Rootstock-regulated gene expression patterns associated with fire blight resistance in apple

Plant Material

Trees for the fire blight tests were purchased from Adams County Nursery Inc. (Aspers, PA) and planted in five replica blocks at The Pennsylvania State University Fruit Research and Extension Center, Biglerville, PA. Trees consisting of 'Gala' scions grafted to a range of rootstocks were planted and conventionally managed with pesticides to control weeds, fungal diseases and insects in preparation for inoculation. For the greenhouse-grown tree first-generation microarrays, fresh bench grafts were grown as described previously [18]. For fire blight susceptibility tests, 'Crimson Gala' (Waliser cultivar) on the same seven rootstocks were planted in 2005 at the Fruit Research and Extension Center in Biglerville, PA. The rootstocks were, from the least to the most vigorous, Malling 27 EMLA (M.27), Budagovsky 9 (B.9), Malling 9 Fleuren 56 (M.9F56), Geneva 30 (G.30), Malling 7 EMLA (M.7), Supporter 4 (S.4), Malling Merton 111 EMLA (MM.111). Throughout the text, plants are described as scion/rootstock combinations. For example, a 'Gala' scion on an M.7 rootstock is designated as 'Gala'/M.7. The trees used for the second-generation microarray experiment were from a segregating population from an 'Ottawa 3' × 'Robusta 5' cross and were grown in an orchard in Geneva, NY [19]. This population had been previously characterized for resistance to fire blight [19].

Fire Blight Susceptibility

For the fire blight tests conducted in Biglerville, PA, actively growing scion shoot tips of three-year old trees were wounded by using scissors to cut across the midribs of the youngest leaves, and a drop of phosphate buffer (10 mM, pH 7) containing 1 × 10⁶ cfu/ml of E. amylovora (strain Ea581a or HKN06P1) was placed on the cut surface. The shoot tip was then covered for 24 hours with a plastic bag containing a wet piece of filter paper to maintain a humid environment and promote infection. Necrotic region measurements were taken over the course of the disease progression. Disease severity was calculated as the length of the blighted section of an inoculated shoot as a percentage of the total shoot length. Four replicate trials were conducted, each replicate consisting of 10 trees of each of the 7 scion/rootstock combinations. At least 5 shoots per tree were inoculated. The fire blight susceptibility of the 'Ottawa 3' × 'Robusta 5' cross progeny used for the second-generation microarray experiment, to two E. amylovora strains (Ea273, Ea2002a), was published previously [19]. The susceptibility data for a third E. amylovora strain (Ea4001a) is unpublished.

RNA isolation and microarray analysis

The methods and results for our first-generation microarray, including RNA isolations from greenhouse-grown trees and microarray analysis, are described elsewhere [18]. The first generation array contained probes designed to detect 55,230 unique transcripts, representing up to 95% coverage of the apple genome. The sequences for all of the contigs used to develop the probes for the arrays can be found at the Gene Expression Omnibus (GEO) dataset website [20]. For the present study, we developed and used a second-generation apple DNA NimbleGen expression microarray that was designed based on our first-generation NimbleGen array [18] and used it to analyze RNA samples isolated from the progeny of the 'Ottawa 3' × 'Robusta 5' cross grown in Geneva, NY. The expression levels for each tree were analyzed on a single array only, with no biological replicates for any individual tree. The second-generation array was a 12-plex array containing 135,000 probes per plex, representing 26,017 transcripts, enabling us to query a relatively large number of samples. The probes for this array represent a subset of those included in the first-generation array. The second-generation array represents the transcripts with the best-performing probe sets from the first-generation array and includes the transcripts that showed differential expression between any two scion/rootstock combinations. Transcripts that had high variability among their probes were left off of the second-generation array. The five best-performing probes of the original six probes per transcript were used in the second-generation array to increase the number of different transcripts that could be queried by the second-generation array.

DNA microarray analysis on our second-generation microarrays was performed by the Penn State Genomics Core Facility at University Park, PA. Briefly, one microgram of total RNA from each sample was amplified using the Ambion (Life Technologies) Amino Allyl MessageAmp II aRNA Amplification Kit (AM1753) following the manufacturer's protocol for one cycle amplification. Fifteen micrograms of aRNA was dye coupled with either Cy3 or Cy5 (GE Health Care #RPN5661), as appropriate. Following quenching and cleanup of dye coupling reactions, 1.5 μg of a Cy3 labeled sample is combined with 1.5 μg of a Cy5 labeled sample and fragmented using RNA Fragmentation Reagents (Ambion AM8740) according the manufacturer's instructions. After fragmentation, samples are dried down completely in a speed-vac and then resuspended in tracking controls and hybridization solution according to the microarray manufacturer's instructions (Roche NimbleGen). Pairs of samples were hybridized overnight at 42°C with active mixing in a MAUI Hybridization System. Following hybridization, microarrays were washed and scanned according to the manufacturer's protocol (Roche NimbleGen). Images were burst, gridded, and pair files generated using NimbleScan software. The gene expression data from the hybridization experiments using the second-generation DNA microarray were normalized using R software and un-adjusted p-values were calculated by regression analysis using R software [21, 22].

Multiple regression analysis

Stepwise multiple regression analysis was used to identify genes whose expression levels were related to fire blight severity. The response variable was mean fire blight severity, calculated as the length of blighted section of an inoculated shoot as a percentage of the entire shoot and averaged across 5 replicates for each of 48 (Geneva trees). The explanatory variables were gene expression levels, expressed as arbitrary units of fluorescence intensity, for a set of 60 candidate genes selected based on potential for involvement in fire blight susceptibility. A full stepwise regression model was implemented so that each of the explanatory variables was evaluated in the regression with significance level for staying in the model set at P = 0.15. For both up-regulated and down-regulated candidate genes, separate stepwise regression analyses were computed for fire blight severity data for each of three E. amylovora strains. Stepwise multiple regression analysis were implemented using the REG procedure of SAS 9.2 (SAS Institute Inc., Cary, NC), and the resulting models were evaluated for goodness of fit based on standard regression analysis procedures [23].

Rootstock-dependent differences in fire blight susceptibility of 'Gala' scions

Significant differences in the relative size of the necrotic regions were observed within 15 days of inoculation of the 'Gala' shoot tips with two different strains of E. amylovora (Figure 1). For both strains, 'Gala'/G.30 and 'Gala'/M.111 were the least susceptible and 'Gala'/B.9 and 'Gala'/M.27 were the most susceptible. Interestingly, strain-dependent differences in fire blight susceptibility were observed for 'Gala'/M.7 and 'Gala'/M.9F56 trees. 'Gala'/M.7 susceptibility to E. amylovora strain Ea581a was similar to that of the most susceptible trees ('Gala'/B.9 and 'Gala'/M.27), while 'Gala'/M.9F56 susceptibility to strain Ea581a was similar to that of the most resistant trees ('Gala'/G.30 and 'Gala'/MM.111). The results were reversed with E. amylovora strain HKN06P1, with 'Gala'/M.7 susceptibility being similar to that of the most resistant trees and 'Gala'/M.9F56 susceptibility being similar to that of the most susceptible trees. E. amylovora strain Ea581a is a moderately virulent isolate, while HKN06P1 is a hypervirulent isolate [24].

Clustering of trees according to phenylpropanoid pathway gene expression

After obtaining fire blight susceptibility ratings for all the 'Gala'/rootstock combinations used in the study, it became possible to ask whether expression levels of genes involved in specific biochemical pathways related to fire blight resistance might contribute to the rootstock-regulated fire blight resistance phenotype. Gene expression levels in scions of all the 'Gala'/rootstock combinations used in this study were previously profiled on a large scale using our first-generation DNA microarray [18], and these microarray data were mined in the present study for genes related to fire blight susceptibility. Because the phenylpropanoid biosynthesis pathway has been implicated in fire blight resistance [11, 14], a search of the apple genome was conducted to find the predicted phenylpropanoid biosynthetic pathway genes. A total of 67 transcripts on our first-generation array were identified as belonging to the phenylpropanoid biosynthetic pathway, among other pathways (Additional File 1, Table S1).

A complete linkage cluster analysis using the expression values for these genes was conducted to evaluate the potential contribution of the expression levels of genes in this pathway to resistance (Figure 2). There were two main clusters of trees based on the expression patterns of putative phenylpropanoid biosynthetic genes. The two least susceptible scion/rootstock combinations ('Gala'/G.30 and 'Gala'/M.111) were paired in one cluster, along with a branch containing one scion/rootstock combination displaying strain-dependent susceptibility ('Gala'/M.7). The second main cluster contained both of the highly susceptible scion/rootstock combinations ('Gala'/M.27 and 'Gala'/B.9) as well as a scion/rootstock combination with strain-dependent susceptibility ('Gala'/M.9F56) and the moderately susceptible scion/rootstock combination ('Gala'/S.4). Thus, the clustering of trees according to phenylpropanoid pathway gene expression closely followed the pattern of susceptibility to the highly virulent E. amylovora strain HKN06P1 (Figure 1).

Clustering of trees according to sugar metabolic pathway gene expression

The AraCyc metabolic pathways tool [25] was used to identify 93 Arabidopsis genes involved in sugar metabolism, and the potential homologs of these Arabidopsis genes in the apple genome were identified. A BLAST search of the set of the Arabidopsis genes involved in sugar metabolism to the apple genome resulted in the identification of 227 unique apple coding sequences. A total of 219 of these transcripts were represented on our first-generation array. A complete linkage cluster analysis using the expression values for the 219 identified sugar metabolism genes did not closely follow the pattern for the levels of fire-blight susceptibility for the various scion/rootstock combinations (compare Figures 1 & 3). Instead, the clustering by sugar metabolism gene expression closely resembled clustering based on data for all the transcripts represented on the first-generation array [18].

Identification and categorization of candidate rootstock-regulated, fire blight susceptibility-associated transcripts

Constitutive gene expression levels in scions of all the 'Gala'/rootstock combinations used in this study were profiled on a large scale previously using our first-generation apple DNA microarray [18]. Using the fire blight susceptibility ratings from the field trials, the microarray data were sorted to identify those transcripts whose expression levels correlated with the differences in fire blight susceptibility among the apple trees being studied. A diagram showing the comparisons used to sort the data is shown in Figure 4. Transcripts of interest were selected based on a fold-expression difference cutoff and a statistical strength cutoff. For a transcript to be selected, every possible pairwise comparison between the two least susceptible ('Gala'/G.30 and 'Gala'/M.111) and the two most susceptible scion/rootstock combinations ('Gala'/B.9 and 'Gala'/M/27) had to have at least a 1.5 fold difference in expression and a q-value of less than 0.05. This selection program resulted in a list of 665 transcripts with higher expression levels in the less susceptible trees (Additional File 1, Table S2), and 25 transcripts with higher expression in the more susceptible trees (Additional File 1, Table S3) for a total of 690 candidate rootstock-regulated, fire blight susceptibility-associated transcripts. The Malus genome BLAST hits and corresponding e-values are included in Additional File 1, Tables S2 and S3.

Genes differentially expressed during fire blight infection are disproportionately represented among the candidate rootstock-regulated, fire blight susceptibility-associated transcripts

Interestingly, of the 54 transcripts identified in common with Norelli et al., over half were down-regulated post infection [11]. However, we found that these same genes had higher expression levels in the less susceptible trees. A few of these were initially up-regulated in the Norelli study at early time points, but were down-regulated at later time points. In Sarowar et al., again over half of the transcripts in common were identified as being down-regulated upon infection; they were expressed at higher levels in the least susceptible trees in the present study. Only 2 of 10 of the genes in common with Baldo et al. were down-regulated upon infection. A KEGG analysis of the genes in Table 3 is included in Table 2. Those pathways that had the highest confidence level for being overrepresented relative to the total transcriptome include protein processing in the endoplasmic reticulum (ko04141) and endocytosis (ko4144).

Analysis of expression patterns of candidate fire blight susceptibility-associated genes in an apple rootstock breeding population

To further analyze which candidate rootstock-regulated, fire blight susceptibility-associated transcripts might prove to be the best indicators of resistance, we examined the expression levels of a subset of the candidate transcripts in a test population of 48 individual, non-grafted apple lines, grown in Geneva, NY, that were offspring from a single 'Ottawa 3' × 'Robusta 5' cross segregating for fire blight resistance. Expression was measured using a second-generation, 135,000 feature microarray (representing ~26,000 transcripts) developed from the original, larger microarray used for rootstock-regulated gene expression profiling [18]. The second-generation microarray was designed before the 690 fire blight-associated candidate transcripts had been identified; of the 690 candidate transcripts, 429 were represented on the second-generation microarray.

For the transcripts identified as having higher levels of expression in less susceptible trees, 3 had an un-adjusted p-value below 0.05 for two strains, and 36 had an un-adjusted p-value below 0.05 for one strain. For those transcripts that had higher expression levels in the more susceptible scion/rootstock combinations, only 3 transcripts had un-adjusted p-values below 0.05. The transcript list in Table 5 includes 3 transcripts previously shown to be differentially expressed upon E. amylovora infection [11] (Table 3). The functional annotation of these genes is included in Table 2. Those pathways, as determined by the KEGG analysis, that had the highest confidence level for being disproportionately represented include: photosynthesis-antenna proteins (ko00196), protein processing in the endoplasmic reticulum (ko04141), and endocytosis (ko4144).

Gene expression patterns correlating with fire blight susceptibility

A total of 13 genes with higher transcript levels in more resistant trees were identified as being significantly related to the level of fire blight caused by at least one of the three E. amylovora strains, with the resulting regression models accounting for between 40 and 60% of the variation in disease severity (Additional File 2, Table S5). Of these, increased transcription of one gene (APPLE0F000020273) was negatively related to levels of fire blight caused by all three strains. Increased levels of transcripts of APPLE0F000027501, APPLE00R00018643 and APPLE0F000019968 were related to lower levels of fire blight caused by at least two of the E. amylovora strains, while increased transcript levels of APPLE0F000019334, APPLE0F000018558, APPLE0F000020583 and APPLE0F000023953 were noted in tissues with higher levels of disease for at least two of the strains (Additional File 2, Table S5).

The models relating the expression of genes with higher levels of expression in more susceptible trees to levels of fire blight were much weaker, only accounting for between 29 and 32% of disease severity (0.031 ≤ P ≤ 0.009; Additional File 2, Table S6). In all, lower expression levels of eight of the genes were related to levels of fire blight for at least one of the E. amylovora strains, with APPLE0FR00081295, APPLE0FR00067567 and APPLE0F000016771 being related with higher levels of fire blight caused by Ea 273 (Additional File 2, Table S6). Only the lower transcript levels of APPLE0F000026657 were related to lower levels of fire blight for all three E. amylovora strains, while lower transcript levels of APPLE0FR00066754 and APPLE0FR00063520 were associated with lower levels of disease caused by Ea273 and Ea4001a, respectively (Additional File 2, Table S6).

Pathways and processes

Three transcripts on the array corresponding to genes in the phenylpropanoid pathway were on our preliminary list of rootstock-regulated candidates, including two for chalcone synthase (APPLE0F000017774, APPLE0F0000178640) and a chalcone isomerase (APPLE0F000056938) (Additional File 1, Table S2). However, these genes do not appear on any of our subsequent lists. Nevertheless, the expression pattern of the genes in the phenylpropanoid pathway as a whole (Figure 2) is consistent with the proposed role of this pathway in the response to E. amylovora infection [14, 15]. This suggests that the expression of the phenylpropanoid pathway as a whole might be a good predictor of fire blight resistance.

Sorbitol dehydrogenase (SDH) (APPLE0F000058028, APPLE0F000007408, APPLE0F000017030), was found to be expressed at higher levels in the trees that were least susceptible to fire blight (Table 3 and Additional File 1, Table S4). Sorbitol is a major form of translocated sugar in apples [29]. SDH converts sorbitol to fructose in sink tissues [30]. For E. amylovora, sorbitol is an important factor in determining host specificity [31]. It may be that higher SDH levels reduce the availability of sorbitol to E. amylovora. However it has also been shown that high sorbitol levels can inhibit the development of disease symptoms [32]. Our analysis of the recently released apple genome suggests that there may be up to 28 genes encoding SDH enzymes, as opposed to a single copy of SDH in Arabidopsis (AT5G51970). This complexity points to the importance of SDH to apple physiology.

Not surprisingly, there was a GO annotation enrichment among the genes having higher expression in the least susceptible trees (Table 1), with genes predicted to be involved in responses to stress and biotic and abiotic stimuli being disproportionately represented relative to the known apple transcriptome as a whole. Upon further analysis of these transcripts (Table 2), more transcripts than expected, relative to the proportion among all of the transcripts represented on the array, were identified in the secretory pathway, including several heat shock proteins (APPLE0F000018523, APPLE0F000020273, APPLE0F000060823), suggesting that protein processing in the endoplasmic reticulum may be more active in the trees least susceptible to fire blight. Heat shock proteins are important for protein processing in the endoplasmic reticulum and have also been shown to play critical roles in signal transduction in defense responses in tobacco [33]. Additionally, higher levels of the transcripts encoding a predicted calnexin (APPLE0F000027501) and a Sec61 homolog (APPLE0F000020583) were associated with reduced fire blight susceptibility by stepwise multiple regression analysis. Both of these proteins have functional annotations indicating involvement in protein processing in the endoplasmic reticulum. Consistent with these findings, it has been shown that the entire secretory pathway is up-regulated during systemic acquired resistance (SAR) [34] as well as during the N-mediated viral defense response in tobacco [35].

Similarly, from the same data set, a greater than expected number of genes involved in both the endocytosis and peroxisomal pathways were identified, suggesting that these processes may be more active in the least susceptible trees. Intriguingly, endocytosis processes are involved in plant recognition of pathogen-associated molecular patterns (PAMPs) [36]. For example, the bacterial flagellin protein FLG22 is recognized in plants by the FLS2 receptor kinase [37], which leads to endocytosis of the FLS2 receptor and its subsequent degradation by the proteosome [38]. Higher levels of a transcript encoding a predicted fatty acyl CoA ligase (APPLE0F000019968) were associated with reduced fire blight susceptibility by the stepwise multiple regression analysis. The protein encoded by this gene has a functional annotation indicating involvement in endocytosis and fatty acid metabolism. We also identified a transcript encoding a putative phosphatidylinositol-4-phosphate 5-kinase (PIP5K) (APPLE0F000017691) that had higher expression levels in the least susceptible trees. The product of PIP5K, phosphoinositol 4,5-bisphosphate (PI(4,5)P2), is a key regulator of clathrin-mediated endocytosis [39].

Additional candidate genes of particular interest

A major objective of this study was to identify candidate genes in apple that are potentially involved in determining fire blight resistance prior to an infection event. Interestingly, quite a few of the candidate genes identified in our study had previous links to disease resistance processes, including differential expression upon E. amylovora infection in apple [11–13]. This includes several heat shock proteins, a leucine-rich repeat transmembrane kinase, and sorbitol dehydrogenase. Strikingly, over half of the transcripts that we identified as being expressed at higher levels in less susceptible trees are down-regulated during E. amylovora infection [11–13]. It is possible that the expression of these genes is down-regulated by the pathogen to promote disease. Additionally, eight of the genes in both Tables 3 have also been shown to be phosphorylated upon infection [40], which offers another level of regulation in addition to changes in transcript abundance.

Only one gene was found in common between our data set and those of Norelli et al. [11], Baldo et al. [12], and Sarowar et al [13]. This gene had higher steady-state expression levels in resistant trees and was up-regulated in all three pathogen induction studies. The Arabidopsis homolog of APPLE0F000017734/APPLE0F000060312 (AT2G31880 or SOBIR1) (Tables 3 and 4, Additional File 1, Table S4) encodes a putative leucine rich repeat transmembrane protein that is expressed in response to Pseudomonas syringae infection in Arabidopsis. Overexpression of SOBIR1 in Arabidopsis caused a constitutive upregulation of PR-1 and PR-2, and the plants showed enhanced resistance to P. syringae DC3000, suggesting that elevated levels of SOBIR1 lead to a constitutive activation of disease-resistance responses [41]. SOBIR1 overexpression also resulted in the activation of cell death. It has been proposed that SOBIR1 may play a role in the regulation of the golgi apparatus, particularly during periods of cellular stress [42].

We also identified a transcript encoding a putative cell death regulator, inositolphosphorylceramide synthase (APPLE0F000021750, Additional File 1, Table S4), that had higher expression levels in the least susceptible trees. Inositol can be modified by inositolphosphorylceramide synthase to produce inositolphosphorylceramide. Inositolphosphorylceramide has been shown to be involved in the regulation of programmed cell death during the plant defense response [43].

Jasmonic acid has been shown to play an important role in the response to pathogens, in a pathway parallel to that of salicylic acid [44, 45]. In particular, ethylene and jasmonate have been shown to play an important role in defense against necrotrophic pathogens like E. amylovora [46]. We also identified homologs of the JAZ1 gene in Arabidopsis (APPLE0F000019494/APPLE0F000070531, Table 3, Additional File 1, Table S4) that had greater expression in the least sensitive trees. The JAZ1 protein of Arabidopsis is part of the COI1/JAZ jasmonate receptor complex [47].