Plants employ receptor kinases (RKs) and receptor proteins as pattern recognition receptors (PRRs) to sense microbes. To counter plant defense mechanisms, pathogens utilize cytoplasm targeted effector proteins to alter plant cell function and metabolism to aid infection. Many bacterial effectors target PRRs and their signaling components to prevent immune recognition of the pathogen (Jones and Dangl, 2006). Progress into deciphering how effector proteins elicit their response on the plant immune system is being made (Dodds and Rathjen, 2010; Deslandes and Rivas, 2012), with recent evidence indicating effectors converge on central hubs in the plant immune system network through effector redundancy (Mukhtar et al., 2011; Weßling et al., 2014). Effector proteins are useful tools for dissecting plant biology due to their abilities to target key pathways in the host cell. A significant portion of these effector interactions occur at the plasma membrane interface, where the biophysical environment changes due to restriction from three-dimensional space to two dimensions. Proteins anchored in a membrane may display strong interactions but when the cell is homogenized and the membrane is no longer acting to stabilize the interface then the interaction may be lost. As a result, there are likely many interactions that are overlooked by traditional protein–protein interaction screens. Understanding the spectrum of plant immunity components influenced by effector protein interactions is paramount to identifying solutions for engineering plant disease resistance.

The effector AvrPto from the tomato pathogen Pseudomonas syringae pv. tomato (Pst) is known to interact with a number of proteins associated with plant immunity (Scofield et al., 1996; Xiang et al., 2008; Luo et al., 2009). For example, AvrPto binds to the Pto/Prf resistance protein complex that comprises a Pto kinase and the nucleotide-binding leucine rich-repeat protein (NB-LRR) Prf. AvrPto binding to Pto/Prf initiates resistance to Pst by stimulating a hypersensitive response (HR) that results in localized cell death (Ntoukakis et al., 2014). Several studies hypothesize that the plant Pto kinase acts as a decoy for AvrPto binding that stimulates a defensive signaling cascade via Prf to induce localized cellular necrosis (van der Hoorn and Kamoun, 2008; Xiang et al., 2008).

AvrPto is targeted to the plasma membrane of plant cells by post-translational modification of its N-terminus (Met-Gly-X-X-Cys) by myristoylation of Gly-2 (following Met-1 cleavage) and palmitoylation of Cys-5 (Shan et al., 2000; de Vries et al., 2006). The addition of 14-carbon and 16-carbon saturated fatty acid moieties to Gly-2 and Cys-5, respectively, confers stable anchoring of the protein to membranes (Resh, 1999) with myristoylation necessary for AvrPto functionality (Shan et al., 2000). Mechanistic studies of AvrPto do not uniformly agree with its deciphered targets and function (Xiang et al., 2010). There is a general consensus that AvrPto binds to the kinase domains of the PRRs FLAGELLIN SENSING 2 (FLS2) and EF-TU RECEPTOR (EFR) to block signaling after perception of the PAMPs flagellin and EF-Tu, respectively (Xiang et al., 2008). Both FLS2 and EFR interact with, and require, the co-receptor BRI1-ASSOCIATED KINASE 1 (BAK1) (Zipfel et al., 2006; Chinchilla et al., 2007; Heese et al., 2007) to form a complex which is required for PAMP triggered immunity (PTI). Extension of this using yeast two hybrid (Y2H) screening approaches in Arabidopsis thaliana produced a single common protein from two independent studies (cysteine-rich receptor-like protein kinase 45, At4g11890) (Speth et al., 2009; Mukhtar et al., 2011) and a large number of hits that are potentially false positives due to the nuclear localization of the proteins identified.

The plasma membrane is perhaps the most important signaling interface for plant cells. Plants as sessile organisms must transfer endogenous and environmental signals across this barrier to induce a response. Membrane-imbedded proteins whilst extremely important have historically been difficult to study due to low abundance and the poor solubility associated with hydrophobic transmembrane regions. Although in some cases it has been possible to purify plant plasma membrane proteins and their interactors directly and identify them using mass spectrometry (Gutierrez et al., 2010), the most successful approach has been to use forward genetic screens for specific phenotypes, though this is often not possible.

A more promising approach for detecting protein interactions in vivo has recently been developed using biotin ligase (BirA) fusions (Roux et al., 2012). The technique, called BioID, expresses a protein target (the ‘bait’) fused in frame with a mutated form of BirA from Escherichia coli. Using biotin as a substrate, BirA generates a reactive biotinoyl-5′-AMP intermediate which is released and covalently bonds to the nearest primary amine - usually a lysine residue or the N-terminus of a neighboring protein labeling these proximal proteins with biotin (Figure 1A). The lifetime/stability of the biotin analog is very limited such that it only interacts with ‘prey’ proteins in very close proximity to BirA, typically within ≤10 nm radius of the BirA-bait fusion protein (Kim et al., 2014). BioID has proven particularly versatile at monitoring prey proteins that interact with membrane bound proteins in vivo (Couzens et al., 2013).

The BioID system has previously been utilized in rice protoplasts and most recently in stably transformed Arabidopsis plants (Lin et al., 2017; Khan et al., 2018). Here we have developed the BioID proximity protein labeling approach to work in planta and identify plant immune proteins targeted by AvrPto. The system utilizes transient Agrobacterium tumefaciens based transformation of N. benthamiana to express the BirA fusion proteins in the leaf and requires the exogenous application of biotin for efficient labeling. By expressing a range of BirA fusion proteins within N. benthamiana we identify five plant proteins targeted by AvrPto. One protein, NbRIN4, has previously been found to interact with AvrPto and is a central immune regulatory component whilst targeted silencing of the other proteins by viral induced gene silencing (VIGS) reduced the susceptibility of N. benthamiana to infection by the bacterial pathogen P. syringae tabaci, suggesting roles for the BioID-selected proteins in pathogen defense. Thus, BioID provides a useful addition to methods for studying plasma-membrane associated proteins with potential roles in pathogen defense.

To set up the BioID system in plants we modified the binary plant expression vector pT70 (Balmuth and Rathjen, 2007) to express bait proteins fused to BirA with a c-Myc tag. The c-Myc tag allowed for western blotting-based assessment of protein expression levels. In order to probe the interactions of AvrPto we designed four constructs which could differentiate between indiscriminate protein interactions at the plasma membrane versus target specific interactions at this interface (Figure 1C). Firstly, AvrPto was fused to the N-terminus of BirA to produce AvrPto_BirA (i). Secondly the 8 amino acid AvrPto myristoylation/palmitoylation (MP) motif was fused to the N-terminus of the BirA protein to produce MP_BirA (ii), targeting the protein to the plasma membrane (Figure 1B). Thirdly BirA was expressed by itself to differentiate interactions in the cytoplasm (iii) and finally an empty vector (EV) control was utilized to differentiate natively biotinylated proteins and non-specific interactions (iv). The four constructs were transiently expressed in N. benthamiana leaf tissue using A. tumefaciens mediated transformation. Leaf tissue contains a number of natively biotinylated proteins as detected by western blotting using streptavidin-HRP (Figure 1D). Expression of BirA alone in a plant cell results in the labeling of a large number of proteins within the cytoplasm, but significantly less labeling was observed with the BirA fusion constructs.

Retention of AvrPto function when fused to the biotin ligase was confirmed through the use of a hypersensitive response assay. Expression of AvrPto-BirA did not seem to change the function of the protein, as it was still recognized by Pto in N. benthamiana producing a severe HR response and no such response was observed in the absence of Pto (Supplementary Figure S1) (Scofield et al., 1996).

Expression of BirA in leaf tissue produced very little labeling without the addition of exogenous biotin substrate (Figure 2A). To overcome the substrate limitation, solutions with varying biotin concentrations from zero to 150 μM were infiltrated directly into the leaf tissue 24 h prior to sampling. The infiltrated biotin was then taken up by the cells to enable BirA-mediated biotinylation. Plants contain a number of natively biotinylated proteins which are observed in all samples including the empty vector sample where BirA was not expressed. The optimum concentration of biotin to infiltrate into the leaf was found to be 75 μM which produced significant labeling of a large range of proteins within the cytoplasm of the plant leaf cells (Figure 2A). Using a time course for biotin infiltration from 3 to 48 h before harvest of the leaf tissue, we found that 24 h was the optimum time point to produce significant biotinylation, though longer time periods did produce more labeling (Figure 2B). For all subsequent experiments Agrobacterium was first infiltrated into leaf tissue, then 48 h later a 75 μM biotin solution was similarly infiltrated into the same tissue and the leaves were harvested 24 h after biotin infiltration.

We describe here adaptation of the BioID system for rapid detection of proteins proximal to a protein bait in live plant tissue. Utilizing transient Agrobacterium-based gene expression removes the need to construct stably transformed plants. This makes the technique relatively quick and easy to use. The introduction of biotin into the leaf by infiltration also provides a simple way to supply the substrate. Once the transient expression of proteins was optimized and favorable biotin concentrations were determined, the key innovation in adapting BioID for use in live plants was the development of a rapid methodology to remove excess biotin from leaf extracts using a disposable size exclusion column (see section “Materials and Methods” section).

The use of the BioID system for labeling proximal proteins within plant cells was found to be both time and substrate concentration dependent revealing that the BirA enzyme is rate limited by access to the substrate within plant tissues. Increases in biotinylation through increased substrate concentration were observed up to 500 μM, but at the highest biotin concentrations transient expression of the bait protein was inhibited.

We observed that BirA expressed alone labeled more proteins than did the fusion proteins, as observed by western blot (Figure 1D) and by the number of proteins discovered using mass spectrometry (Figure 3A). This can be partly explained by the higher levels of expression observed for BirA. It is also likely due to the fact that the fusion proteins are directed to a specific region within the cell, thus reducing the subset of proteins they can interact with. This is supported by the reduction in the number of proteins labeled when BirA was directed to the plasma membrane (MP_BirA) where it is presumably restricted in the number of proteins available for interaction. Additionally, this restriction was further exaggerated when AvrPto was fused to BirA, with even fewer hits identified (Figures 1D, 3A,B). The EV and BirA constructs produced a significant list of proteins which were excluded from all further analysis. This list of background contaminants will be of use for future experiments to avoid false positives (Supplementary Table S1).

Of the total dataset of biotinylated proteins, we found that the vast majority of the predicted membrane proteins were in the MP_BirA and AvrPto_BirA samples (Figure 3B). This corroborates the evidence that AvrPto is found at the plasma membrane. The lack of overlap observed for the hits from the AvrPto_BirA and MP_BirA constructs (Figure 3A) is consistent with AvrPto_BirA specifically labeling proteins in close proximity to AvrPto and not all proteins at the plasma membrane, confirming the high spatial resolution of the BioID technique (Van Itallie et al., 2013).

Of the five APPs discovered using BioID it was encouraging that RIN4 (APP1) has previously be shown to interact with AvrPto using both Y2H and co-immunoprecipitation techniques (Luo et al., 2009). Additionally, it is worth noting that this protein is also found at the plasma membrane, anchored in place by a C-terminal palmitoylation group (Kim et al., 2005). RIN4 is degraded in the presence of AvrPto and this degradation in Solanaceous plants is dependent on the Pto/Prf resistance protein complex. Additionally, the degradation of RIN4 is reliant on the presence of conserved AvrRpt2 cleavage sites at the N and C-terminus (Ntoukakis et al., 2014). NbRIN4 contains the conserved AvrRpt cleavage sites at the N and C-terminus and has 79% amino acid identity to Solanum lycopersicum RIN4 which has been extensively studied (TC174419) (Luo et al., 2009).

APK1 (APP2) is a member of the leucine rich repeat (LRR) VIII class of LRR-RKs, which have not previously been implicated in immunity. (Monaghan and Zipfel, 2012). APK1 from N. benthamiana is most closely related to the Arabidopsis gene At5g49760 and contains 55% amino acid identity. APK1 is predicted to encode a transit peptide followed by an extracellular region containing 14 leucine rich repeats, a transmembrane domain, and an intracellular protein tyrosine kinase domain. The kinase domain of APK1 has high homology to Pto which is a known target of AvrPto. This suggests that APK1 may also act as a bait for AvrPto (Supplementary Figure S2), although this could not be confirmed by coimmunoprecipitation.

TOM1 (APP3) contains both a VHS domain and a plant like GAT-GGA domain and has 23% homology with the closest human homolog, TOM1-like protein 2 isoform 3 (NP_001076437.1). For human TOM1, the VHS domain is associated with the plasma membrane (Misra et al., 2000) and contains a positively charged phosphatidylinositol 5-phosphate (PtdIns5P) binding motif. The PtdIns5P is responsible for recruiting TOM1 onto signaling endosomes and regulates endosomal maturation (Boal et al., 2015). We found that the positively charged PtdIns5P binding motif is conserved in NbTOM1 (Supplementary Figure S3), suggesting that TOM1 in plants could perform a similar role in PtdIns5P signaling pathways. The pathways so far found to be regulated by PtdIns5P and its binding proteins in mammalian cells include responses to pathogen invasion, stress responses, apoptosis and autophagosome biogenesis (Shisheva, 2013; Vicinanza et al., 2015). NbTOM1 also has a plant like GAT-GGA domain which in mammalian TOM1 is critical for trafficking soluble proteins from the trans-Golgi network to endosomes/lysosomes through interactions with clathrin, GTPases, and TGN-sorting receptors (Shiba et al., 2003). In plants it has recently been shown that clathrin dependent internalization of ligand-activated receptor kinases into the endosomal pathway is required to activate immunity (Mbengue et al., 2016).

The most closely related Arabidopsis gene to APP4 is At4g14200. This protein is documented as being a pentatricopeptide repeat protein (PPR) which in Arabidopsis makes it a member of one of the largest gene families. PPRs have been demonstrated to have a role in plant growth and development. The Arabidopsis PPR protein called PENTATRICOPEPTIDE REPEAT PROTEIN FOR GERMINATION ON NaCl (PGN) has been shown to function in plant defense against a necrotrophic fungus and is involved in abiotic stress tolerance (Laluk et al., 2011). The knockdown of APP4 produced a severely stunted phenotype and inhibited growth of Pstab on N. benthamiana, and the precedent for the involvement of related genes in immunity and stress regulation point to it having a related function.

APK2 (APP5) which is a MAPKKK is most closely related to the Arabidopsis gene At3g22750. At3g22750 was identified in a phosphoproteomic screen of plasma membrane proteins present after treatment with the fungal elicitor xylanase (along with RIN4 in the same study) (Benschop et al., 2007). This protein will be of interest in follow up studies to determine its function in planta.

The upregulation of the APP genes in response to pathogenic bacteria, and even more so to AvrPto (Figure 5), suggests that they are specifically upregulated in response to the effector. This might be expected for genes encoding immune proteins that are targeted by AvrPto. AvrPto did not interact directly interact with APPs using CoIP assays. This may be because BioID is able to detect weak or transient interactions which would normally be missed by CoIP, or because it detects proximity rather than direct interactions per se. Encouragingly, we found that two APPs, APK1, and TOM1, interacted with each other (Figure 4B). This is likely the result of the AvrPto_BirA protein labeling multiple members of a protein complex at the plasma membrane. The fact that APK1 and TOM1 form a complex at the plasma membrane implies that they work cooperatively in the immune response. AvrPto’s tendency to target the kinase domains of receptor like kinases (Xiang et al., 2008) would point to APK1 as the binding target.

The significant decrease in bacterial growth on plants silenced for either TOM1 or APK1 suggests that this protein complex acts to negatively regulate the immune response to Pstab in N. benthamiana. Silencing of TOM1 was found to have a greater effect on inhibition of bacterial growth than APK1, and as such may play a more important role in plant immunity. Something worth following in later work would be to determine if TOM1 has a role in recruitment of the activated receptor-like kinase to the endosomal pathway to facilitate signaling, thus linking the receptor to the clathrin dependent internalization pathway (Mbengue et al., 2016).

Silencing APP4 produced significantly less bacterial growth, which may be evidence for immune related function but could also be biased by the stunted phenotype. The decrease in bacterial growth when APP4 was silenced again suggests negative regulation of the immune response in N. benthamiana.

Comparison of this study with previous uses of the BioID technology for assessing protein–protein interactions reveals some significant differences in approach. The first use of BioID published in plants (Lin et al., 2017) utilized protoplast transformation which simplifies the process of getting the biotin substrate into the cells and also of removing extraneous biotin before affinity capture of biotinylated proteins. The downside of this technique is the difficulty of making protoplasts and the fact that the plant cells are not in their native state which may change protein expression and interactions especially for immune related proteins which may sense cell wall status. The second published study of BioID in plants (Khan et al., 2018) utilized stably transformed Arabidopsis plants and infiltrated biotin into the leaves to achieve biotinylation. This process involves making stably transformed plants which enables biotin labeling to occur in the whole leaf. The major drawback of this technique is the laborious and time-consuming effort required to make stably transformed plants. Khan et al. (2018) also use this system to determine the interactions of an effector protein at the plasma membrane but fail to include the critical control where BirA is targeted to the plasma membrane which would potentially eliminate more false positives. Khan et al. (2018) utilize centrifugal filtration to buffer exchange the sample and remove biotin. We found this to take too long and didn’t remove all biotin and so utilized single use size exclusion columns to remove biotin from the sample before affinity capture of biotinylated proteins. Khan et al. (2018) infiltrated 2 mm biotin into the leaf to promote biotinylation by BirA; we found that at such high concentrations Agrobacterium based transient expression of proteins was diminished and so utilized lower concentrations of biotin.

We have developed a novel strategy to track proteins proximal to a bait protein within living plant tissue. It is particularly amenable for users to determine the interactions between pathogenic effector proteins and plant host molecules. We have used this technique through the AvrPto effector protein to delineate several new proteins which are potentially involved in the plant immune network. Two of the most interesting proteins found in this study were APK1 and TOM1, which form a stable complex with each other at the plasma membrane. Silencing either gene resulted in reduced growth of pathogenic bacteria, suggesting that they act to negatively regulate the plant immune response.

BioID is a powerful new technique with high potential for determining novel plant protein interactions. The ability to tag proximal prey proteins with biotin then directly isolate them at a later time point under denaturing conditions is a major advantage compared to traditional co-immunoprecipitation techniques, which rely on maintaining the interaction during purification. The technique appears to be particularly good for defining membrane protein interactions at the plasma membrane interface. Though the technique has been used here to delineate the plant immune proteins targeted by bacterial pathogens it will be useful for finding a whole range of other protein–protein interactions in plants.

BC and JR designed the study. BC completed the study and wrote the manuscript. TS and JG completed the mass spectrometry. IS provided critical feedback on the manuscript.

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