Gene Expression Signatures from Three Genetically Separable Resistance Gene Signaling Pathways for Downy Mildew Resistance^,

Abstract

Resistance gene-dependent disease resistance to pathogenic microorganisms is mediated by genetically separable regulatory pathways. Using the GeneChip Arabidopsis genome array, we compared the expression profiles of approximately 8,000 Arabidopsis genes following activation of three RPP genes directed against the pathogenic oomycete Peronospora parasitica. Judicious choice of P. parasitica isolates and loss of resistance plant mutants allowed us to compare the responses controlled by three genetically distinct resistance gene-mediated signaling pathways. We found that all three pathways can converge, leading to up-regulation of common sets of target genes. At least two temporal patterns of gene activation are triggered by two of the pathways examined. Many genes defined by their early and transient increases in expression encode proteins that execute defense biochemistry, while genes exhibiting a sustained or delayed expression increase predominantly encode putative signaling proteins. Previously defined and novel sequence motifs were found to be enriched in the promoters of genes coregulated by the local defense-signaling network. These putative promoter elements may operate downstream from signal convergence points.

Genetic screens in Arabidopsis have defined a complex network of pathways controlling local immune responses. These appear to be broadly conserved across all plants analyzed to date. Proteins encoded by disease resistance (R) genes mediate specific molecular recognition of pathogenic microorganisms and trigger signaling cascades that activate defense reactions (Dangl and Jones, 2001; Hammond-Kosack and Parker, 2003). Members of the largest R protein class feature nucleotide binding sites and Leu-rich repeats (NB-LRR). In dicots, NB-LRR proteins can be subdivided into those expressing either putative coiled-coil (CC) domains or a domain with homology to the cytoplasmic tail of animal signaling proteins called TIR at the N terminus.

R-mediated pathogen recognition is often associated with a localized hypersensitive cell death response (HR) of cells directly in contact with, or very near to, the invading pathogen. In Arabidopsis, mutation analysis has defined several distinct defense signaling pathways (Aarts et al., 1998; McDowell et al., 2000). Some R functions require accumulation of salicylic acid (SA; Klessig et al., 2000). Genetic evidence suggests that there are at least two separable R-dependent signaling branches. One requires EDS1 and PAD4 (Falk et al., 1999; Feys et al., 2001), proteins with limited homology to lipases, and the other requires NDR1, a protein of unknown biochemical function (Century et al., 1997). The EDS1/PAD4 signaling pathway is typically associated with TIR-NB-LRR proteins, while the NDR1 pathway is typically associated with CC-NB-LRR proteins. There are, however, exceptions to this generality (McDowell et al., 2000; Bittner-Eddy and Beynon, 2001).

R functions can differ in their requirement for the genetically defined defense regulators mentioned above. For example, RPP7 encodes a CC-NB-LRR protein that recognizes the Hiks1 isolate of the oomycete pathogen Peronospora parasitica (McDowell et al., 2000). RPP7 function is SA and PAD4 independent (McDowell et al., 2000). RPP7 is also unaffected in transgenic plants expressing NahG, a bacterial gene encoding an enzyme that degrades SA (Gaffney et al., 1993; McDowell et al., 2000), eds16/sid2 (T. Eulgem and J.L. Dangl, unpublished data), a mutant defective in the isochorismate synthase enzyme required for SA biosynthesis (Wildermuth et al., 2001), or eds5 (J.M. McDowell, unpublished data). RPP7 resistance does require SGT1b, a putative regulator of proteasome-dependent protein degradation (Austin et al., 2002; Azevedo et al., 2002; Tör et al., 2002). RPP8 encodes a CC-NB-LRR protein that recognizes the P. parasitica isolate Emco5. RPP8 is also SA independent by the criteria defined above (McDowell et al., 1998, 2000) but differs from RPP7 because it is SGT1b independent. In fact, RPP8 function is unaltered by ndr1, rar1, sgt1b, eds1, pad4, sid2, NahG, npr1, ein2, or coi1 mutations (McDowell et al., 1998, 2000). RPP4 encodes a TIR-NB-LRR protein that requires SA accumulation, PAD4, and SGT1b function. RPP4 was reported to be weakly compromised in cotyledons of ndr1 and npr1 mutants (van der Biezen et al., 2002). Yet, under our experimental conditions (see “Materials and Methods”), we did not observe any reduction of RPP4 function in these plants (Table I; data not shown).

Table I.

Peronospora interactions examined by expression profiling

Pathway	Plant Line	P. parasitica Isolate	Sporangiophores/Cotyledon	Interaction
RPP4	Col-0 (RPP4)	Emoy2	0.2 ± 0.1	I
RPP4	ndr1-1	Emoy2	0.5 ± 0.3	I
RPP4	npr1-1	Emoy2	0.1 ± 0.1	I
RPP4	pad4-1	Emoy2	14.3 ± 2.2	C
RPP4	NahG	Emoy2	7.1 ± 1.4	C
RPP7	Col-0 (RPP7)	Hiks1	0.0 ± 0.0	I
RPP7	rpp7-3	Hiks1	16.0 ± 1.3	C
RPP7	sgt1b	Hiks1	13.2 ± 1.3	C
RPP8	Col-0::RPP8	Emco5	0.06 ± 0.04	I
RPP8	Col-0 (rpp8)	Emco5	15.9 ± 1.0	C

Two-week-old seedlings were sprayed with 10⁵ spores of the respective P. parasitica isolate/mL. The number of sporangiophores/cotyledon (n ≥ 20) was determined 7 dpi. I, incompatible, C, compatible.

In addition to R-dependent signaling pathways that mediate rapid and strong resistance responses, plants express a basal defense that is pathogen nonspecific (Glazebrook et al., 1996, 1997a, 1997b). Basal defense limits the growth of virulent pathogens. There is genetic overlap between loci required for R-mediated and basal defenses, suggesting that they may share components (for review, see Glazebrook et al., 1997a). Disease resistance mediated by SA signaling can additionally require NPR1, a nuclear transported protein required for a significant portion of the overall defense gene activation during systemic acquired resistance (Kinkema et al., 2000).

Differences in global gene expression patterns between incompatible (plant resistant) and compatible (plant susceptible) interactions are quantitative and temporal rather than qualitative (Maleck et al., 2000; Tao et al., 2003). For example, approximately 30 genes were found to be induced to higher expression levels during incompatible than during compatible interactions of Arabidopsis with P. parasitica (Maleck et al., 2000). Thus, R-dependent signaling accelerates and amplifies the regulation of a large suite of defense genes that largely overlap those induced by the basal defense system. Interruption of R signaling by mutation should alter the expression amplitude and/or timing of these genes. Here, we characterize the transcriptional response following stimulation of three genetically separable R-gene pathways, comparing resistant Arabidopsis lines to isogenic lines defective in the respective pathways.

We present a comparative analysis of global gene expression patterns triggered by three different R-dependent defense pathways: (1) the canonical RPP4 that is dependent on PAD4, SA accumulation, and SGT1b; (2) the RPP7 pathway that is dependent on SGT1b but independent of PAD4 or SA accumulation; and (3) the unique RPP8 pathway that is independent of PAD4, SA accumulation, or SGT1b. Despite the genetic disparity for signaling downstream from these recognition events, we found that all three pathways trigger up-regulation of common sets of target genes, indicating signal convergence upstream of these target genes. RPP4 and RPP7 trigger at least two distinct temporal patterns of gene activation, each targeting genes enriched for defined functional categories. Potential binding sites of at least three different types of transcription factors were found to be conserved in promoters of genes coregulated by the local defense-signaling network.

RESULTS

Definition of Gene Sets Controlled by RPP4, RPP7, or RPP8

We sought to define sets of genes controlled by three genetically separable defense signaling pathways. We infected wild-type and mutant plants disrupted in RPP4-, RPP7-, or RPP8-mediated resistance with the P. parasitica isolates that trigger each of the respective R genes (Table I). RPP8 was originally cloned from accession Landsberg erecta. The rpp8 allele in Columbia (Col-0) does not recognize any known pathogen (McDowell et al., 1998). Because the majority of relevant signaling mutants were derived in Col-0 and because there is a weak second resistance locus in Landsberg erecta against P. parasitica Emco5, we used an RPP8 transgene under the control of its own promoter in Col-0 for these experiments.

Because the P. parasitica infection process is asynchronous and because the timing to HR and cessation of pathogen growth is slightly different for each resistance response we assayed, we chose time points based on microscopic observations for each RPP gene. For example, infection with P. parasitica Emoy2 and P. parasitica Hiks1 results in hyphal growth by 48 h postinfection (hpi) during compatible interactions, while RPP4- and RPP7-dependent HR are clearly visible at this time point during incompatible interactions (Fig. 1). The timing of P. parasitica Emco5 infections differs substantially from that of P. parasitica Hiks1 and P. parasitica Emoy2 infections. P. parasitica Emco5 spore germination and hyphal growth are already visible during compatible interactions, at 12 hpi following P. parasitica Emco5 infection (Fig. 1), and RPP8-mediated HR is clearly detectable at this time point during incompatible interactions (Fig. 1). Thus, regulatory events and physiological responses responsible for the differences between resistant and susceptible outcomes must occur within the first 48 hpi, but certainly occur on different time scales for each interaction. We therefore determined RNA profiles of all plant lines listed in Table I at 0, 12, or 48 hpi with the respective P. parasitica isolates using Affymetrix Arabidopsis genome arrays representing one-third of the Arabidopsis genome.

Figure 1.

Temporal progression of P. parasitica infection. Trypan blue-stained cotyledons of 2-week-old wild-type, mutant, or transgenic seedlings after infection with P. parasitica isolates Emoy2, Hiks1, or Emco5 at the indicated time points. dpi, days postinfection. Trypan blue stains HR sites (orange arrowheads) and P. parasitica hyphae (red arrows) dark blue.

For each experimental condition, we performed three independent biological repetitions on approximately 50 seedlings per genotype per repetition and pooled equal portions of the corresponding total RNA preparations. Hence, the data from each chip reflect the average of three independent biological experiments covering approximately 150 plants and thousands of interaction sites. Expression data generated for the analysis of each RPP signaling pathway were examined separately.

As inclusion criterion for further analysis, we demanded that a given probe set (oligonucleotide probes representing a defined gene) display at least two independent ≥2.5-fold expression differences within the experiments for each RPP signaling pathway. These conservative inclusion criteria will underestimate the total number of genes responding to each P. parasitica infection and will miss expression changes unique to one treatment at one time point. But this treatment will ensure that the included genes are likely to be true positives (Maleck et al., 2000). We selected 88 genes that meet these criteria in comparisons of Col-0 wild-type plants with isogenic ndr1.1, npr1.1, pad4.1, or NahG at 0, 12, or 48 hpi with P. parasitica Emoy2 (the RPP4 set; Supplemental Table Ia). Similarly, 72 genes were selected that exhibit at least two ≥2.5-fold expression differences between Col-0 wild-type plants and rpp7-3 or sgt1b at 0, 12, or 48 hpi with P. parasitica Hiks1 (the RPP7 set; Supplemental Table Ib). Finally, 182 genes were selected displaying at least one 2.5-fold expression difference between a transgenic Col-0:RPP8 line (McDowell et al., 1998) and Col-0 plants at 0, 12, or 48 hpi with P. parasitica Emco5 (the RPP8 set; Supplemental Table Ic). For this last example, the lack of additional loss of resistance Col-0 mutants altering RPP8 function forced us to adopt this simpler inclusion regimen.

The asynchronous nature of P. parasitica infection probably strengthens the robustness of inclusion for those genes that are in our data set. Pathogen-induced gene expression changes are most dramatic in plant tissue directly in and around the infection site (Schmelzer et al., 1989; Eulgem et al., 1999; Kirsch et al., 2001; Rushton et al., 2002). Because we extracted RNA from homogenized aerial seedling tissues, the expression changes detected in our experiments constitute the average of infected and uninfected cells in the aerial tissues at the time point of harvest. As depicted in Figure 1, most plant cells are not in contact with pathogen structures and may not respond. Thus, our observed expression changes must underestimate the expression changes in and near infection sites that one can achieve with uniform stimuli. However, the selection criteria for the inclusion of genes in our data sets were as stringent as (and therefore more conservative than) those typically applied in experiments where more uniform responses are achieved throughout entire plants (e.g. after stimulation with chemicals, abiotic stress, or high titer bacterial infections; Maleck et al., 2000; Kreps et al., 2002; Glazebrook et al., 2003; Tao et al., 2003).

The three datasets defined above (Supplemental Table I) were separately subjected to hierarchical clustering, using average linking (Eisen et al., 1998). The resulting clustergrams defined two categories of genes controlled by each RPP pathway: (1) genes showing elevated expression during all incompatible interactions relative to all compatible interactions at one or more time points (RPP4, RPP7, or RPP8 elevated; Supplemental Table II), and (2) genes showing reduced expression during all incompatible interactions relative to all compatible interactions at one or more time point (RPP4, RPP7, or RPP8 reduced; Supplemental Table II). These genes are controlled by the respective RPP pathways since genetic disruption (or absence) of each pathway alters their expression characteristics.

Importantly, the expression levels for these genes are typically not altered in the mutants that affect another of the three RPP pathways assayed. For example, RPP4 elevated genes are consistently more strongly expressed following infection of P. parasitica Emoy2 resistant Col-0 plants than in P. parasitica Emoy2 susceptible pad4 mutants or NahG plants, but their expression is not altered following infection of mutants like ndr1 or npr1 that are irrelevant for RPP4 function in our tissues. This strict correlation strengthens their definition as RPP4 controlled genes.

The RPP response clustergrams in Figure 2 reveal several interesting characteristics of RPP4, RPP7, or RPP8 elevated genes. Engagement of any of the three R genes induces elevated expression of target genes at 12 and/or 48 hpi (represented in Fig. 2 by red signal in 12 and 48 hpi columns). The pad4 mutation and NahG transgene have a more pronounced effect on RPP4 elevated genes at 48 hpi, while the rpp7 and sgt1b mutations predominantly affect RPP7 elevated genes by 12 hpi. The absence of RPP8 in Col-0 plants mainly affects RPP8 elevated genes by 12 hpi.

Figure 2.

Profiling of RPP4-, RPP7-, and RPP8-mediated gene expression responses defines sets of R-associated genes. Hierarchical clustergrams with 88 genes (A; represented by 95 probe sets) that show at least two ≥2.5-fold expression differences between Col-0 and ndr1, npr1, pad4, or NahG after inoculation with P. parasitica Emoy2, 72 genes (B; represented by 81 probe sets) that show at least two ≥2.5-fold expression differences between Col-0 and rpp7 or sgt1b mutants after inoculation with P. parasitica Hiks1, or 182 genes (C; represented by 212 probe sets) that show at least one ≥2.5-fold expression difference between Col-0:RPP8 and Col-0 after inoculation with P. parasitica Emco5. Expression ratios are displayed at the indicated time points postinfection; red, positive ratios; green, negative ratios. As illustrated by the color bar in the lower left corner, the brightest color intensity represents a ≥6-fold expression difference. Clusters defining genes more strongly expressed in resistant as compared to susceptible plants (elevated) and genes less strongly expressed in resistant as compared to susceptible plants (reduced) are marked by red and green bars, respectively. Signals suggesting derepression of defense genes in npr1 are encircled. Signal intensities of all genes represented in this figure are listed in Supplemental Table I. In addition, signal intensities of RPP4, RPP7, or RPP8 elevated and reduced genes are listed separately in Supplemental Table II.

This general trend may reflect PAD4 and SA action in RPP4 signaling at a later stage than SGT1b, RPP7, and RPP8 in their respective regulatory cascades. This is consistent with genetic and biochemical analyses of pad4 and sgt1b with respect to other defense signaling events (Rusterucci et al., 2001; Aviv et al., 2002; Tör et al., 2002). Expression of the NahG transgene clearly has a more profound effect on RPP4 elevated genes than the pad4 mutation, as it more efficiently blocks P. parasitica-dependent elevated expression at 12 hpi. Yet by 48 hpi expression profiles of pad4 and NahG plants are almost identical. This may suggest the existence of early signaling events that are independent of PAD4 but affected by NahG.

Interestingly, each of the RPP response pathways appears to influence transcriptional response leading to elevated target gene expression in uninfected plants. A significant number of genes are differentially expressed in defense-compromised backgrounds prior to infection, compared to resistant lines (see 0 hpi columns of Col-0/pad4 and Col-0/NahG [Fig. 2A], Col-0/rpp7 and Col-0/sgt1b [Fig. 2B], and Col-0:RPP8/Col-0 [Fig. 2C]). Comparisons of Col-0 to Col-0:RPP8 are particularly striking in this regard. These results suggest that NB-LRR proteins might have a constitutive regulatory activity. Another, less likely possibility is that the RPP8 transgene insertion causes these transcriptional changes.

NPR1 is a key regulator of systemic disease resistance (Cao et al., 1997; Ryals et al., 1997; Dong et al., 2001; Mou et al., 2003). We previously demonstrated that its importance in systemic acquired resistance is reflected by its regulatory effect on a large number of defense-related genes (Maleck et al., 2000). NPR1 is not necessary for resistance to P. parasitica Emoy2 and RPP4-dependent up-regulation of defense genes (Table I). We nevertheless observed that the majority of genes activated by RPP4 exhibited elevated expression levels in npr1-1 before infection (Fig. 2A). One possible explanation for this may be that NPR1 acts as a negative regulator of some defense genes. This is consistent with recent reports on repression of some defense responses by NPR1 or NPR1-dependent TGA transcription factors (Spoel et al., 2003; Zhang et al., 2003). Derepression of such genes in npr1-1 may partly be responsible for elevated P. parasitica Emoy2 resistance in this mutant as compared to other mutants with defects in SA signaling, such as pad4.

RPP4 and RPP7 Simultaneously Trigger Two Distinct Temporal Patterns of Gene Expression

To examine the kinetic behavior of RPP4, RPP7, and RPP8 elevated genes defined in Figure 2 in detail, we further categorized them using k-means clustering (GeneSpring, Silicon Genetics; see “Materials and Methods”). k-means clustering is a nonhierarchical clustering algorithm that assigns each gene to one of a user-defined number of clusters based on its distance to the centroid of each cluster (Knudsen, 2002). We used this method because it allowed us define common qualitative patterns in gene expression changes over time. Normalized mRNA levels (not expression ratios) detected by probe sets representing 54 RPP4, 50 RPP7, or 135 RPP8 elevated genes were separately clustered into three sets of k-means clusters based on their temporal pattern of expression (Supplemental Fig. 1). We defined a set of genes that display a pronounced mRNA increase between 0 and 12 hpi, after which the expression levels decline or remain constant (early and transient; Table II). By contrast, we found a second set of genes that exhibit a substantial expression increase between 12 and 48 hpi (late or sustained; Table II). In both cases, the timing or amplitude of the expression response is altered in the appropriate susceptible mutant plants (Supplemental Table III). While we appreciate the limited utility of defining temporal patterns of gene regulation based on a two time points, we use these terms to group genes roughly according to apparently shared temporal expression patterns. These groupings proved useful in the subsequent definition of both common putative regulatory elements and common putative cellular functions (see below).

Table II.

Early/transient or late/sustained up-regulation genes controlled by RPP4 or RPP7

AGI No.	Probe Set IDs	Gene Product	GO Annotations	Functional Category	RPP*
Genes Showing Early/Transient Up-Regulation
At1g65970	15116_f_at	Peroxiredoxin TPx2	Antioxidant activity	Defense/metabolism	7, 8
At2g15390	12642_at	Putative xyloglucan fucosyltransferase	Fucosyltransferase activity	Defense/metabolism	4
At2g29460	19640_at	Putative glutathione S-transferase	Glutathione transferase activity	Defense/metabolism	7, 8
At2g43510	19171_at	Putative trypsin inhibitor	Trypsin inhibitor activity	Defense/metabolism	7, 8
At2g43620	18928_at	Putative endochitinase	Chitinase activity	Defense/metabolism	4, 7
At2g45220	20269_at	Putative pectinesterase	Pectinesterase activity	Defense/metabolism	4, 7, 8
At3g26830	14248_at	PAD3, cytochrome P450 emb|CAA50677.1	Indole phytoalexin biosynthesis	Defense/metabolism	4, 7, 8
At3g49120	14638_s_at	Peroxidase	Peroxidase activity	Defense/metabolism	4, 8
At3g54640	17487_s_at, 14672_s_at	Trp synthase α-chain	Trp synthase activity	Defense/metabolism	4, 8
At5g05730	20291_s_at, 12889_s_at	Anthranilate synthase alpha subunit	Anthranilate synthase activity	Defense/metabolism	4
At5g39580	18946_at	Peroxidase ATP24a	Peroxidase activity	Defense/metabolism	4, 8
At5g57550	18968_at, 18969_g_at	Endoxyloglucan transferase	Xyloglucan:xyloglucosyl transferase activity	Defense/metabolism	7, 8
At5g64120	17413_s_at	Peroxidase	Peroxidase activity	Defense/metabolism	4
At3g50770	13217_s_at, 17500_s_at	Calmodulin-like protein	Calcium ion binding	Signaling/transcription	7
At4g17500	12904_s_at, 16063_s_at	EREBP 1	Transcription factor activity	Signaling/transcription	4, 7
At4g33050	19182_at	Protein with calmodulin binding motif	Calmodulin binding	Signaling/transcription	4, 7
At4g36990	16105_s_at	Heat shock transcription factor 4	Transcription factor activity	Signaling/transcription	7, 8
At4g01870	13656_at	Protein of unknown function	Molecular function unknown	Unclear	7
At1g17740	15629_s_at	Phosphoglycerate dehydrogenase	Molecular function unknown	Unclear	4
AT1G27020	18235_at	Unknown	Molecular function unknown	Unclear	4
At2g30140	14614_at	Putative glucosyltransferase	UDP-glycosyltransferase activity	Unclear	4, 8
At2g38860	18255_at, 15866_s_at	Similarity of pfpI-like protein (protease)	Molecular function unknown	Unclear	4
At4g12480	16150_s_at	pEARLI 1	Lipid transport	Unclear	7
At4g15610	17899_at	Unknown	Molecular function unknown	Unclear	7, 8
At4g17470	13949_s_at	Thioesterase like protein	Palmitoyl-hydrolase activity	Unclear	7
At5g13490	15978_at	Adenosine nucleotide translocator	ATP:ADP antiporter activity	Unclear	4
At5g24780	15125_f_at	Vegetative storage protein, VSP1	Acid phosphatase activity	Unclear	7, 8
Genes Showing Late/Sustained Up-Regulation
At1g05300	19718_at	Putative Fe(II) transport protein	Cation transporter activity	Defense/metabolism	4
At1g75040	16153_s_at, 14636_s_at	Thaumatin-like protein	PR5	Defense/metabolism	4
At2g14610	14635_s_at, 17128_s_at	PR-1-like protein	Molecular function unknown	Defense/metabolism	4, 8
At4g08870	17187_at	Arginase	Hydrolase activity	Defense/metabolism	7, 8
At5g42980	13189_s_at	Thioredoxin	Thiol-disulfide exchange intermediate activity	Defense/metabolism	7, 8
At5g64120	17413_s_at	Peroxidase	Response to oxidative stress	Defense/metabolism	7 (4 early up)
At1g21250	15616_s_at	wall-associated kinase 1	Kinase activity	Defense/metabolism	4, 7, 8
At1g28370	14232_at	ERF11	Transcription factor activity	Signaling/transcription	7
At1g33960	12879_s_at, 17544_s_at	AIG1	Nucleotide binding (GO:0000166)	Signaling/transcription	7, 8
At1g68050	14196_at	F-box protein FKF1/ADO3, AtFBX2a	Ubiquitin-protein ligase activity	Signaling/transcription	4
At1g69490	18590_at	NAC domain transcription factor	Transcription factor activity	Signaling/transcription	7
At1g72930	18003_at	Toll/interleukin-1 receptor-like protein	Signal transduction (GO:0007165)	Signaling/transcription	4
At2g40080	18272_at	Similar to RNA polymerase subunit PB2	Positive regulation of circadian rhythm	Signaling/transcription	4
At2g41090	17917_s_at	Calcium binding protein (CaBP-22)	Calcium ion binding	Signaling/transcription	4, 8
At2g46430	17499_s_at	Cyclic nucleotide gated channel, CNGC3	Ion channel activity/calmodulin binding	Signaling/transcription	4, 7, 8
At3g56710	14148_at	Sigma factor A binding protein	Protein binding	Signaling/transcription	4, 7
At4g11280	12891_at, 12892_g_at, 16817_s_at	ACC synthase 6	Ethylene biosynthesis/response to external stimulus	Signaling/transcription	7, 8
At4g21380	16360_at	Receptor-like Ser/Thr protein kinase ARK3	Kinase activity/receptor activity	Signaling/transcription	4
At5g04340	15665_s_at	Putative c2h2 zinc finger transcription factor	Transcription factor activity	Signaling/transcription	7
At5g52310	15611_s_at	Similar to RNA polymerase subunit PB2	Response to abiotic stimulus	Signaling/transcription	4
At1g31580	16439_at	Transmembrane protein with similarity to CD8 C	Response to biotic stimulus	Unclear	4
At1g76960	14096_at	Putative transmembrane protein	Molecular function unknown	Unclear	4, 8
At2g14560	14704_s_at, 15846_at, 15847_g_at	Unknown	Molecular function unknown	Unclear	4, 7, 8
At3g22240	14691_at, 14709_at	Unknown	Molecular function unknown	Unclear	7, 8
At4g14400	20429_s_at	Transmembrane protein with ankyrin repeats	Protein binding	Unclear	4, 7, 8
At4g35480	17047_s_at	RING-H2 finger protein RHA3b	Molecular function unknown	Unclear	4
At5g10760	14145_at	CND41, chloroplast nucleoid DNA binding protein	Proteolysis and peptidolysis	Unclear	4

Genes showing an early/transient or late/sustained up-regulation by RPP4 and/or RPP7 as defined in Supplemental Figure 1. Some of these genes are also controlled by RPP8. Only probe sets that could be clearly assigned to Arabidopsis Genome Initiative (AGI) numbers were included in this table. Gene Ontology (GO) annotations were retrieved from TAIR. We selected for each gene the GO annotation most informative to describe its molecular function and manually assigned it to one of three categories: (1) unclear, (2) defense execution or metabolism, and (3) signaling/transcription. *, RPP genes triggering elevated expression.

RPP4 and RPP7 Early/Transient Genes Predominantly Encode Proteins That Execute Defense Reactions

The majority of both RPP4 and RPP7 early/transient genes encode proteins putatively involved in metabolic processes (approximately 75% of all genes that were assigned to the classes defense/metabolism or signaling/transcription in Table II), many of which are typically associated with defense. Genes with predicted signal transduction or gene regulation functions are much less represented (approximately 20%). Several RPP4 and RPP7 early/transient genes encode cell wall modifying enzymes such as pectin esterase and endoxyloglucan transferase, as well as peroxidases whose enhanced expression may be related to oxidative cross-linking of cell wall components.

Plant defense responses can involve synthesis of phytoalexins, secondary metabolites with potential anti-microbial function. In Arabidopsis, the indole-derivate camalexin that accumulates during pathogen infections can act as a phytoalexin in vitro (Glazebrook and Ausubel, 1994; Rogers et al., 1996; Slusarenko and Schlaich, 2003). Biosynthesis of camalexin requires the cytochrome P450 monooxygenase CYP71B15 encoded by PAD3 (Zhou et al., 1999). Radiolabeled tracer experiments suggested that anthranilate as well as indole are precursors of camalexin (Zook and Hammerschmidt, 1997; Zook et al., 1998), and it was speculated that camalexin synthesis may involve a Trp synthase α-chain as well as the PAD3 product converting shikimate pathway derived indole-3-glycerolphosphate via indole to camalexin (Zook et al., 1998; Zhou et al., 1999). A Trp synthase α-chain gene (TSA1) and PAD3 exhibit pronounced RPP4- and RPP7-mediated early/transient up-regulation. The standard correlation of the TSA1 expression profile to that of PAD3 over all P. parasitica treatments is 0.89. A gene encoding an anthranilate synthase α-subunit that catalyzes a step in the shikimate pathway is also coregulated with TSA1 and PAD3 (standard correlation to PAD3 profile over all P. parasitica treatments = 0.82). PAD3 is required for RPP4 function, while RPP7 function is only modestly reduced in a pad3/pad1 double mutant (Glazebrook et al., 1997). Hence, this early/transient pattern of gene activation initiated by RPP4 and RPP7 consists of genes whose functions are potentially involved in stopping pathogen growth. Some of these genes are also up-regulated by RPP8 (Table II), providing additional support for their potential importance in disease resistance against P. parasitica.

RPP4 and RPP7 Late/Sustained Genes Predominantly Encode Putative Regulatory Proteins

Surprisingly, the majority of both RPP4 and RPP7 late/sustained up-regulated genes (approximately 75% of all genes that were assigned to the classes defense/metabolism or signaling/transcription in Table II) appear to be involved in signaling or gene regulation, whereas genes putatively involved in metabolism are much less represented in this set (Table II). Several members of this category have Ca²⁺-binding motifs and may therefore act downstream from cellular Ca²⁺ fluxes. A large body of evidence points to a role of Ca²⁺ fluxes in defense signaling (Jabs et al., 1997; Zimmermann et al., 1997; Kim et al., 2002), but direct genetic evidence for a contribution of Ca²⁺ fluxes to disease resistance is still lacking. Five of 27 late/sustained genes are also constitutively expressed in the presence of RPP8 (At2g46430, At4g14400, At1g21250, At2g14560, and At2g41090 show at least 2.5 times higher expression levels in Col-0:RPP8 as compared to Col-0). Increased expression of these five genes is strictly associated with disease resistance mediated by three different RPP genes and may therefore control processes executing shared defense functions (see “Discussion”).

The RPP4, RPP7, and RPP8 Pathways Converge in the Up-Regulation of Overlapping Sets of Genes

To uncover commonalities among the responses triggered by the three pathways examined, we performed hierarchical clustering combining all experimental conditions represented by the 21 expression ratios in Figure 2, A to C. Hierarchical clustering was performed in two dimensions (dimensions of genes and experimental conditions) with 419 genes (549 probe sets) that show at least one 2.5-fold expression difference over all 21 comparisons (Fig. 3; Supplemental Table IV). Requiring only one 2.5-fold expression difference for inclusion in this analysis allows the broadest comparisons but at a probable cost to the robustness of any gene expression ratio change that occurs only once in Figure 3. Our goal, however, was to identify common patterns, not to ascribe meaning to expression changes of single genes in single treatments.

Figure 3.

RPP4, RPP7, and RPP8 control common sets of target genes. Hierarchical clustering in gene and treatment dimensions with 549 probe sets representing 419 genes that show at least one 2.5-fold expression difference over all displayed 21 experimental comparisons. Maximal color intensity represents a 3-fold or higher expression difference. Treatments of the RPP4 set (P. parasitica Emoy2 infections) are in blue; treatments of the RPP7 set (P. parasitica Hiks1 infections) are in red; and treatments of the RPP8 set (P. parasitica Emco5 infections) are in green. Two clusters comprising genes that are commonly up-regulated by RPP4, RPP7, and/or RPP8 signaling activities are marked by red bars (clusters I and II). A cluster of genes showing reduced expression in resistant plant lines is marked by a green bar (cluster III). The dendogram above the clustergram represents the relatedness of the overall expression pattern between the different experimental conditions. The shorter the branches that connect two given conditions, the more closely related are the expression profiles associated with them. Experimental conditions that have the strongest impact on RPP4, RPP7, or RPP8 elevated genes (pad4 48 hpi, NahG 12 hpi and 48 hpi, rpp7 12 hpi, sgt1b 12 hpi, and Col-0 [rpp8] 12 hpi) are labeled with asterisks. The dendogram to the left of the clustergram represents the relatedness of expression patterns of individual genes and gene clusters. Branches corresponding to clusters I, II, and III are highlighted in light purple. Some higher order branches were cut off to reduce complexity of the figure. Signal intensities for treatments and probe sets represented in this figure are listed in Supplemental Table IV. In addition, data for all cluster I, II, and III genes are separately listed in Supplemental Table V.

Clustering in the dimension of experimental conditions clearly separated treatments into two sets, A and B (defined by the first node of the dendogram in Fig. 3). Strikingly, the expression profiles associated with conditions that appear to have a strong impact on RPP4-, RPP7-, or RPP8-dependent increases in gene expression occur within set A (pad4 48 hpi and NahG 12 and 48 hpi infected with P. parasitica Emoy2; rpp7 12 hpi and sgt1b 12 hpi with P. parasitica Hiks1 as well as Col-0 12 hpi with P. parasitica Emco5; asterisks in Fig. 3, top). These six key conditions affect two largely overlapping sets of genes, clusters I and II, defined by distinct nodes in the dendogram of genes (Fig. 3, marked by red bars; Supplemental Table V). Many genes within these two clusters are commonly up-regulated by two or all three of the examined pathways, strongly supporting convergence of RPP4-, RPP7-, and RPP8-dependent signaling.

Interestingly, a large number of these genes are also affected simply by the presence of intact RPP4, RPP7, or RPP8 signaling pathways in uninfected tissue, and conditions that define this constitutive activity (all the 0 hpi ratios) also cocluster within set A (Fig. 3). Furthermore, the vast majority of genes controlled by RPP4, RPP7, and RPP8 exhibit elevated expression in npr1-1 plants (green signal in Col-0/npr1 0 hpi column) and may be controlled by NPR1-dependent repression. Hence, P. parasitica-induced elevated expression (at 12 and/or 48 hpi) correlates with derepression in npr1. In support of these findings, 67% of our cluster I and II genes are included in the set defined by Tao et al. (2003) to be differentially expressed following infection with Pseudomonas syringae. These genes typically exhibited elevated expression during incompatible interactions involving avrB/RPM1 and avrRpt2/RPS2 interactions in that dataset (compared to compatible interactions; data not shown).

There are also some displacements between the responses of mutations that affect a particular RPP response and the experimental conditions dendogram in Figure 3. For example, as described above, NahG blocks RPP4-triggered gene expression more efficiently than pad4 at 12 hpi. Perhaps reflecting this, these two conditions are separated by the first node in the dendogram. Alternatively, this separation may reflect the recently described pleiotropy of NahG (Heck et al., 2003; van Wees and Glazebrook, 2003). Expression profiles of rpp7 and sgt1b at 48 hpi are also separated by this node. It is clear that sgt1b affects expression of a larger set of genes than rpp7 (e.g. genes within cluster IIII), consistent with recent reports of a broad role in cellular signaling for Sgt1b (Gray et al., 2003).

Cluster III comprises 79 genes exhibiting reduced expression levels in plants with intact RPP4, RPP7, or RPP8 signaling (Fig. 3; Supplemental Table V). Expression of these genes is elevated in Arabidopsis lines defective in the respective pathways. This effect is most pronounced in NahG and pad4 at 48 hpi following P. parasitica Emoy2 infection, but members of this cluster show the same trend in the other set A conditions. These genes may be commonly down-regulated by all three defense pathways. Alternatively, their expression may be directly or indirectly induced by the growth of P. parasitica in the susceptible plant lines. Several genes involved in photosynthesis and primary metabolism are present in this set (Supplemental Table V).

Figure 4 shows temporal expression profiles of cluster I and II genes from Figure 3. Expression levels (not ratios) normalized to a median of 1 over all tested conditions and time points are presented. Members of cluster I (59 genes, 64 probe sets) predominantly exhibit a pattern of RPP4, RPP7, and RPP8 early/transient up-regulation. Most RPP4 (14 of 16) and RPP7 (9 of 16) early/transient genes as defined in Supplemental Figure 1 are included in cluster I. Genes in cluster II (38 genes, 46 probe sets) predominantly exhibit a pattern of RPP4- and RPP7-triggered late/sustained up-regulation, and most RPP4 (14 of 18) and RPP7 (10 of 14) late/sustained genes as defined in Supplemental Figure 1 are included in cluster II (Supplemental Table II). Although the timing of P. parasitica Emco5 infections is different from that of P. parasitica Emoy2 and P. parasitica Hiks1, some members of cluster II show also a late/sustained pattern after triggering of the RPP8 pathway.

Figure 4.

Temporal expression profiles of cluster I and II genes. Normalized mRNA levels (not ratios) for cluster I and II genes as defined in Figure 3 at 0, 12, or 48 hpi with the respective P. parasitica isolate. Highlighted in red is the weighted average of each gene set (weighted by a control factor for each gene; see “Materials and Methods”).

Our combined analysis of all three RPP signaling pathways identified two gene sets, illustrated as clusters I and II in Figures 3 and 4, that are commonly targeted by the RPP4, RPP7, and RPP8 pathways. Members of each of these two clusters exhibit a defined pattern of temporal expression, early/transient or late/sustained, again indicating that the local defense pathways we examined converge upstream of two distinct temporal patterns of defense-associated gene expression.

Several Genes Tightly Coregulated with PAD4 Encode Signaling Proteins

PAD4 is a regulator of SA biosynthesis (Zhou et al., 1998; Jirage et al., 1999). Thus, one could expect a subset of RPP4-controlled genes that are up-regulated downstream of PAD4 but upstream or independent of SA accumulation. These should be affected in their RPP4-triggered expression in pad4-1 plants but not or less affected in NahG. Interestingly, the PAD4 gene itself shows this type of expression pattern (Fig. 5; normalized mRNA levels; not expression ratios). PAD4 autoregulation has been reported before (Jirage et al., 1999). The pad4-1 mutation is a single nucleotide exchange in the coding region leading to a functionally compromised protein but not a shortened transcript (Jirage et al., 1999). The probe sets representing PAD4 on the chip we used (14249_I_at and 14250_r_at) detect the 3′ end of the PAD4 transcript, which appears not to be altered in pad4-1. PAD4 exhibits a strong and sustained up-regulation in Col-0, ndr1, npr1, and NahG (Fig. 5). In pad4-1 plants, there is an early expression increase, but the sustained expression is blocked. Interestingly, we defined a set of seven PAD4 coregulated genes (correlation coefficient ≥0.85; Fig. 5). EDS1, encoding a defense signaling component that acts in the same pathways as PAD4 (Feys et al., 2001), is one of them. Its sustained up-regulation is blocked in pad4 but not in NahG. Both EDS1 and PAD4 encode putative lipases that were shown to physically interact (Feys et al., 2001). Strict coregulation of a variety of other genes encoding interacting signaling proteins has been shown before (Cooper et al., 2003) and appears to be a common principle in signal transduction processes (Marcotte et al., 1999). Some of the other genes coregulated with PAD4 encode proteins with putative signaling functions (Fig. 5) that we predict will participate in EDS1/PAD4-dependent regulatory processes.

Figure 5.

Genes coregulated with PAD4 are probable functional targets for RPP4-dependent regulation. Expression behavior of PAD4 (represented by the two probe sets in blue) and seven other tightly coregulated genes (correlation coefficient ≥ 0.85). Normalized mRNA levels (not ratios) at 0, 12, or 48 hpi for each indicated plant line are displayed. The weighted average pattern of these genes is shown in red. The individual genes are listed below the graph.

Sets of Coregulated RPP4, RPP7, and RPP8 Target Genes Contain Known and Novel Conserved Promoter Motifs

The high degree of coregulation of early/transient and late/sustained genes suggests common regulatory mechanisms for each of these gene sets. We used the Gibbs Sampling algorithm AlignACE (Hughes et al., 2000) to search for conserved sequence motifs in the promoters of cluster I (early/transient) and cluster II (late/sustained) genes defined in Figure 3 and 4. Functional cis-elements on plant promoters are typically found within the first 1 kb upstream from the translation start site (Rombauts et al., 2003), and we previously used this cutoff to identify cis-elements enriched in pathogen coregulated gene clusters (Maleck et al., 2000). Therefore, we downloaded 1 kb upstream from the inferred translational start site for each gene on this chip (The Arabidopsis Information Resource [TAIR], http://www.arabidopsis.org/tools/bulk/sequences/index.html). Since most known cis-elements consist of 6 to 12 bp, each promoter set was subjected to a series of AlignACE runs varying the parameter width from 12 down to 6. Based on the frequencies of potentially conserved motifs in 1 kb upstream sequences from all Arabidopsis genes as a reference, we calculated P values using the Poisson distribution to describe the likelihood of the observed sequence conservation occurring by chance.

To examine specifically the enrichment of binding sites of transcription factors known to control defense-related genes, only motifs with TGAC, GCC, ACC, and ACGT core sequences were considered. TGAC-containing sequences are known to interact with multiple types of transcription factors, such as members of the WRKY or TGA-bZIP families (Schindler et al., 1992; Eulgem et al., 2000; Jakoby et al., 2002). Binding sites for PBF2 consist of TGAC preceded by four As or Ts (Desveaux et al., 2002, 2004). The triplets GCC or ACC are frequently present in binding sites of ERF (Ethylene Response Factors)-type transcription factors (Rushton and Somssich, 1998; Rushton et al., 2002). ACGT is present in many binding sites of b-ZIP-type transcription factors (Jakoby et al., 2002).

We examined six sets/subsets of genes, the full sets of genes contained in clusters I and II of Figures 3 and 4 and two subsets derived from each of these clusters. The expression profiles of the genes we included in each subset exhibited a standard correlation of 0.90 or 0.95 to the weighted average profile of the respective cluster (see “Materials and Methods”). Thus, they represent genes with the most closely related expression profiles from within each cluster. If conserved promoter motifs are responsible for the coordinated expression of sets of coregulated genes, then their degree of conservation should correlate with the degree of coordinated target gene expression. Consistent with this, we did not find any significant conservation of TGAC-, GCC/ACC-, or ACGT-containing motifs in the full sets of clusters I and II or in genes showing only a correlation of 0.90 to the weighted average profiles of these clusters. However, we found strong conservation of such motifs in the subsets of genes defined by a correlation coefficient of 0.95 or higher to the weighted averages profiles of clusters I and II (Table III). Motif I is conserved in a subset of cluster I genes. It has a strictly conserved TGAC core sequence and, hence, may interact with WRKY or TGA-bZIP-type factors. Motif III is conserved in the respective subset of cluster II genes. This motif has conserved GCC or ACC core sequences, which are typically present in binding sites of transcription factors of the ERF family.

Table III.

Promoter motifs conserved in early/transient or late/sustained genes

Motif	Cluster	Consensus	Observed Frequencya	Expected Frequencyb	P Value	No. of Promotersc
I	I (correlation = 0.95)	TN (G,T) TGACNNG	0.86/1 kb	0.18/1 kb	1.1 E−5	10/14
II	I (correlation = 0.95)	CATGT (C, G) NA	1.21/1 kb	0.25/1 kb	1.9 E−6	8/14
III	II (correlation = 0.95)	(G, A) CCAAAA (G, A)	1.6/1 kb	0.49/1 kb	8.5 E−6	10/13

^aObserved frequency of motifs fitting the consensus sequence per 1 kb of upstream sequence.

^bAverage observed frequency in 1-kb sequence stretches upstream of all Arabidopsis genes.

^cNumber of promoters of respective gene cluster that have at least one copy of a motif fitting the respective consensus sequence.

In addition, we observed strict correlation of a motif containing the invariant palindrome CATG in genes showing a correlation of 0.95 to the average profile of cluster I (Table III, motif II). A search against the plant cis-regulatory elements database PLACE (http://www.dna.affrc.go.jp/htdocs/PLACE/) revealed that numerous plant cis-elements contain CATG motifs. We found in PLACE a series of 21 CATG-containing elements mediating responses to auxin, ethylene, abscisic acid, light or developmental stimuli, bound by VP1, EIN3, bZIP-type, or unknown factors. However, beyond their CATG core, none of these elements has any obvious similarity to the motifs we identified.

We did not find conservation of TGAC-, GCC/ACC-, or ACGT-containing motifs in any subsets derived from cluster III or among the set of genes coregulated with PAD4 (Fig. 5). However, a close inspection of all RPP4 early/transient genes (as defined in Supplemental Fig. 1A) revealed a moderate overrepresentation of one possible PBF2 binding motif (A/TTTTTGAC, P value = 1.5E−3). The PBF2 component AtWHY1 interacts with its binding sites in a transient manner following SA treatment and is required for full defense responses (Desveaux et al., 2004). PBF2 may, therefore, participate in the regulation of the SA-dependent RPP4 response pathway.

DISCUSSION

We profiled gene expression responses triggered by three different disease resistance signaling pathways. RPP4 function is compromised by the pad4-1 mutation as well as the NahG transgene, which do not affect RPP7 or RPP8 function. A Col-0 derived rpp4 mutant has not been described yet. The rpp7-3 and sgt1b mutations fully eliminate RPP7 function (J.M. McDowell, unpublished data; Tör et al., 2002). None of these mutations alters RPP8 function (B. Holt III and J.L. Dangl, unpublished data). By comparing gene expression profiles triggered by each P. parasitica isolate on resistant lines and comparing these to isogenic susceptible lines, we defined gene sets controlled by three genetically separable RPP response pathways (see below). Considerable differences between gene expression profiles from different Arabidopsis ecotypes have been reported recently (Zhu et al., 2001); therefore, we limited our study to Col-0 derived plants.

Technical Validity

We observed very similar expression responses between related conditions (Figs. 2 and 3). Our analyses identified genes whose expression is altered in comparisons of resistant and susceptible lines. We accept that there may be other genes whose transcriptional activity is altered by these infections. We may have missed transcriptional events necessary for resistance mediated by any of the tested resistance specificities. There also may be genes differentially regulated, for example, in P. parasitica Emoy2 infected pad4 that are not altered in P. parasitica Hiks1 infected sgt1b. However, such transcriptional changes, if they occur, are insufficient in sum to result in resistance. This study was designed to focus on sets of genes whose transcriptional change is strictly correlated with RPP function in each case.

We limited our analysis mainly to genes showing elevated expression associated with RPP-dependent disease resistance. Yet, there is a cluster comprising 79 genes (Fig. 3, cluster III) whose expression levels are lower in resistant plants than in susceptible mutants. Higher relative expression levels in susceptible mutants might reflect either more intense P. parasitica-induced gene expression in these lines or elevated basal transcription in the absence of a putative negative regulatory function of the relevant NBS-LRR R protein. In particular, the majority of RPP4 reduced genes (Fig. 2; also included in cluster III of Fig. 3) display a pronounced up-regulation in pad4-1 and NahG following infection (data not shown). Many of these genes encode proteins with putative roles in defense, such as chalcone synthase, thaumatin (PR5), and cell wall modifying enzymes. Strong up-regulation of these genes may be triggered by a PAD4 and SA-independent part of the basal defense system responding to extensive growth of P. parasitica Emoy2.

Shared and RPP-Specific Temporal Patterns of Coregulation

Progression of P. parasitica Emoy2 and P. parasitica Hiks1 infection events occurs with similar speed. Therefore, the timing of gene expression responses mediated by RPP4 and RPP7 can be compared. These pathways converge and trigger two distinct temporal patterns of gene expression. A set of early and transiently up-regulated genes encodes many proteins that might directly fight the invading pathogen. Many of these genes are also controlled by RPP8-dependent signaling. Thus, at least one convergence point of signals derived from all three tested R genes must exist upstream of these early/transient genes. The second pattern of gene activation triggered by RPP4 and RPP7 results in later, or sustained, up-regulation of genes predominantly encoding signaling proteins. Many of these genes are also up-regulated constitutively in the presence of RPP8. Hence, RPP4, RPP7, and RPP8 derived signals must also converge upstream of late/sustained genes.

These convergence points may be mediated by common regulatory molecules such as signaling proteins or small molecule messengers. Reasonable candidates include MAP kinase pathways terminating in the Arabidopsis MPK3 and MPK6 proteins (Asai et al., 2002) and the transcription factors that are presumably the targets of their activities. Tobacco (Nicotiana tabacum) orthologs of these MAP kinases were also shown to be involved in defense signaling cascades (Zhang and Klessig, 2001). Members of the large family of WRKY transcription factors were suggested to operate downstream from MPK3 and MPK6 in Arabidopsis (Asai et al., 2002). Consistent with this, we found a potential binding site for WRKY factors conserved in promoters of genes early/transiently up-regulated by the pathways we examined (Figs. 3 and 4, cluster I). Different signaling routes may activate different sets of transcriptional regulators that then target separate or common promoter elements in the genes of each regulon. Alternatively, convergence of defense signals may occur in parallel at multiple points, each controlling only a subset of defense responses. The sets of genes defined here are commonly targeted by more than one of the examined RPP signaling pathways and will be important tools for the future dissection of the local defense signaling network.

RPP4 and RPP7 late/sustained genes predominantly encode signaling proteins. Their roles in the plant immune system are enigmatic. Sustained up-regulation of these genes may control a long-lasting activation of some local defense reactions. However, their elevated expression at a time point (48 hpi), at which the mRNA levels of many genes executing defense reactions have returned to their ground states, is partially unexpected. Thus, another potential role of these late/sustained signaling genes could be to shut down defense responses and to reset the local defense system. Alternatively, their massive transcript increase at later time points may be a result of a delayed activation in tissue surrounding the infection sites, suggesting that these genes play roles in controlling aspects of disease resistance beyond the infection site and perhaps systemically.

Coregulated Responses Feature Enriched Putative Control Elements

We identified promoter motifs strongly conserved in early/transient or late/sustained up-regulated genes. These will serve as a starting point for the cloning of transcription factors potentially participating in the coordinated regulation of these genes. In particular, the novel CATG-containing motif that is enriched in early/transient genes may allow the identification of important transcription factors of the plant immune system. We anticipate that these hypothetical CATG interacting factors may control physiological responses directly affecting pathogen viability, such as camalexin biosynthesis and cell wall modifications.

This analysis also identified two potential binding sites of transcription factors known to regulate defense-related genes. These motifs contain TGAC or GCC/ACC core motifs that are likely to interact with WRKY or ERF-type transcription factors, respectively. Members of each of these families of plant-specific transcription factors have been demonstrated to regulate defense genes (Rushton et al., 1996; Zhou et al., 1997; Eulgem et al., 1999). In each case positions outside the respective core motifs were found to be highly conserved. WRKY- and ERF-type factors are represented by large families in Arabidopsis (Riechmann et al., 2000). Their members typically share a preference for the core motif common to each family's binding site repertoire. The extended conserved motifs that we identified may constitute specific binding sites of individual members within each family. Permutations fitting the consensus sequences of these two conserved motifs are present in promoters of the majority, but not all members, of the respective gene set. Derivatives of the conserved motifs with slightly altered sequences may be present in the remaining promoters. Further analysis may lead to the definition of cis-elements that are targeted by one or more distinct RPP response pathways. Future experiments will address whether RPP4, RPP7, and RPP8 signaling converges upstream of common cis-elements or if each pathway targets a specific ensemble of regulatory promoter elements. Molecular mechanisms operating at the interface between R gene signaling and defense gene regulation are still largely unknown. Our expression profiling data will facilitate systematic studies to uncover basic principles and details of this important regulatory circuit.

MATERIALS AND METHODS

Arabidopsis Lines and Peronospora parasitica Isolates

All plants used in this work are in the Col-0 genetic background. The Arabidopsis mutant or transgenic lines ndr1-1 (Century et al., 1995), npr1-1 (Cao et al., 1994), pad4-1 (Glazebrook et al., 1997), NahG (Delaney et al., 1994), sgt1b (Tör et al., 2002), and Col-0:RPP8 (McDowell et al., 1998) have been described. In the rpp7-3 allele, codon 796 (encoding Glu) of RPP7 is deleted (X.J. Wang and J.M. McDowell, unpublished data). The Peronospora parasitica isolates Emoy2, Hiks1, and Emco5 were described previously (Holub and Beynon, 1996).

Infection of Arabidopsis Seedlings, Staining of Cotyledon Tissue, RNA Preparation, and GeneChip Data Generation

P. parasitica was grown and propagated as described previously (McDowell et al., 2000). Arabidopsis seedlings were grown on soil for 14 d in a clean growth chamber (10 h day, 14 h night, 21°C; 100 μE m⁻² s⁻¹) and sprayed with 100,000 spores/mL of the respective P. parasitica isolate. Trypan blue staining of infected Arabidopsis cotyledons was performed as described previously (McDowell et al., 2000). Untreated Arabidopsis seedlings or seedlings at 12 and 48 hpi with the respective P. parasitica isolate were shock-frozen in liquid nitrogen. Preparation of total RNA, processing of RNA for GeneChip experiments, hybridization, calculation of signal intensities (average difference values; calculated using Affymetrix Microarray Suite version 4.0; Santa Clara, CA), and overall intensity normalization were performed as described previously (Zhu and Wang, 2000).

GeneChip Data Analysis

Raw data for all chips are deposited at TAIR under the accession number ME00313, according to the MIAME guidelines. Data from each individual chip were normalized against each other by setting their target intensity (average signal intensity) to 100 (Zhu and Wang, 2000). The signals from most of the negative control probe sets were below 25, so we defined signal intensities of 25 (25% of target intensity) as the noise level using our normalization procedure. All signals below 25 were therefore raised to 25 for further data transformation and analysis. The number of false 2-fold expression changes between technical replicates was previously found to be 0.22% using this regime (Zhu and Wang, 2000). This data handling scheme has been used by a variety of authors analyzing diverse biological responses (Harmer et al., 2000; Zhu and Wang, 2000; Zhu et al., 2001; Chen et al., 2002; Kreps et al., 2002; Glazebrook et al., 2003; Laule et al., 2003; Tao et al., 2003). Using Microsoft Excel, average difference values <25 were raised to 25 and ratios of expression levels were calculated. Filtering of expression ratios and hierarchical clustering was performed as described previously (Maleck et al., 2000) using Cluster and Treeview (Eisen et al., 1998). k-means clustering was applied to normalized mRNA levels by GeneSpring 3.5 (Silicon Genetics, Redwood City, CA) using standard correlation as a means to calculate distances between expression profiles. The median of all data points for each chip as well as the median of all data points for each gene were set to 1. Subclusters of clusters I, II, and III were defined using GeneSpring by selecting genes with normalized mRNA profiles showing a standard correlation of ≥0.90 or ≥0.95 to the weighted average profile of the respective cluster. Genes were included in the respective subclusters, if represented by at least one probe set, showing the required correlation to the weighted average. We used the weighted average of each gene set (weighted by a control factor of each gene) for the definition of subclusters. This control factor is a measure of liability and reflects the absolute signal strength. Genes with higher control values are more reliable. As a result, the weighted average gives less weight to potential noise and outliers (http://www.silicongenetics.com/cgi/TNgen.cgi/GeneSpring/GSnotes/Notes/want_average).

Promoter Analysis

For each member of clusters I, II, and III (Figs. 3 and 4) as well as for all Arabidopsis genes as a reference, 1 kb of genomic DNA sequence upstream from the inferred translational start site was downloaded from the TAIR Web site (http://www.arabidopsis.org/tools/bulk/sequences/index.html). To search for conserved motifs within the promoter sequences of each cluster or subcluster, we used AlignACE (Hughes et al., 2000; http://atlas.med.harvard.edu/). For each promoter set, we performed a series of AlignACE runs with the following set of parameters: number of columns to align = 12 to 6; number of sites to expect = number of promoters in respective input set; fractional background GC content = 0.32 (we found the GC content of Arabidopsis promoters to be approximately 32%). The resulting AlignACE outputs were scanned for conserved motifs containing core sequences of known defense-related transcription factor binding sites as well as the tetramer CATG that we found to be highly enriched in cluster I genes. For each conserved motif containing any of these core sequences, we determined the frequency in all Arabidopsis promoters (expressed as average occurrences/1 kb). Using this value, we calculated the expected frequency of each motif in the set of promoters it was originally derived from. P values expressing for each conserved motif the probability for the observed enrichment to occur by chance were calculated by Microsoft Excel using the Poisson distribution with the following set of parameters: number of events = number of occurrences of the respective motif in all promoters of its gene cluster; mean = expected number of occurrences of this motif in this gene cluster based on its average frequency in all Arabidopsis promoters; cumulative = false. Only conserved motifs with P values < 1E−5 were considered further.