative contribution of selected features to the prediction of gene-level protein abundance across tissues (n = 30 tissues) using our model. Violin shapes show the kernel density estimation of the data distribution and the median as white dot. Thick black bars denote the interquartile range.d, Specific nucleotide sequence motifs in 5′ UTRs of mRNAs contribute to the prediction of protein levels in a subset of tissues. Clustering tissues based on the presence or absence of detected 5′ UTR motifs shows that several features are repeatedly selected for inclusion in the model while others appear to be more tissue-specific.e, On the basis of the observation that thed_N/d_S ratio between orthologous ofA. thaliana andA. lyrata contributed to the prediction of protein levels (c), we analysed this feature in more detail. Left, distribution of thed_N/d_S ratio for orthologous genes inA. thaliana andA. lyrata. The distribution is plotted for the example of ‘leaf distal’ (n = 6,447 genes). To compare evolutionarily conserved genes (defined by lowd_N/d_S ratios) and genes that evolve neutrally or are under positive selection (highd_N/d_S ratios), we selected the bottom 5% and top 5% of thed_N/d_S ratio distribution, respectively. Right, evolutionarily conserved genes (lowd_N/d_S ratio) show 10–20 times higher protein abundance than genes under evolutionary pressure. Boxes contain 50% of the data and show the median as a black line. Whiskers denote 1.5 times the interquartile range. Outliers were omitted from the plot for clarity.f, Time-course analysis of median protein abundance changes after treatment with CHX (translation block) or MG132 (proteasome block) versus time-matched DMSO control samples (Methods). Boxes contain 50% of the data and show medians as black lines. Whiskers denote 1.5 times the interquartile range. Outliers were omitted from the plot for clarity but were included in the statistical tests below. All proteins in the experiment (n = 8,920, grey), proteins that have a high PTR in seed (n = 425, red) or a low PTR in seed (n = 254, blue) (defined as in Fig.3d) are shown. Differences between time points were tested for significance within each subset (all; high PTR; low PTR) using one-way ANOVA and the post hoc Tukey HSD test. ***P < 0.001 (all_CHX8–CHX16:P < 1 × 10⁻⁷; all_CHX8–CHX24:P < 1 × 10⁻⁷; all_CHX16–CHX24:P = 0.0002; highPTR_CHX8–CHX24:P = 0.0003; lowPTR_CHX8–CHX16:P = 0.0000004; lowPTR_CHX8–CHX24:P < 1 × 10⁻⁷; lowPTR_CHX16–CHX24:P < 1 × 10⁻⁷).g, Representative images of seeds after 4 days of incubation with CHX, MG132 or DMSO control medium (n = 1). Germination was completely inhibited by CHX and partially inhibited by MG132, showing that the drug treatments were effective.

Source data

a, Median PTRs across tissues plotted against the inter-tissue variation of these PTRs (expressed as MAD; proteins and transcripts had to be detected in at least 10 matching tissues to be included in the analysis). Arrows denote examples of genes with high PTRs (rbcL andpetA) and low PTRs (IAA8 andIAA13). Bar plot shows the MAD range segmented into five quantiles, each containing the same number of genes (coloured bars and dashed lines). Most genes have reasonably stable PTRs across tissues.b, As ina (datasetn = 14,069) but for transcript (left) and protein (right) measurements. There is more variation in protein levels across tissues than there is mRNA variation (80% of all transcripts show a MAD of <1; 80% of all proteins show a MAD of 1.2). There is also more variation in the protein levels across tissues for low abundant proteins. This may in part be due to technical limitations as low abundance proteins can generally be less accurately quantified.c, As ina but for the ratio of phosphorylation site versus protein abundance. P-sites and proteins had to be detected in at least 10 matching tissues to be included in the analysis (n = 13,793).d, As inb (datasetn = 13,793) but for P-site abundance. P-site abundance shows greater variation across tissues than protein abundance (60% of all P-sites show MAD <1 compared with 80% of all proteins; seeb). Again, this may in part be due to technical limitations as P-site quantification is performed on a peptide level and does not benefit from aggregating multiple peptide quantifications into one value for protein quantification.

Source data

a, Scatter plot of Pearson’s correlation coefficients (r) as a measure for co-expression across tissues for all pairs of proteins (x axis) and all pairs of transcripts (y axis) (core dataset only,n = 5,043) along with their marginal histograms. Colours denote the log₁₀-normalized STRING scores of individual gene pairs as a measure of known or predicted direct (physical) or indirect (functional) associations. Strong co-expression of transcripts or proteins or both are more strongly related (physically or functionally) than transcripts and proteins that are not.b, Co-expression analysis of duplicated genes (pairs had to be detected in at least 10 matching tissues to be included in the analysis). The density plots show the distribution of Pearson’s correlation coefficients (r) of co-expressed transcripts (grey) or proteins (blue) for genes that arose by whole-genome duplications (WGD), local duplications or transposon-mediated duplications. Randomly selected gene pairs are shown as a control. Medians are given and displayed as dotted lines. There is substantial co-expression of duplicated genes, indicating that these genes probably have redundant functions.c, Left, protein-level Pearson’s correlation coefficient (r) values (fromb) for all duplicate gene pairs (WGD, local, transposed) plotted against the protein abundance ratio of each pair (average across 30 tissues) (Methods). Blue arrows denote an example of a high or low ratio of protein production for the duplicated genes. Right, example for tissue-resolved protein intensity proportions (top-3) (Methods) for the duplicate pairMAC5A andMAC5B. Irrespective of the tissue,MAC5A is always much higher expressed thanMAC5B. Tissues are coloured as in Fig.1.d, Top, ranked protein abundance ratio for selected duplicate pairs (mean ± s.d.;n = 30) and annotated for phenotypic effects (bottom) in the loss-of-function mutant for either duplicate 1 or duplicate 2 (+). Minus symbols denotes absence of a phenotypic effect. Asymmetric protein production within duplicate pairs can be associated with the occurrence of a phenotype in the loss-of-function mutant of the higher expressed duplicate protein, indicating a dominant functional role of the more highly expressed protein. Blue arrows highlight MAC5A–MAC5B and PHB3–PHB4 as examples.e, Inference of physical protein–protein interactions from co-expression data. Distribution of pairwise Pearson’s correlation coefficients (r) of co-expressed proteins across (at least 10) tissues that are subunits of selected protein complexes.r > 0.5 (shaded in grey) was chosen as a cut-off for the selection of proteins for subsequent analysis to make sure that proteins present in well-characterized protein complexes are retained. CONSTITUTIVE PHOTOMORPHOGENESIS9 SIGNALOSOME (CSN), CELLULOSE SYNTHASE (CESA).f, Recovery of annotated protein–protein interactions by co-expression analysis. Distribution of Pearson’s correlation coefficients (r) of pairs of transcripts (grey) or protein (blue) that are annotated to interact physically in the AtPIN database³³ (pairs had to be detected in at least 10 matching tissues to be included in the analysis). Subsets of the AtPIN database, namely interactions detected by the yeast two-hybrid (Y2H) method, by affinity purification–mass spectrometry (AP–MS) or both.r > 0.5 are shaded in blue (protein). Dotted lines denote median values. Co-expression only recovers a minority of annotated physical interactions andinteractions supported by more than one line of experimental evidence also tend to show stronger co-expression.

Source data

a, Molecular mass (MW) of monomeric proteins (determined from sequence) plotted against the mass determined from the apex of the elution profile for proteins identified by SEC–MS fractions of flower tissue (sFL). Inset shows the molecular mass calibration of the SEC column using a protein calibration standard (mass between 44 and 690 kDa). The distribution of proteins annotated in Araport11 is shown at the top. Many proteins show a much higher apparent molecular mass than would be expected from their sequences (data points above thex = y line). This suggests that these proteins engage in physical protein interactions that are sufficiently stable during SEC separation.b, SEC traces of proteins from five well-characterized protein complexes for flower, leaf and root tissue. Although the resolution of SEC separations is not very high, the complex subunits show very strong co-elution behaviour and the SEC separations of the five complexes are reproducible between tissues. CoA carboxylasen = 4 proteins; CDC48n = 3 proteins; RubisCOn = 4 proteins; prefoldinn = 6 proteins; SCSn = 3 proteins.c, Intensity-normalized SEC elution profile of proteins for flower tissue. Proteins are ordered based on the SEC fraction in which their intensity peaks and the data are displayed as a heat map (n = 2,485 protein traces). Co-eluting proteins were grouped into ‘trace modules’ (Methods). Proteins in trace modules may represent members of protein complexes and thus serve as candidates for further experimental validation.d, To quantify how well protein complexes can be detected using co-expression analysis from data in the tissue atlas (TA) or by SEC–MS, a summary statistic termed ‘complex index’ was calculated (Methods). The complex index is 1 when all subunits of a complex are identified in the same module and no other proteins are contained in the module. Bar plots show examples for complex indices obtained from the different datasets and are divided into large (>4 subunits) and small (≤4 subunits) protein complexes (according to UniProt). Co-expression alone generates many candidates of interactors, but combining co-expression and SEC–MS analysis is an efficient way to prioritize candidates for follow-up experiments.e, Subunit heterogeneity within the coatomer complex. The coatomer complex consists of seven subunits, five of which (α, β, β′, ε and ζ) can be provided by twelve paralogues of these five genes. Plots show the protein proportions of these paralogues in all 30 tissues (data from tissue atlas). The coatomer complex has a similar composition in most tissues. A notable exception is seed tissues, in which production of subunit ζ-1 dominates over the two other paralogous proteins, suggesting that the coatomer complex in seed tissue also preferentially contains the ζ-1 subunit. Tissues are coloured as in Fig.1.f, Absolute SEC intensity traces of individual complex subunits for determining subunit stoichiometry. Examples from left to right: the chaperonin complex (flower, 8 proteins, ratio of all subunits: 1:1), the 26S proteasome core and lid (flower, 14+17 proteins, ratio of all subunits: 1:1), the COP9 signalosome (flower, CSN; 8 proteins, ratio of all subunits: 1:1) and the CESA1–CESA3–CESA6 complex (root, 3 proteins, ratio of all subunits: 1:1). CSN3 and CSN5 were detected both as part of the CSN complex and in monomeric form.g, Top, total intensity of protein complex subunits across all tissues for the complexes shown inf (subunit intensities from the tissue atlas). Middle, relative proportion (mean ± s.d.;n = 30 tissues) of subunits across tissues (Methods). For the CESA complex, ratios were calculated for the subunit combinations CESA1–CESA3–CESA6 and CESA4–CESA7–CESA8. The stoichiometries determined from the tissue expression data are generally well-aligned with the expected 1:1 ratio of subunits in these complexes. As noted inf, a substantial amount of CSN5 was detected as a monomer in the SEC analysis, and the tissue expression atlas also shows higher relative expression of this protein compared with all other complex partners. This suggests that the protein is produced in excess over what is required for the COP9 complex (as observed previously¹²³), and may therefore indicate an additional function within the cell.

Source data

a, Percentage of annotated kinases and phosphatases family members detected at the protein or phosphoprotein level. Parentheses denote the number of genes in each family in theArabidopsis genome.b, Tissue-resolved combined intensity (that is, protein abundance) of families of kinases (left) and phosphatases (right). Tissues are coloured as in Fig.1. Several tissues (notably pollen) stand out in terms of the expression of kinases and phosphatases, which indicates that these tissues are particularly active in phosphorylation-mediated dynamic signalling.c, Top, pie chart of specificity categories for kinases and phosphatases (see Methods for definition). Bottom, distribution of tissue-enhanced kinases and phosphatases across the 30 tissues. Several tissues (such as pollen) stand out in terms of the expression of certain kinases and phosphatases, which indicates tissue-specific signalling.d, Pie charts showing the proportion of proline-directed, acidic, basic and other motif categories for phosphorylated Ser (pS), Thr (pT) and Tyr (pY) residues. Only class I P-sites (localization score > 0.75;Methods) were considered in this analysis.e, Example motif logo plots for motifs such as proline-directed, acidic and basic. P-site motifs were identified using the motif-X algorithm (see Supplementary Table2 for all 266 motifs).n denotes the number of phosphorylation sites that contain the respective motif; ‘fc’ denotes the fold change (that is, enrichment) of the motif in phosphorylated versus unmodified peptides (Methods).f, Enrichment of proline-directed (yellow), acidic (red), basic (blue) and other (grey) sequence motifs (circles) in the serine P-site dataset versus the same motifs detected in the background dataset of unmodified peptides (Methods). Motifs are shown for two, three and four fixed amino acid positions. The P-site in each motif example is underlined. ‘X’ denotes any amino acid.g, Number of identified P-sites for a given protein plotted against the sequence lengths of the same protein. LEA proteins are shown.h, Schematic of the LEA protein sequences (black bars). Pink denotes phosphorylated and blue denotes unphosphorylated STY residues. Almost all STY residues in LEA proteins can be phosphorylated.i, Schematic of the sequences and domain topology of the receptor-like kinases SRF4, FER and CERK1. P-sites often preferentially occur in specific domains, notably the juxtamembrane domain. Protein sequence regions covered by identified peptides are marked in blue, and P-sites are marked in pink.

Source data

a, P-site localization within the structure of RCAR10. The RCAR10 structure (blue) was modelled using the RCAR11 protein crystal structure (cornflower blue) as a template¹¹⁵. ABA-binding loops are shown in turquoise, P-sites in pink, and ABA ligand in yellow.b, RCAR10 expression across tissues at the protein (blue, iBAQ), transcript (grey, TPM) and P-site (pink, intensity) level.c, Tissue-resolved total protein intensity and relative proportions of the members of the PP2C co-receptor family. Seed tissues stand out in terms of overall expression as well as the dominance of AHG1 in these tissues.d, Measurement of the ABA response after expression of RCAR10 or phosphomimetic mutant variants in combination with different PP2C co-receptors in protoplasts (Methods). Columns display the average ABA response (mean ± s.d.,n = 3) and grey dots indicate individual measurements. Co-expression of the phosphatases HAI1–HAI3 leads to similar responses in both phosphomimetic mutants, whereas other co-expressed phosphatases show diverse responses.e, QKY expression across tissues at the protein (blue, iBAQ), transcript (grey, TPM) and P-site (pink, intensity) level.f, Members of the MCTP family clustered by sequence similarity (left) and schematic of their domain structures along with detected P-sites (right). MCTP11a, MCTP12 and MCTP13 were not detected (n.d.) in this study. MCTP15 (also known as QKY) is in bold.g, Number of independent transgenic plant lines (qky-9 mutant background) transformed with wild-type QKY, phosphomutant (S262A, SA; blue) or phosphomimetic (S262E, SE; purple) constructs that show complete, partial or no rescue of the mutant phenotype. qPCR results (mean ± s.d.,n = 3; individual data points as grey dots) show the relative transgene expression in wild-type,qky-9 mutant and selected transgenic lines.h–j, Representative confocal images of six-day-oldqky-9pQKY::mCherry:QKY (WT QKY;n = 14 roots),qky-9pQKY::mCherry:QKY(S262A) (phosphomutant;n = 21 roots),qky-9pQKY::mCherry:QKY(S262E) (phosphomimetic;n = 15 roots) root epidermal cells of the meristematic zone. The punctate signal along the cell circumference shows the expected localization of the QKY protein. Arrows indicate punctate structures. Scale bars, 5 μm.

Source data

This file contains Supplementary Table 1: List of validated peptide spectra for uncertain proteins including mirror plots between experimental and predicted spectra. Supplementary Table 2: Position weight matrix logo plots for all identified serine, threonine and tyrosine phosphorylation motifs.

Supplementary Data 1: This file contains tissue sample names and a description of their growth conditions.

Supplementary Data 2: This file contains the tissue atlas expression values on transcriptome, proteome and p-sites level.

Supplementary Data 3: This file contains a summary of expression evidence for gene isoforms and sORF sequences.

Supplementary Data 4: This file contains the results for 1D GO term enrichment analyses using transcript abundance or protein to RNA ratio (PTR) values as input.

Supplementary Data 5: This file contains a description of feature analysis parameters used in the protein abundance prediction model.

Supplementary Data 6: This file contains protein expression values for selected paralog genes and information about their mutational phenotypes.

Supplementary Data 7: This file contains size exclusion chromatography protein elution profiles for flower, leaf and root tissue.

Supplementary Data 8: This file contains GO term enrichment analysis for WGCNA protein and trace modules and lists gene identifiers in protein and trace module intersections.

Supplementary Data 9: This file contains annotations of selected complexes and their identification in tissue atlas or size-exclusion chromatography experiments.

Supplementary Data 10: This file lists all identified serine, threonine and tyrosine phosphorylation motifs.

Supplementary Data 11: This file contains a list of all primer sequences and gene IDs used in this study.