Advertisement

Cytosolic HSP90 associates with and modulates the Arabidopsis RPM1 disease resistance protein

David A. Hubert, Pablo Tornero, Youssef Belkhadir, Priti Krishna, Akira Takahashi, Ken Shirasu, Jeffery L. Dangl

Author Affiliations

  1. David A. Hubert1,
  2. Pablo Tornero1,2,
  3. Youssef Belkhadir1,
  4. Priti Krishna3,
  5. Akira Takahashi4,
  6. Ken Shirasu4 and
  7. Jeffery L. Dangl*,1,5
  1. 1 Department of Biology, University of North Carolina at Chapel Hill, Chapel Hill, NC, 27599, USA
  2. 2 Present address: IBMCP, Universidad Politecnica de Valencia, Camino de Vera s/n, 46022, Valencia, Spain
  3. 3 Department of Biology, University of Western Ontario, London, ON, N6A 5B7, Canada
  4. 4 Sainsbury Laboratory, John Innes Centre, Colney, Norwich, NR4 7UH, UK
  5. 5 Curriculum in Genetics and Department of Microbiology and Immunology, Coker Hall, Room 108, CB 3280, University of North Carolina at Chapel Hill, Chapel Hill, NC, 27599, USA
  1. *Corresponding author. E-mail: dangl{at}email.unc.edu
  1. D.A.Hubert, P.Tornero and Y.Belkhadir contributed equally to this work

Abstract

The Arabidopsis protein RPM1 activates disease resistance in response to Pseudomonas syringae proteins targeted to the inside of the host cell via the bacterial type III delivery system. We demonstrate that specific mutations in the ATP‐binding domain of a single Arabidopsis cytosolic HSP90 isoform compromise RPM1 function. These mutations do not affect the function of related disease resistance proteins. RPM1 associates with HSP90 in plant cells. The Arabidopsis proteins RAR1 and SGT1 are required for the action of many R proteins, and display some structural similarity to HSP90 co‐chaperones. Each associates with HSP90 in plant cells. Our data suggest that (i) RPM1 is an HSP90 client protein; and (ii) RAR1 and SGT1 may function independently as HSP90 cofactors. Dynamic interactions among these proteins can regulate RPM1 stability and function, perhaps similarly to the formation and regulation of animal steroid receptor complexes.

Introduction

Our understanding of disease resistance specificity in plants centers on the structure and function of pathogen‐specific Resistance (R) gene products. R proteins confer resistance to pathogen strains expressing a molecule that specifically triggers its action (Dangl and Jones, 2001). The largest class of R protein contains a nucleotide‐binding site (NB) and leucine‐rich repeats (LRRs), and are termed NB‐LRRs (Dangl and Jones, 2001). The repertoire of R proteins, though deployed with broad population polymorphism, may still not be sufficiently diverse to mediate direct recognition of all relevant pathogens. The question of repertoire size, among others, drove the formulation of the ‘guard hypothesis’. Here, pathogen molecules that trigger R action are most easily thought of as virulence factors whose presence is sensed by the host cell. Experimental evidence supports this model, though it is not yet fully generalizable (Dangl and Jones, 2001; Holt et al., 2003). NB‐LRR activation may include a large conformational change, perhaps akin to the ‘jackknife model’ leading to proximity‐induced activation as proposed for Apaf‐1 (Moffett et al., 2002; Hwang and Williamson, 2003). This, in turn, leads to a series of cellular events that collectively form the defense response (Holt et al., 2003). It is unclear what portion of this diverse defense response is actually required to halt pathogen growth.

Arabidopsis RPM1 is an NB‐LRR protein that confers recognition to bacterial strains expressing either of two divergent type III effector genes, avrRpm1 or avrB (Grant et al., 1995). These two type III effector proteins have virulence function on hosts lacking RPM1 (rpm1; disease susceptible; Ashfield et al., 1995; Ritter and Dangl, 1995). Recognition of AvrRpm1 or AvrB by RPM1 may be the consequence of their action on Arabidopsis RIN4 (RPM1‐interacting protein 4). RIN4 is a protein of unknown function and is phosphorylated in response to the presence of either AvrRpm1 or AvrB (Mackey et al., 2002). RIN4 can interact with RPM1, AvrRpm1 and AvrB in vivo. All four of these proteins localize to the plasma membrane (Boyes et al., 1998; Nimchuk et al., 2000; Mackey et al., 2002).

A limited set of genetically defined proteins are broadly required for the action of Arabidopsis R gene subsets (Dangl and Jones, 2001). The ndr1 mutation compromises the function of a subset of NB‐LRR R proteins (Century et al., 1997). NDR1 is a putative glycosylphosphatidylinositol (GPI)‐anchored protein (B.Staskawicz, personal communication). Mutations in RAR1 and SGT1b, one of two SGT1 orthologs in Arabidopsis, compromise the function of many R proteins (Azevedo et al., 2002; Muskett et al., 2002; Tör et al., 2002; Tornero et al., 2002b). Plant SGT1 proteins share similarity with the yeast SGT1 protein, a regulator of the SCF ubiquitin ligase complexes in a variety of cellular processes (Kitigawa et al., 1999). RAR1 and SGT1 can interact in vivo, and presumably function together (Azevedo et al., 2002).

RPM1 function is compromised by ndr1 and by rar1 but, surprisingly, is not compromised by either sgt1a or sgt1b. Because an sgt1a/sgt1b double mutant is lethal (A.Takahashi and K.Shirasu, in preparation), overlapping contributions of these genes to RPM1 function cannot be determined. We performed a large‐scale screen for loss of RPM1‐mediated hypersensitive cell death (hypersensitive response; HR) in response to conditional expression of an avrRpm1 transgene (Tornero et al., 2002a). We describe here four mutant alleles of the Arabidopsis HSP90.2 gene that caused loss of RPM1‐specified HR and disease resistance in that screen. This represents the first phenotype attributed to mutation of a plant cytosolic HSP90.

We demonstrate that RPM1 is the first client protein described for plant cytosolic HSP90, using as criteria (i) in vivo HSP90–RPM1 interaction; (ii) modulation of RPM1 function by particular mutant alleles of the HSP90.2 isoform; and (iii) greatly decreased steady‐state RPM1 levels in these hsp90.2 mutant backgrounds. We describe genetic interactions between HSP90.2 and both RPM1 and NDR1. We provide evidence for in vivo interactions between HSP90s and both RAR1 and SGT1. Thus, we describe a possible mechanism by which RAR1 and SGT1 affect disease resistance protein signaling through probable cofactor interactions with cytosolic HSP90s.

Results

lra2 mutations specifically affect RPM1‐mediated pathogen recognition

We screened ∼500 000 ethyl methanesulfonate (EMS)‐mutagenized Arabidopsis M2 individuals for mutants affecting recognition of avrRpm1. Among others, we identified four allelic mutations that we called lra2‐1, lra2‐2, lra2‐3 and lra2‐4 (lra2; loss of recognition of avrRpm1; Tornero et al., 2002a). Two of these, independently isolated, were later found to carry the same mutation.

Both the lra2‐2 and lra2‐3 alleles have an intermediate effect on RPM1 function (as did the other two alleles, data not shown) measured by pathogen growth and disease symptoms following challenge with Pseudomonas syringae pv tomato (Pst) strain DC3000 expressing either avrRpm1 or avrB (Figure 1; data not shown). We did not see a significant effect on other R genes active against different type III effectors from P.syringae (Figure 1). RPM1‐mediated HR was also altered, but not completely abolished, in the lra2 mutants in response to infiltration with Pst DC3000(avrRpm1). This is similar to the effect of rar1 on RPM1‐mediated HR (Tornero et al., 2002b). While normal RPM1‐mediated HR occurs 5–8 h after inoculation, we observed a low frequency of HR on leaves from all lra2 alleles by 20 h (data not shown). The effect of lra2 mutations on the HR was also specific to RPM1.

Figure 1.

Mutations in lra2 specifically affect RPM1 signaling. Growth of Pst DC3000 containing the indicated avirulence genes in lra2 mutants and corresponding controls (used throughout). The a11 line is the Col‐0 parent containing the estradiol‐inducible avrRpm1 transgene (Tornero et al., 2002a). The a11;rpm1‐1 line is the a11 transgene crossed into an isogenic rpm1‐1 background. Bacterial numbers here and in Figures 2 and 3 are expressed as the log10 of c.f.u/mg fresh weight (f.w.) (Tornero and Dangl, 2001). Error bars indicate ±SE. Growth of Pst DC3000(vector) did not show a significant difference in growth in lra2 mutants. Pst DC3000(avrRpm1) and Pst DC3000(avrB) exhibited consistent increased growth in lra2 mutants intermediate to growth observed in rpm1 mutants in four independent experiments.

Basal resistance against virulent pathogens (Pst DC3000) was not significantly affected (Figure 1). We also observed no alteration in the responses of lra2 alleles to infection with a series of Peronospora parasitica isolates (Holub et al., 1994). This included isolates that were either specifically recognized by various R genes in the lra2 parental background [Cala2 (recognized by RPP1a), Emoy2 (RPP4), Emwa1 (RPP4) and Hiks1 (RPP7)] or caused downy mildew disease (Emco5 and Noco2) (data not shown).

We observed that lra2‐2 and lra2‐4 displayed a partial penetrance phenotype for both HR and onset of disease symptoms. After inoculation with low doses of Pst DC3000(avrRpm1), lra2‐2 and lra2‐4 plants either developed symptoms characteristic of disease (in 0–80% of plants) or were completely asymptomatic (data not shown). Progeny from self‐fertilization of either symptomatic or asymptomatic individuals displayed similar variable penetrance in the next generation (data not shown). The partially penetrant disease phenotype did not influence the standard deviation in our bacterial growth assays (see Figure 1).

LRA2 and RPM1 interact genetically

We made test cross F1s between rpm1 and all lra2 alleles and assayed two phenotypes for allelism: the HR resulting from estradiol induction of the avrRpm1 transgene contained in these lines (Tornero et al., 2002a) and bacterial symptoms resulting from Pst DC3000(avrRpm1) infection. Both of these assays initially suggested that lra2 was allelic to rpm1 in the F1 generation. However, F2 progeny from these F1s contained a large percentage of phenotypically wild‐type individuals. This is inconsistent with allelism, and is consistent with non‐allelic non‐complementation. In this condition, F1 individuals of a cross between two unlinked recessive mutants (each mutation thus heterozygous) display a phenotype similar to either homozygous single mutant. This often indicates that the two genes act together, that the protein products physically interact or are part of the same protein complex, or that half the wild‐type dose of the two, in combination, is insufficient for wild‐type function (e.g. Belanger et al., 1994; Larkin et al., 1999).

We assayed HR response and bacterial growth in the lra2 alleles and in various F1 progeny (Figure 2) to address the apparent genetic interaction between LRA2 and RPM1. The wild‐type parental plant line, containing the estradiol‐inducible avrRpm1 transgene (called a11; Tornero et al., 2002a) responded to estradiol with a strong HR (Figure 2A). The isogenic a11;rpm1‐1 control (Tornero et al., 2002a) and the tested lra2 alleles exhibited no HR. The obvious induction of HR in F1 progeny of backcrosses to a11 demonstrated that both rpm1 and the tested lra2 alleles were recessive in this assay. Strikingly, F1 progeny of (a11;rpm1‐1 × lra2‐2 or × lra2‐3) did not respond to estradiol (Figure 2A). Furthermore, lra2‐2 and lra2‐3 partially compromised RPM1 inhibition of bacterial growth, and were fully recessive for this phenotype when assayed as F1s backcrossed to the a11 parental line (Figure 2B). The F1 progeny of (a11;rpm1‐1 × lra2‐x) exhibited modest, but reproducible, reduction of RPM1 function. These plants allowed bacterial growth between that of the lra2 parent and the (a11 × a11;rpm1‐1) F1 control. This genetic interaction is specific to lra2 and rpm1, as we did not observe non‐allelic non‐complementation in other trans‐heterozygous combinations of mutants affecting the RPM1 pathway tested (tested were rar1 × rpm1, ndr1 × rpm1, rar1 × ndr1, rar1 × lra2 and ndr1 × lra2; data not shown). The data in Figure 2 strongly support the conclusion that rpm1 and lra2 exhibit non‐allelic non‐complementation. Our data thus suggest that the respective wild‐type RPM1 and LRA2 proteins work in the same pathway, and potentially interact physically.

Figure 2.

LRA2 and RPM1 interact genetically. (A) Trypan blue staining of an HR assay following estradiol induction of avrRpm1 expression. Two‐week‐old plants were treated with 10 μM estradiol with 0.02% silwet, and stained with trypan blue 2 days later. Row 1 displays parental responses to conditional expression of avrRpm1. Row 2 displays that the lra2 and rpm1 mutants are recessive. Row 3 shows that rpm1;lra2 trans‐heterozygotes do not express HR. Three independent repetitions were performed. (B) Growth of Pst DC3000(avrRpm1) in the same genotypes as (A). Error bars indicate ±SE of triplicates from this experiment. The experiment was performed six times with similar results.

ndr1 and lra2 display both synergistic and epistatic interactions

Like lra2, ndr1 supported an intermediate level of bacterial growth when challenged with Pst DC3000(avrRpm1) (Figure 3). Additionally, the RPM1‐mediated HR is severely attenuated in lra2 (see above), but not compromised in ndr1 (Century et al., 1995; Tornero et al., 2002b). These incomplete phenotypic effects allowed us to assay for genetic interaction between lra2 and ndr1. We constructed lra2;ndr1 double mutants and tested them for bacterial growth and HR. The lra2;ndr1 double mutants were completely compromised for RPM1 function, allowing as much pathogen growth as an rpm1 mutant following application of Pst DC3000(avrRpm1) (Figure 3; note log scale). The lra2;ndr1 double mutant also displayed full loss of Pst DC3000(avrRpm1)‐triggered HR, suggesting that LRA2 is required for the RPM1‐dependent HR remaining in ndr1 mutants. Note that while the ndr1 allele is a null (Century et al., 1997), the lra2 mutants presented here are not (see below).

Figure 3.

lra2 and ndr1 affect RPM1 function synergistically. Growth of Pst DC3000(avrRpm1) in lra2;ndr1 double mutants with corresponding controls, as in Figures 1 and 2. Both lra2 and ndr1 single mutants exhibit intermediate growth, while the double mutant exhibits complete susceptibility.

LRA2 is HSP90.2

We cloned LRA2 based on its map position. Our mapping population of disease‐susceptible (lra2;lra2) individuals narrowed the LRA2 interval to an ∼52 kb region on the bottom arm of chromosome V (see Materials and methods and Supplementary figure 1, available at The EMBO Journal Online). We sequenced candidate genes from this interval in the lra2‐1 mutant and found a G/A transition at position 21 937 (nucleotide positions relative to the published sequence of P1 clone MDA7; GenBank accession No. AB011476). This created a G95E mutation in the cytosolic HSP90.2 (At5g56030; Figure 4A). We sequenced this gene from the other three independently isolated lra2 alleles and found mutations in lra2‐2 (C21952T; S100F, independently isolated in lra2‐4) and in lra2‐3 (G21785A; D80N) (Figure 4A). To avoid confusion and to follow accepted nomenclature conventions, we re‐designate the lra2 alleles hsp90.2‐1, hsp90.2‐2 and hsp90.2‐3 for lra2‐1, lra2‐2 and lra2‐3, respectively.

Figure 4.

Clustering of hsp90.2 mutations in the highly conserved N‐terminal ATPase domain. (A) Alignment of the N‐terminal ATPase domains of HSP90 orthologs across diverse kingdoms, highlighting the residues mutated in our hsp90.2 alleles in red. Proteins compared: AtHSP90.2 (swissprot id: HS82_ARATH), ScHSP82 (swissprot id: HS82_YEAST), DmHSP83 (swissprot id: HS83_DROME), HsHSP90a (swissprot id: HS9A_HUMAN) and EcHptG (swissprot id: HTPG_ECOLI). (B) Hsp90.2 threaded over the structure of yeast HSP90 bound to ADP (center space filling structure) viewed from two different angles. Residues that make direct interactions with ADP are in violet. The three residues mutated in our hsp90.2 alleles are in yellow. Note that D80 makes direct contacts with ADP, and is mutated in hsp90.2‐3.

All mutations were within the conserved ATPase domain of this HSP90 (Figure 4A). The hsp90.2‐3 change (D80N; yellow in Figure 4B) alters a residue previously shown to make multiple ATP contacts in the crystal structure of yeast HSP90 (Prodromou et al., 1997). The hsp90.2‐1 (G95E) and hsp90.2‐2 (S100F) changes are both adjacent to residues that make direct contact with ATP (N93 and R99). The G95E change should alter the local charge density, while the S100F change results in addition of a large hydrophobic side chain. We used molecular markers based on the mutation to outcross hsp90.2‐3 from the conditional avrRpm1 expression transgenes (Materials and methods). This line expressed the same significant reduction of RPM1 function as its parent line, measured by both disease symptoms (Figure 5A, compare with Figure 1) and pathogen growth (Figure 5B). Thus, the transgenes carrying the estradiol‐inducible avrRpm1 system have no effect on the mutant phenotype. Finally, an insertion allele, hsp90.2‐5 (a T‐DNA insertion at approximately nucleotide 23 453, amino acid position 601 of 699, SiGNAL line SALK_058553; Supplementary figure 1) was viable and exhibited no alteration of RPM1‐mediated resistance (Figure 5B). We did not detect a truncated form by western blot (data not shown), and assume that this allele is a null. We infer that one of the other three highly homologous cytosolic HSP90s compensates for the loss of HSP90.2 (Borkovich et al., 1989). HSP90.2 is constitutively expressed, especially in flower structures and roots (Yabe et al., 1994). We observed only very modest pleiotropic developmental changes in the hsp90.2 mutants (Queitsch et al., 2002) (Supplementary figure 2). Thus, the hsp90.2 alleles identified in our screen are rare, specific, and not compensated for by other HSP90 isoforms.

Figure 5.

RPM1 function is not compromised by an hsp90.2 insertion allele, and the hsp90.2‐3 phenotype is independent of the conditional avrRpm1 expression system. (A) Bacterial growth assay with Pst DC3000(avrRpm1). The hsp90.2‐5 (Salk 058553) T‐DNA insertion allele displays a wild‐type response to Pst DC3000(avrRpm1). (B) In contrast, the hsp90.2‐3 allele outcrossed from the transgenic, conditional avrRpm1 expression system is still compromised for RPM1 function. This line, and controls, were spray infected with Pst DC3000(avrRpm1), and still exhibit symptoms intermediate to a full loss‐of‐function rpm1 mutant.

HSP90 associates with RPM1 in vivo

An RPM1‐myc epitope‐tagged protein is a peripheral plasma membrane protein (Boyes et al., 1998), and fails to accumulate in rar1 plants (Tornero et al., 2002b). We crossed an RPM1‐myc transgene (Boyes et al., 1998) into both the fully penetrant hsp90.2‐3 allele and the partially penetrant hsp90.2‐2 allele. Both alleles exhibited greatly decreased RPM1‐myc levels compared with wild‐type plants (Figure 6A). Curiously, hsp90.2‐2 consistently accumulated less RPM1‐myc than hsp90.2‐3. This suggests that the partial penetrance phenotype of hsp90.2‐2 is not simply correlated to RPM1 levels.

Figure 6.

HSP90 interacts with RPM1. (A) hsp90.2 mutations severely affect RPM1‐myc accumulation. A 40 μg aliquot of total protein was loaded and western blots were probed with anti‐myc monoclonal antibody. (B) Top: anti‐myc and co‐IPs demonstrate that HSP90 associates with RPM1‐myc in planta. The relative amounts of protein from the immune pellet and the total extracts are not equivalent. The pellet is over‐represented by 20‐fold. This experiment is representative of six independent replicates for RPM1‐myc. Bottom: control showing specificity of IP reagents, and that RPM1‐myc is extracted from transgenic lines in appropriate mutant backgrounds under the conditions used for the co‐IP.

Given the genetic interaction between rpm1 and hsp90.2, we asked whether HSP90s could co‐immunoprecipitate (co‐IP) with RPM1‐myc. We probed protein blots of anti‐myc IPs with an antibody raised against cytosolic HSP90 from Pharbitis nil, which should detect all four isoforms of cytosolic HSP90 in Arabidopsis (Krishna et al., 1997). We successfully detected HSP90 in IPs from RPM1‐myc plant extracts, but not in IPs from rpm1 mutant extracts (Figure 6B). The HSP90 antibody failed to detect a clear difference in protein levels between our point mutants or insertion allele and wild‐type (data not shown). We therefore cannot ascertain whether the HSP90 detected in our co‐IPs is, or contains, HSP90.2. We were unable to detect HSP90 in anti‐RIN4 co‐IPs, but could co‐IP RPM1‐myc with anti‐RIN4 from the same extracts (data not shown). These data suggest that RIN4 and HSP90 interactions with RPM1 might be mutually exclusive.

Disease signaling components RAR1 and SGT1 also associate with HSP90 in planta

RAR1 and SGT1 can be co‐immunoprecipitated from plant cell extracts (Azevedo et al., 2002). There are also structural similarities between plant SGT1 and animal proteins required for HSP90 assembly and function (Dubacq et al., 2002; Garcia‐Ranea et al., 2002; see Discussion). These observations prompted us to explore the physical relationships between RAR1, SGT1 and HSP90.

We used antibodies raised against either RAR1 or SGT1 (Azevedo et al., 2002) to IP HSP90 from total plant extracts. We found that anti‐RAR1 is able to co‐IP HSP90 from wild‐type extracts, but not from rar1‐20 extracts (Figure 7A). Arabidopsis contains two SGT1 genes encoding proteins of slightly different mobility (Austin et al., 2002; Tör et al., 2002); both are detected by our antisera under the IP conditions used. We found that anti‐SGT1 antibody is consistently able to co‐IP HSP90 from wild‐type Col‐0 and Ws‐0 extracts (six out of six experiments; Figure 7B). However, we detected no, or only very low amounts, of HSP90 in anti‐SGT1 IPs from sgt1b extracts in multiple experiments (Figure 7B). This suggests that the majority of HSP90 we detected is associated with SGT1b, and that SGT1a may weakly associate with HSP90. We invariably detected HSP90 in IPs from sgt1a extracts (Figure 7B), strengthening this conclusion. Thus, in our experimental conditions, there is a preference for HSP90 to associate with SGT1b compared with SGT1a, consistent with yeast two‐hybrid data (A.Takahashi and K.Shirasu, unpublished).

Figure 7.

HSP90 associates with RAR1 and SGT1. (A) Hsp90 associates with RAR1 in planta (top). The relative amounts of protein from the co‐IP pellet and the total extracts are not equivalent. The pellet is over‐represented by 10‐fold. This experiment is representative of two independent replicates. An extraction control is displayed below the co‐IP. (B) HSP90 displays an apparent preference for SGT1b over SGT1a. (C) HSP90 interacts with RAR1 and SGT1b independently. The N‐terminal portion of RAR1 containing the CHORD I domain is sufficient for interaction with HSP90. The relative amounts of protein from the co‐IP pellet and the total extracts are not equivalent. The pellet is over‐represented by 10‐fold. This experiment is representative of three out of seven independent replicates for sgt1b and four out of four independent replicates for sgt1a. Extraction controls beneath the co‐IPs demonstrate the differential mobilities of SGT1a and SGT1b and the effect of the respective mutations.

Additionally, we detected HSP90 in anti‐SGT1 IPs from rar1 mutant extracts and in anti‐RAR1 co‐IPs from sgt1b mutant extracts (Figure 7C). This suggests that RAR1 and SGT1b can associate independently with HSP90. We corroborated this finding in the converse experiment (IP with anti‐RAR1 and blot with anti‐HSP90) using the null rar1‐20 allele (data not shown). We were, surprisingly, able to detect consistently HSP90 from anti‐RAR1 IPs in rar1‐21 mutant extracts (Figure 7C). The rar1‐21 mutation introduces a stop codon near the end of the CHORD I domain (Tornero et al., 2002b), and consequently lacks the CHORD II domain, known to be the RAR1–SGT1 interaction platform (Azevedo et al., 2002). While we were unable to detect this small RAR1 fragment by direct immunoblot, it appears sufficient to IP HSP90, consistent with two‐hybrid interaction data presented elsewhere (Takahashi et al., 2003).

Artifactual detection of HSP90 in immunoblots is a concern, because it can represent up to 1–2% of total cellular protein (Lai et al., 1984). We therefore used several different antisera to test this, and found that none of them could co‐IP HSP90 (see Materials and methods), thus suggesting strongly that our data represent specific interactions.

Discussion

We provide genetic and biochemical data demonstrating that RPM1 is a cytosolic HSP90 client. We demonstrate that mutations in the Arabidopsis HSP90.2 can specifically modulate RPM1 function. RPM1 accumulation is greatly diminished in specific hsp90.2 missense mutants. We provide pairwise co‐IP data demonstrating interactions between HSP90s and RPM1, RAR1 and SGT1. We show that the HSP90 interaction with RAR1 does not require SGT1, nor does the HSP90 interaction with SGT1 require RAR1. Surprisingly, we did not find an association of HSP90s with RIN4, a protein clearly implicated in RPM1 and RPS2 function (Axtell and Staskawicz, 2003; Mackey et al., 2003), suggesting that RIN4 and HSP90 association with RPM1‐myc might be mutually exclusive.

Our data suggest that RAR1, SGT1 and HSP90 may work together to coordinate RPM1 function. The key questions emerging from our work are: why do specific hsp90.2 missense mutations result in diminution of RPM1 levels, and hence RPM1 function? Is the requirement for HSP90 in RPM1 signaling indicative of a common regulatory mechanism among the NB‐LRR class of R proteins? Do RAR1 and SGT1 act as cofactors for HSP90 in R protein signaling?

Genetic interactions support a quantitative function for HSP90.2 in R‐mediated disease resistance

Our four hsp90.2 missense alleles are recessive and significantly reduce, but do not eliminate, RPM1 function. They do not reproducibly affect any of the seven other R functions tested. The hsp90.2 alleles exhibit non‐allelic non‐complementation with rpm1, suggesting that the two wild‐type proteins act together. The fact that far less steady‐state RPM1‐myc protein accumulates in two hsp90.2 alleles provides a simple, dosage‐based mechanistic explanation for this genetic observation.

Our results suggest that specific HSP90.2 missense mutations alter RPM1 function. The four hsp90.2 alleles reported here compare with 95 rpm1 alleles also identified in this screen (Tornero et al., 2002a). RPM1 is 926 amino acids long, HSP90.2 is 699; thus target size cannot explain this mutant ratio. Furthermore, the insertional hsp90.2‐5 allele is viable, apparently null, and exhibited full RPM1 function. We assume that one of the other three cytosolic HSP90s can compensate for the full loss of HSP90.2, as observed in other systems (Borkovich et al., 1989). Thus, our hsp90.2 alleles also prohibit functional compensation. Significantly, we did not recover alleles in any of the other HSP90 genes as loss of RPM1 function mutants. Thus, our hsp90.2 mutants are rare, and suggest a preferential utilization of HSP90.2 in RPM1 accumulation and, hence, in RPM1 function.

Importantly, a mutation exactly orthologous to hsp90‐2.3 (D80N) has been studied in yeast HSP90 (D79N). This mutant yeast protein homodimerizes properly, and a wild‐type–mutant mixed dimer exhibits wild‐type levels of ATP hydrolysis (Richter et al., 2001). However, mutant dimers are unable to bind or hydrolyze ATP (Obermann et al., 1998; Panaretou et al., 1998), and in competition assays are unable to interfere with ATP hydrolysis, even in 8‐fold excess (Richter et al., 2001). Interestingly, a transcriptionally inducible yeast hsp90 (D79N) also prevented phenotypic compensation by a wild‐type, constitutively expressed isoform (Panaretou et al., 1998). This may explain how we identified our mutants in plants that normally contain four copies of cytosolic HSP90, as well as suggesting that HSP90 isoforms might have non‐overlapping functions in Arabidopsis.

Our other two mutations, hsp90.2‐1 (G95E) and hsp90.2‐2 (S100F), have not been identified in screens in other systems. Both are solvent exposed and adjacent to ATP‐interacting residues. It is possible that these mutations also interfere with ATP binding. Altered ATP binding/hydrolysis can also directly affect client protein binding and release by HSP90, as well as the interaction of HSP90 with its co‐chaperones, thus de‐stabilizing the complex (Obermann et al., 1998; Panaretou et al., 1998).

Our evidence of both synergism (in pathogen growth assays) and epistasis (in an HR assay) between hsp90.2 and ndr1 suggests that HSP90 and NDR1 act together in the RPM1‐dependent disease resistance response. A requirement for HSP90.2 early in RPM1 signaling is consistent with our proposal that it is part of, or required for the assembly of, a poised RPM1 receptor complex. These results suggest an interesting parallel with animal innate immunity. NDR1 is a putative GPI‐anchored protein. In animal systems, HSP90 co‐localizes with the GPI‐anchored protein CD14, and both act in the TLR4‐dependent innate immune response to lipopolysaccharide (Triantafilou et al., 2002). Also, HSP90 inhibitors prevent the innate immune response activation by bacterial DNA (Zhu and Pisetsky, 2001). We propose that NDR1 and HSP90 function together, perhaps transiently, during RPM1 signaling.

Could RAR1 and SGT1 function as HSP90 cofactors in disease resistance pathways?

Several observations indicate that RAR1 and HSP90 function together in RPM1‐dependent HR. We demonstrated that RAR1 and SGT1 can associate with HSP90 in vivo. We noted a severe attenuation of RPM1‐dependent HR in hsp90.2 mutants and we found decreased RPM1‐myc stability in hsp90.2 mutants, two phenotypes observed in rar1 plants (Tornero et al., 2002b). The epistasis of hsp90.2 over ndr1 with respect to HR was also observed with rar1;ndr1 double mutants assayed for RPM1 function, and for the synergistic interaction of rar1 and ndr1 in RPP7‐mediated resistance (Tornero et al., 2002b). These data are, in sum, consistent with RAR1 and HSP90 acting together in both RPM1‐dependent HR and, possibly more broadly, in NB‐LRR function. Our data are further consistent with recent findings demonstrating that RAR1 and NDR1 contribute quantitatively to the function of various NB‐LRR R proteins (Muskett et al., 2002; Tornero et al., 2002b).

Because an sgt1a;sgt1b double mutant is lethal in Arabidopsis (A.Takahashi and K.Shirasu in preparation), it is impossible to determine if SGT1 isoforms have overlapping function in RPM1 signaling. Recent gene silencing experiments corroborate a role for HSP90 in the function of several R genes that also require SGT1 and RAR1 for their function (Lu et al., 2003). Structural modeling also supports the contention that RAR1 and SGT1 might act as cofactors of HSP90 (Shirasu and Schulze‐Lefert, 2003). SGT1 and RAR1 homologs in animals have predicted structural homology to the HSP90 partner protein, p23 (Dubacq et al., 2002; Garcia‐Ranea et al., 2002; see Supplementary figure 3). Additionally, the TPR domain of SGT1 shares structural homology with other HSP90 partner proteins including HOP/STI1 (Garcia‐Ranea et al., 2002).

We propose that RAR1 and HSP90 normally not only act to maintain RPM1 in a signal competent conformation, but also stabilize RPM1 against degradation. This is reminiscent of the assembly of activation‐competent steroid receptors with an HSP90 isoform homodimer and various co‐chaperones (Pratt and Toft, 2003). HSP90 binding to the steroid receptor is not sufficient to render the receptor competent; cofactor binding and continual ATP turnover are required to maintain the steroid‐binding cleft in a receptive conformation (Pratt and Toft, 2003). Further conformational change accompanies ligand binding.

Inhibition of ATP binding and/or turnover in our hsp90.2 mutants should result in a locked HSP90 conformation, bound to RPM1 but unable to hold it appropriately, thus leading to RPM1 disappearance. This would mimic the effect on client proteins observed after treatment with the ATP‐binding inhibitor geldanamycin in other systems. RPM1 instability is consistent with results showing that HSP90 can rapidly shut off transcriptional responses by binding transcription factors and causing their degradation (Freeman and Yamamoto, 2002). Steroid receptor levels, like RPM1 (Boyes et al., 1998), drop after signaling (Lange et al., 2000; Wallace and Cidlowski, 2001).

Evidence for a second HSP90 function in protein stability is also emerging. First, human SKP2, a member of the SCF complex, is able to co‐immunoprecipitate HSP90b in mouse NIH‐3T3 cells (Lyapina et al., 1998). Degradation of HSP90 client proteins, triggered by either geldanamycin treatment or overexpression of the E3 ligase CHIP, can be inhibited by the addition of proteasome inhibitors (Whitesell and Cook, 1996; Segnitz and Gehring, 1997; Connell et al., 2001). However, full steroid binding is not recovered in these experiments, also suggesting two functions for HSP90. One, reversible by lactacystin, is required for degradation, and another, not completely reversed by lactacystin, is required to mold a steroid‐binding complex. Initiation of RPM1 function leads to RPM1 degradation (Boyes et al., 1998), perhaps involving SGT1 that is first recruited to an HSP90–RPM1 complex and then guides RPM1 to the proteasome. RAR1 may normally block this conversion, perhaps in conjunction with (or antagonistically to) SGT1. This notion is consistent with our findings that RAR and SGT1 do not require each other to associate with HSP90.

A model where HSP90, in association with RAR1 and SGT1, controls levels of properly poised R protein complexes is consistent with the two functions of HSP90 discussed above—conformational molding and trafficking to the proteasome. Because of both R protein sequence polymorphism and the possibility that R proteins might associate with additional cellular proteins, one could expect differential functional requirements for maintenance of this poised complex. For example, we also could co‐IP HSP90 with anti‐hemagglutinin (HA) monoclonal antibody detecting an RPS2‐HA fusion (data not shown), suggesting that at least one other NB‐LRR protein can interact with a cytosolic HSP90. However, we observed no change in RPS2 function in our mutants. This model is consistent with the fact that RPM1‐mediated HR is very fast compared with others, including RPS2, and that RPM1 is degraded following triggering, unlike RPS2 (Axtell and Staskawicz, 2003). A requirement for a finely tuned conformational poise in NB‐LRR R protein function was suggested recently using split Rx molecules (Moffett et al., 2002).

We thus favor an overall model whereby different NB‐LRR R proteins, perhaps in association with the cellular proteins they guard, are kept in active sentinel mode by varying degrees of dynamic re‐shaping and maintenance of appropriate steady‐state levels driven by HSP90, RAR1 and SGT1. This model can encompass a continuous quantitative function for HSP90 in both the assembly of conformationally charged R protein complexes and the regulation of signal flux through those complexes.

Materials and methods

Plant lines

Transgenic Arabidopsis ecotype Columbia (Col‐0) (line a11) and rpm1‐1 (line a11r) containing estradiol‐inducible avrRpm1 have been described in Tornero et al. (2002a). Mutant lines used (all in Col‐0 unless noted) were ndr1‐1 (Century et al., 1997), rar1‐20; a null allele, originally pbs2 (Warren et al., 1999), rar1‐21 (Tornero et al., 2002b), rps2‐101C (Mindrinos et al., 1994), rps5‐2 (Warren et al., 1998), ecotype RLD (Hinsch and Staskawicz, 1996) as an rps4 mutant control, sgt1a (T‐DNA insertion in Ws‐0 ecotype; A.Takahashi and K.Shirasu, in preparation), and sgt1b (edm1‐1; Tör et al., 2002). We constructed double mutants of hsp90.2 and rpm1‐1 by identifying F2 individuals susceptible to Pst DC3000(avrRpm1) which were molecularly heterozygous for rpm1‐1, yet did not give rise to resistant offspring in the next generation. The F3s from such a line were then selected for a homozygous rpm1‐1 mutant. These lines were confirmed after identification of the hsp90.2 mutations using PCR‐based markers. A similar procedure was used for creation of hsp90.2 and ndr1 double mutants. A homozygous insertion in the SALK tDNA insertion line 058553 was identified by molecular analysis of a segregating pool. The insertion site was confirmed by sequencing of T‐DNA‐specific product. Primer sequences for selection of mutations are available on request.

Bacterial strains, inoculation and growth quantification

Pst DC3000 derivatives containing pVSP61 (empty vector), avrRpm1, avrB, avrRpt2, avrPphB or avrRps4 were maintained as described (Ritter and Dangl, 1996). Plant inoculations and counting of the bacteria were performed as described (Tornero and Dangl, 2001). Where indicated, high concentrations of bacteria (OD600 = 0.075, 3.75 × 107 c.f.u./ml) were infiltrated into the bottom part of the leaf with a blunt syringe to test for the induction of HR.

Estradiol induction

Two‐week‐old plants grown under short day (8 h) conditions were sprayed with 0.02% silwet L‐77 (CKWitco Corporation) and 10 μM β‐estradiol (Sigma E 8875) in distilled water from a 10 mM β‐estradiol stock dissolved in 100% ethanol (Tornero et al., 2002a).

Mapping and tests for disease symptoms

Rough mapping was performed by crossing hsp90.2 mutants and Landsberg erecta (La‐er). F2s were tested for lra2‐like disease symptoms. One‐ to three‐week‐old F2 plants were sprayed with a 10 mM MgCl2 suspension containing PstDC3000(avrRpm1) at a concentration of OD600 = 0.1 (5 × 107 c.f.u./ml) with 0.02% silwet L‐77, covered with a clear lid for 4 h, and assessed for chlorosis and other symptoms of P.syringae infection 4–6 days later under short day conditions. χ2 analyses showed that all mutations were recessive: lra2‐1: 480 wt, 612 mutant χ2 = 3.843; P = 0.05; lra2‐2: 217 wt, 89 mutant χ2 = 2.723, P = 0.099; lra2‐3: 457 wt, 155 mutant χ2 = 0.035, P = 0.85. Susceptible F2 individuals were allowed to self and were confirmed in the F3 generation. DNA (Ausubel et al., 1987) from 41 of these individuals was used in PCR amplification of known PCR‐based molecular markers (http://www.arabidopsis.org) to obtain approximate mapping positions. This interval was refined using molecular markers we developed (available upon request). We used DNA from 939 susceptible F2 individuals to define a 52 kb interval on P1 clone MDA7. The ∼52 kb LRA2 interval lies between a T/A polymorphism at position 551 (Jander et al., 2002) and the published CAPS marker MDA7 (http://www.arabidopsis.org) at position 52 632 (C/A) relative to the published sequence for P1 clone MDA7 (Kaneko et al., 1998). Independent mapping of the lra2‐2 (195 susceptible individuals) and lra2‐3 (312 susceptible individuals) alleles showed similar linkage. All mutations where confirmed by sequencing of both DNA strands.

Alignments and threading analysis

Protein alignments of HSP90 N‐termini were made using Align X [a component of Vector NTI Suite 7.1; Informax, Inc. (Frederick, MD)]. This program uses CLUSTALW to make alignments. Parameters used were: gap opening penalty = 10, gap extension penalty = 0.05, gap separation penalty range = 8, percentage identity for alignment delay = 40, and hydrophobic residue gap = GPSNQEKR.

The sequences of AtSGT1a and AtSGT1b were submitted to the threading Meta server [META] (http://bioinfo.pl/meta/) to identify structural templates for homology modeling. The meta server accessed the following fold recognition servers and reported the consensus: bioinbgu [BIOINBGU], 3D‐PSSM [3D‐PSSM], GenTHREADER [GENTHREADER], FUGUE [FUGUE] and Sam‐T99 [SAMT99]. The meta server identified the crystal structure of the TPR1 domain of Hop (PDB ID 1elw) as a structural template for residues 1–115 of AtSGT1a, and the crystal structure of the human co‐chaperone P23 (PDB ID 1ejf) as a structural template for residues 156–237 of AtSGT1a. Models of the P23 and TPR domains were built using the Modeler module of the InsightII molecular modeling system from Accelrys Inc. (http://www.accelrys.com). Figures were created with SPOCK.

Protein blots and co‐immunoprecipitations

For detection of RPM1‐myc in hsp90.2‐2 and hsp90.2‐3, we introgressed these mutants into plants expressing RPM1‐myc from the native RPM1 promoter (Boyes et al., 1998) as described in Tornero et al. (2002b). Total protein was extracted in 50 mM Tris–HCl pH 8.0, 1% SDS, 1 mM EDTA, 1 mM 2‐mercaptoethanol and 1× plant protease inhibitor cocktail (Sigma). For immunodetection, 40 μg protein samples were electrophoresed on 8% SDS–polyacrylamide gels. Western blots were performed using standard methods and detected with ECL+ (Amersham).

For co‐immunoprecipitations, tissue was first ground in liquid nitrogen with a mortar and pestle. This material was then homogenized by alternate rounds of Polytron (Kinematica) and glass douncer (Kontes Glass Company) in 2 ml of sterile buffer 20 mM Tris–HCl pH 8.0, 0.33 M sucrose,10 mM EDTA, 5 mM dithiothreitol (DTT) and 1× plant protease inhibitor cocktail (Sigma) per 1 g of tissue. Debris was removed by centrifugation at 5000 g for 40 min at 4°C. A 1.5 ml aliquot of this supernatant was first pre‐cleared by adding 50 μl of protein G–agarose (Boehringer Mannheim) and incubated at 4°C for 60 min on an orbital shaker. The cleared supernatant was then removed and combined with one of the following: 5 μl of the anti‐RAR1 (Muskett et al., 2002) antibody, 5 μl of anti‐SGT1 (Azevedo et al., 2002) antibody, 30 μl of a re‐suspended anti‐HA Affinity Matrix (3F10, Roche) or 30 μl of a re‐suspended anti‐c‐Myc agarose (9E10, Santa Cruz Biotechnology). This was followed by incubation at 4°C for 2 h. A 50 μl aliquot of protein G–agarose was then added to the reactions containing the anti‐RAR1 and anti‐SGT1 antibodies. All reactions were then rolled at 4°C overnight. Beads were pelleted at 1000 g for 5 min. This was followed by four washes in 1.5 ml of 50 mM HEPES pH 7.4, 100 mM NaCl, and 10 mM EDTA pH 8.0. Bound proteins were eluted with 50 μl of sample buffer and run on an 8% polyacrylamide gel, and probed with a polyclonal antibody raised against the C‐terminal portion of P.nil HSP90 (Krishna et al., 1997).

HSP90 was not detected in control immunoprecipitations with four different antibodies, three of which do immunoprecipitate HSP90 via RPM1‐myc, RAR1 or SGT1b. Thus, HSP90 is not non‐specifically sticking to protein A/G–agarose beads or other matrix reagents. Furthermore, we used antibodies to actin (soluble), ascorbate peroxidase (soluble), BiP (soluble and endoplasmic reticulum), toposiomerase II (nuclear), RD28 (intergral plasma membrane) (Daniels et al., 1994) and Tip (tonoplast intrinsic protein). None of these co‐immunoprecipitated HSP90. The absence of HSP90 in these IPs (data not shown) argues for the specificity of our co‐IP data.

Supplementary data

Supplementary data are available at The EMBO Journal Online.

Supplementary Information

Supplementary data [emboj7595457-sup-0001.pdf]

Acknowledgements

We thank members of the Dangl lab, especially Drs Ben Holt and Gopal Subramaniam, for technical advice, helpful suggestions and critical reading of the manuscript, and Dr Brenda Temple of the UNC‐CH Center for Structural Biology for assistance with threading analyses. This work was funded by the NSF‐2010 Arabidopsis Project grant IBN‐0114795 to J.L.D. P.K. was supported by a research grant from the Natural Sciences and Engineering Research Council of Canada. A.T. and K.S. were supported by BBSRC and Gatsby Charitable Organization.

References