Arabidopsis G-protein interactome reveals connections to cell wall carbohydrates and morphogenesis.

The heterotrimeric G-protein complex is minimally composed of Gα, Gβ, and Gγ subunits. In the classic scenario, the G-protein complex is the nexus in signaling from the plasma membrane, where the heterotrimeric G-protein associates with heptahelical G-protein-coupled receptors (GPCRs), to cytoplasmic target proteins called effectors. Although a number of effectors are known in metazoans and fungi, none of these are predicted to exist in their canonical forms in plants. To identify ab initio plant G-protein effectors and scaffold proteins, we screened a set of proteins from the G-protein complex using two-hybrid complementation in yeast. After deep and exhaustive interrogation, we detected 544 interactions between 434 proteins, of which 68 highly interconnected proteins form the core G-protein interactome. Within this core, over half of the interactions comprising two-thirds of the nodes were retested and validated as genuine in planta. Co-expression analysis in combination with phenotyping of loss-of-function mutations in a set of core interactome genes revealed a novel role for G-proteins in regulating cell wall modification.

Introduction

Heterotrimeric G-proteins couple a myriad of extracellular signals, including light, ions, peptides hormones, neurotransmitters, and protein ligands, to intracellular changes in ion flux, enzymatic activity, protein proximities, and trafficking. Classically, the G-protein heterotrimer is activated by cell-surface receptors (G-protein-coupled receptors, GPCRs) that trigger the Gα-subunit of the heterotrimer to release GDP, thus enabling the Gα-subunit to bind to GTP. GTP binding is accompanied by structural rearrangements that disengage the Gβγ interaction and result in heterotrimer dissociation (Sprang, 1997). The free subunits then relay signals by interacting with downstream proteins collectively called effectors, because they ‘effect' the cellular changes mentioned above. G-protein signaling is terminated after the Gα-subunit hydrolyzes GTP to GDP and the heterotrimer re-associates. Gα proteins provide specificity between GPCRs and effectors, amplify signal transduction, serve as a point of signal modulation, and act as timing devices that control signaling lifespan.

In metazoans, there are hundreds of GPCRs as well as dozens of effectors, modulators, and scaffold proteins that interact with the heterotrimer or its dissociated subunits (Temple et al, 2010). However, few, if any, of these are encoded in plant genomes sequenced to date (Jones and Assmann, 2004). Because our understanding of G-protein signaling in plants critically depends on knowing the constellation of proteins operating in this network and the relationship of each effector to each other, we chose an ab initio approach to assemble the G-protein interactome. Presumptive roles for G-protein signaling in several new pathways were revealed and one was pursued here.

Results and discussion

Generation of a high-quality G-protein interactome map using yeast two-hybrid screening

Aiming at the identification of regulators and effectors of heterotrimeric G-protein subunits in Arabidopsis thaliana, we performed comprehensive high-throughput yeast two-hybrid (Y2H) screening. Nine prey cDNA libraries made from diverse Arabidopsis tissues were screened 10 times using seven primary baits, including the subunits of the heterotrimeric G-proteins Gα (GPA1), Gβ/Gγ1 (AGB1/AGG1), Gβ/Gγ2 (AGB1/AGG2), (Ullah et al, 2001, 2003; Trusov et al, 2007), and regulator of G-protein signaling 1 (RGS1), Pirin (PRN), N-myc downregulated-like1 (NDL1), and receptor for activated C kinase1A (RACK1A; Chen et al, 2003, 2006; Lapik and Kaufman, 2003; Mudgil et al, 2009). To distinguish between potential effectors that interact with GPA1 in its active conformation only, and regulators interacting with GPA1 irrespective of its activation state, three variants of GPA1 were used as bait constructs: a constitutive active form (Q222L), the wild-type protein, and a mutant version with accelerated GTPase activity (G220A). Potential false positives were excluded both by re-cloning and retesting positive interactions, as well as excluding known common false positives (Methods and Supplementary Experimental Procedures). The complete data set is available in Supplementary Table 1, in Arabidopsis Protein Interaction Database (AtPID, ID:FDS1176001), and in a public database dedicated to Arabidopsis G-protein signaling (AGIdb), http://bioinfolab.unl.edu/AGIdb.

In accordance with published data, our screenings identified AGG1, AGG2, and NDL1 as interactors of AGB1, and thylakoid formation 1 (THF1) as an effector of the constitutive active form of GPA1 (Huang et al, 2006; Mudgil et al, 2009). From the total of 206 unique prey proteins from the first round of screenings, 14 secondary baits (ANNAT1, ARD1, CDC48B, NDL1, RACK1B, RACK1C, SYP23, TGA1, UNE16, VAP27, AT1G05000, AT1G52760, AT4G26410, and AT5G14240) were chosen, because these interacted with two or more of the primary baits. Interestingly, the interaction data reveal a highly interconnected network, with many proteins identified as interaction partners of two or more of the bait proteins. High local connectivity supports the close functional relationship of the proteins involved, and proteins sharing interaction partners have an increased probability of interacting themselves (Milo et al, 2002; Barabasi and Oltvai, 2004). Furthermore, interactions within three- and four-protein-interaction loop motifs are considered reliable and functionally meaningful (Spirin and Mirny, 2003; Wuchty et al, 2003; Yeger-Lotem et al, 2004). Therefore, we focused on the bi-connected core (two-core) of the network composed of 69 proteins interacting with at least two other proteins in the network. Thirty-eight of these proteins were analyzed further by individually testing all possible pair-wise combinations for interaction in yeast. This approach led to the identification of 64 additional interactions, resulting in a highly interconnected core network (68 nodes, 167 edges) with an average node degree of 4.1 (Figure 1). A comparison of the gene ontology annotations of the core network to the genome indicated enrichment in these categories: plasma membrane, nucleus, cytosol, endoplasmic reticulum, ribosome, protein binding, nucleotide binding, structural molecule activity, responses to biotic/abiotic stresses, developmental processes and cell organization, and biogenesis (Supplementary Figure S1). Homologous interacting protein pairs from other organisms (interlogs) were not found.

Figure 1

Arabidopsis heterotrimeric G-protein core interactome. Only those proteins from the interactome data set are shown that possess at least two connections within the network (two-core). Red nodes highlight published components of G-protein signaling in Arabidopsis that were used as primary baits in the Y2H screenings. Proteins used as baits for the second round of screens are shown in brown. Gray lines (edges) represent interactions detected with the Y2H system. Green edges represent interactions shown with both the Y2H system and BiFC. Orange edges represent interactions found in our Y2H screenings that were published and confirmed by FRET analyses or co-immunoprecipitation previously (Lapik and Kaufman, 2003; Adjobo-Hermans et al, 2006; Huang et al, 2006; Mudgil et al, 2009). The AGB1-ARD1 through ARD4 confirmations are as described in Friedman et al (2011). Thick edges indicate protein pairs with significantly correlated expression profiles (Supplementary Table S1).

In planta confirmation of protein–protein interactions

We randomly selected 41 central proteins from the core network and 8 outside the core to test 78 interactions in planta by bimolecular fluorescence complementation (BiFC). In addition, 28 negative controls comprising both soluble and membrane proteins were included. Seventy-four of the 78 tested protein pairs, but none of the negative control pairs, complemented yellow fluorescent protein (YFP) fluorescence, indicating in planta interaction, thereby corroborating the Y2H data. The test pairs were selected after passing stringent experimental and topological filters (Supplementary Experimental Procedures). Our validation rate is higher than expected given the previously observed detection overlap between Y2H and orthogonal validation assays (Braun et al, 2009). This is likely due to the applied filtering of validated interactions, although formally it cannot be excluded that use of a larger and more diverse set of random negative controls would detect a higher background and hence necessitate more stringent scoring. The majority of the BiFC signals were present in the cytoplasm and/or the cell periphery of epidermal pavement cells. Notably, some tested interactions seem to be restricted to the nucleus in planta, e.g., MYC2-ARD1, AGB1-MYB, and UNE16-MYB, while others preferentially take place in guard cells, e.g., RGS1-TGA1 and ARD1-EXGT-A1 (Supplementary Figure S2).

Correlation of gene expression supports the reliability of the G-protein interactome map

Correlated expression frequently is an indicator of co-functionality of genes in common pathways and processes (Persson et al, 2005; Hirai et al, 2007; Humphry et al, 2010), and interacting proteins, particularly proteins that are part of network motifs, are often significantly co-expressed (Bhardwaj and Lu, 2009). We therefore interrogated correlation of expression of the gene pairs derived from our interaction analyses. To establish a threshold of significance for this type of analysis, 69 gene pairs encoding proteins well documented to physically interact were analyzed (‘gold standard', Supplementary Table S1). The thresholds were established using permutation testing to account for multiple testing practices. The expression profiles of the gold standard protein pairs were significantly correlated as compared with a random selection. We then calculated correlation of gene expression for 385 gene pairs from the G-protein interactome across the four GENEVESTIGATOR categories: Anatomy, Development, Mutation, and Stimulus. In total, 90, 15, 178, and 196 of the 385 tested pairs across the four respective categories showed a significant correlation of gene expression (Figure 1, Supplementary Table S1, and AGIdb, http://bioinfolab.unl.edu/AGIdb). Overall, the average expression correlation of the gene pairs was significantly higher than the reliability of the G-protein interaction network made up of the 69 previously documented gene pairs.

The core interactome points to a previously unknown role for G-proteins in cell wall composition

The interactome provided many new leads to test a role of G-proteins in Arabidopsis cell function, some of which are known while many are new. For example, manual inspection of the interactome revealed several potentially interesting interactions between the G-protein core and proteins that may either directly (e.g., enzymes) or indirectly (e.g., transcription factors) regulate cell wall composition or structure (Supplementary Table S1). Among the cell wall-associated enzymes in this group, the predominant candidates have known or predicted functions toward xylose biosynthesis or metabolism. Xylose is present in plant cell walls primarily in two polysaccharides, xyloglucan and xylan, both of which are found in Arabidopsis tissues. Therefore, we hypothesized that the cells lacking a functional G-protein complex would have an altered xyloglucan and/or xylan content or structure.

We first tested for alterations in cell wall composition across a selection of G-protein mutants using infrared microscopy, a rapid method to detect global changes in cell wall composition (McCann et al, 2007), followed by principal component (PC) analysis. Mutants tested included agb1, gpa1, rgs1, suppressor of G-beta1 (SGB1), and a double mutant defective in both GPA1 and AGB1. The gpa1–3, rgs1–2, and sgb1–2 single mutants could not easily be discriminated from Col-0 wild type. However, the agb1–2 mutant and the gpa1–4 agb1–2 double mutant could be discriminated from Col-0 at 80% correct assignment using four PCs for the double mutant, and 78% using five PCs for the agb1 single mutant (Supplementary Figure S3A–D). The loading for PC2 indicates differences in cell wall ester, protein content, and some carbohydrate features between Col-0 and agb1.

As infrared microscopy does not allow unambiguous identification of the altered carbohydrates, we next analyzed wild-type and mutant cell walls for monosaccharide content via gas–liquid chromatography. Small but significant differences in the mole % of one of the major cell wall pentoses, xylose, were observed in both the agb1 mutant and gpa1 agb1 double mutant compared with wild type. While the agb1 single mutant exhibited a slight but statistically significant decrease in xylose mole %, the gpa1 agb1 double mutant showed a statistically significant increase in the proportion of cell wall xylose (Supplementary Table S1). The opposing phenotypes suggest that the action is primarily through GPA1, as loss of AGB1 would increase the activated pool and loss of GPA1 would decrease it (Ullah et al, 2003). Specific increases or decreases in xylose content alone suggest alterations in either xylan or xyloglucan structures.

To determine more precisely the identity of the polysaccharide affected by the mutations, and to gain a broader picture of cell wall changes in the mutant plants, cell walls of wild-type and mutant plants were quantitatively profiled for their glycan epitope content (glycome profiling). Glycome profiling involves enzyme-linked immunosorbent assay-based screening of sequential extracts of cell walls, using a set of ∼150 cell wall glycan-directed monoclonal antibodies that resolve into 18 groups recognizing diverse epitopes present on most major classes of plant cell wall polysaccharides (Pattathil et al, 2010). Leaf and root tissues were isolated from gpa1 and agb1 single mutants, the gpa1 agb1 double mutant, the gpa1 agb1 agg1 agg2 quadruple mutant, and wild-type plants. The glycome profiles of leaf and root cell walls from mutants showed some differences when compared with each other and to the profiles from wild-type cell walls. The altered glycome profiles were particularly evident in the case of the leaf cell wall extracts of the mutants, where the binding patterns of the xylan-3 and xylan-4 groups of antibodies varied in all of the mutants (Figure 2A and Supplementary Table S1 and Supplementary Figure S3E). In the gpa1 agb1 double mutant and gpa1 single mutant, a reduced abundance of xylan-3 (P=7.2e−12 in the double mutant and P=1.8e−14 in the single mutant) and xylan-4 (P=1.8e−5 in the double mutant and P=1.1e−5 in the single mutant) epitopes was observed in the chlorite fraction in comparison with the corresponding fraction from wild-type cell walls (Figure 2A). Analyses were carried out using ANOVA models on each antibody classification group (Supplementary Experimental Procedures), and the results are provided in Supplementary Table S1. Additionally, the chlorite fractions from the leaf cell walls of gpa1 and gpa1 agb1 agg1 agg2 quadruple mutants contained reduced levels of xylan-3 and xylan-4 epitopes. Subtle changes in the glycome profiles for some mutant cell walls were also noted in the binding patterns of antibodies against pectic backbone and xyloglucan-directed antibodies (Figure 2A and Supplementary Figure S3E). Overall, these results indicate that structural changes in the leaf cell walls of mutants were mostly related to the structural features of xylan or to the integration of xylan into the cell walls during its biosynthesis.

Figure 2

Glycome profiling of sequential cell wall extracts prepared from leaves (A) and roots (B) of the agb1, gpa1 mutant and wild-type (Columbia-0, Col-0) plants. Heat maps of enzyme-linked immunosorbent assay (ELISA) data obtained by screening the sequential carbohydrate extracts with 150 glycan-directed monoclonal antibodies (Pattathil et al, 2010). The panel on the right lists the array of monoclonal antibodies used (left-hand side) and groups them according to the principal cell wall glycan (right-hand side) recognized by the antibodies (Pattathil et al, 2010). The reagents used for various extraction steps are identified at the bottom of each column in the heat map. The yellow-black scale indicates the strength of the ELISA signal: bright yellow depicts strongest binding and black indicates no binding. The colored outlines highlight the changes in the antibody binding patterns to the extracts of the mutants compared with WT. (A) The red box outlines the changes in the xylan-3 and xylan-4 antibodies, blue box outlines the changes in the pectic backbone antibodies, and green box outlines the changes in xyloglucan-directed antibodies. (B) Green box outlines the changes in the pectic backbone antibodies, red box outlines the changes in the RG-1/AG antibodies, and the blue box outlines the changes in the AG-1 and AG-2 antibodies. A full data set of the glycome profiles of all mutants can be found in Supplementary Figures S2 and S3.

The mutants also showed differences in their root cell wall glycome profiles compared with that of wild-type plants, although the dissimilarities were in different groups of antibodies than those observed in the leaf glycome profiles. Invariably, in all mutants, there was a notable reduction in the levels of pectic backbone epitopes (homogalacturonan, rhamnogalacturonan I backbone epitopes) in the oxalate and carbonate extracts compared with the corresponding extracts from wild-type walls. There was also a subtle reduction in the release of pectic rhamnogalacturonan I/arabinogalactan (RG-I/AG) epitopes in the 1 M KOH fractions of the mutants. Additionally, altered binding patterns of the AG-1 and AG-2 groups of antibodies to the oxalate and carbonate fractions of mutants was observed (Figure 2B and Supplementary Figure S3F). These data suggest that the mutations in the genes studied here also affect root cell wall structure, primarily by altering the extractability of pectic polysaccharides.

In summary, the glycome profiles of the mutant wall extracts indicate that mutations in heterotrimeric G-protein complex components lead to alterations of the overall cell wall structure, thereby resulting in changes in the extractability of primarily xylan epitopes in leaf tissues and pectic backbone epitopes in root tissues.

Mutants of G-protein interactors share a stomatal density phenotype

Alterations in cell wall composition can confer anomalous morphological features and one quantifiable anomaly is the stomatal pattern of the epidermis (Chen et al, 2009; Guseman et al, 2010). Previous genetic analysis revealed that GPA1 and AGB1 antagonistically modulate stomatal density in Arabidopsis (Zhang et al, 2008). To assess whether any of the newly identified members of the G-protein interactome are co-involved in this process of aberrant morphogenesis, we quantified stomatal density at two different time points in seedlings of 17 plant lines carrying mutations in genes encoding proteins at nodes of the core interactome and compared them to Col-0 wild type. Our data confirmed the previously reported lower and higher stomatal density of gpa1 and agb1 single mutants, respectively, as well as the phenotype of the gpa1 agb1 double mutant, which was reported to be intermediate between the agb1 mutant and Col-0 wild type (Zhang et al, 2008). Five mutants showed either significantly (P<0.1) elevated or reduced stomatal density (Supplementary Table S1). These mutants relate to a diverse set of genes with yet unknown links to cell wall modification and/or morphogenesis.

Conclusions

We established a comprehensive protein–protein interaction network for Arabidopsis G-protein signaling pathway elements, greatly expanding our view of G-protein-coupled signaling. Interactome networks derived from high-throughput interactome projects have proven highly valuable for systems biology and the functional annotation of the genomes of a number of model organisms (Gavin et al, 2002; Li et al, 2004; Formstecher et al, 2005; Rual et al, 2005). An unbiased, partial, plant interactome data set (8000 by 8000 matrix) was published during review of the present work (Arabidopsis Interactome Mapping Consortium, 2011). In Arabidopsis, systematic protein interaction mapping projects focused on pre-selected groups of genes with specific structural or functional relation, such as particular transcription factor families, membrane proteins, virulence effectors, or proteins involved in cell cycle regulation (Zimmermann et al, 2004; de Folter et al, 2005; Lalonde et al, 2010; Van Leene et al, 2010; Mukhtar et al, 2011). Our focused interactome revealed G-protein interacting proteins with different subcellular localization and cell type specificity. A role for the G-protein interactome in cell wall biogenesis/metabolism was deduced from the predicted functions of identified interactors, the biochemical and immunological phenotype of G-protein mutants, and the morphological phenotype of mutants defective in G-protein interactors.

Materials and methods

Y2H methods

All bait constructs used in the primary and secondary screenings were tested for auto-activation, and the addition of 3-amino-1,2,4-triazole eliminated residual growth on selection media. To exclude potential screening artifacts (‘technical false positives'; Finley, 2007; Venkatesan et al, 2009), in vivo recombination cloning of PCR products was used to re-clone the prey cDNAs in yeast, and interactions were pair wise reassessed with their respective baits as well as with negative controls. Only those prey clones that were positive with the bait and negative in the control experiment were taken into account. Preys representing known Y2H artifacts were discarded (Supplementary Experimental Procedures). Further details of the methods used for the interaction screens, including the nine cDNA libraries interrogated, are provided in Supplementary Experimental Procedures.

Arabidopsis G-Signaling Interactome Database

The protein interaction information that we obtained is publicly available through the Web-based database: Arabidopsis G-Signaling Interactome Database (AGIdb, http://bioinfolab.unl.edu/AGIdb). The AGIdb currently includes information from 544 unique protein pairs (currently 1058 interaction data obtained from different libraries and other experimental conditions) and allows users to access and search the entire data set. The interaction data can be downloaded in a table format that can be imported in Cytoscape (http://www.cytoscape.org/; Cline et al, 2007) and other network analysis software. Details of the specifications of the database are provided in Supplementary Experimental Procedures.

Genes that show correlated expression may share cis-acting elements in their promoters (5′ regulatory regions). Known cis-acting elements for each gene in the interactome were retrieved via a dedicated database (AtcisDB, http://arabidopsis.med.ohio-state.edu/AtcisDB/) and added to the interactome database. Comparative analysis revealed sets of shared cis-regulatory elements in case of some co-expressed interactome genes.

Bimolecular fluorescence complementation

BiFC was performed as described in Grigston et al (2008) with modifications as indicated in Supplementary Experimental Procedures. The higher extinction coefficient mediated by these second-generation BiFC vectors enabled analyses at lower expression levels, thus greatly lowering the false-positive rate. A positive-transformation control was included. Leaf samples were imaged using a Zeiss LSM710 confocal laser scanning microscope equipped with an Apochromat × 40 (NA 1.2) water-immersion objective. The positive control for BiFC interaction used in each replicate was the AHP2 dimer (At3g29350). Twenty-eight negative control tests for core proteins against both membrane and soluble proteins were included (Supplementary Figure S1, panels 79–109).

Co-expression analyses

A detailed description of the methods used to determine the statistical significance of paired gene expression patterns is provided in the Supplementary Experimental Procedures material.

Cell wall analyses

Isolation of cell walls, FITR analysis, monosaccharide and linkage analyses, glycome profiling, and the statistical analyses used for glycome profiling are described in detail in the Supplementary Experimental Procedures material.