Kumquat microarray chip hybridization data were assessed for overall signal intensity and consistency of the expression ratio over all time points, which resulted in the exclusion of chips with inconsistent results. Figure 1 is a scatter plot showing M-values from two different biological replicate-hybridizations with Xcc-inducible targets (Cy5-labeled) and mock inoculated non-infected targets (Cy3-labeled) confirming high data consistency levels (R² = 0.921).

The B2GO program 29 was used to assign GO (Gene Ontology) terms for hits obtained through eBLAST homology searches in NCBI. A general view of the similarity of the query set with the NCBI database, the distribution of the cut off for the e-value as well as the distribution of species with similar sequences are shown in Additional files 1, 2, and 3. The GO annotation score is considered to be more intuitive than regular blast e-values since GO annotation is carried out by applying an annotation rule (AR) on the ontology terms. Additionally, query sequence descriptions are obtained by applying a language processing algorithm that extracts informative names and avoids low-content terms such as "hypothetical protein" or "expressed protein". Using Blast2Go suite default parameters, 1042 probes were provided with GO annotations (Additional file 4). Approximately 25% of the transcripts on the array do not show similarity to proteins present in public databases. Some of these could represent exclusive genes of the citrus or kumquat lineages, but a fraction of these uncharacterized sequences may possibly represent low quality or 3'UTR sequences. Similar percentages of unknown sequences have been reported in other small-scale EST projects 303132 and therefore this pattern can be considered characteristic of this approach. Since a citrus genome sequence is now available, future studies will have a wealth of citrus genomic sequence information that can be utilized to identify kumquat-specific as well as novel citrus genes involved in diverse defence mechanisms 28.

Gene ontology analysis provided an extremely informative snapshot of the Xcc/kumquat interaction. The hierarchical structure for the gene ontology of a group of sequences can be visualized as a tree by means of directed acyclic graphs (DAG) 33. For instance, the molecular functions of the network implicated in the kumquat response to Xcc infection is illustrated in the DAG presented in Figure 2. The graph demonstrates a tree controlled by the Seq filter that organizes the number of nodes to be displayed. Seq is the number of different sequences annotated at the child GO term. On the whole, the biological meaning for different sequences in the data set was best illustrated in terms of three GO gene categories; the biological processes (Figure 3) underlying molecular functions (Additional file 5), and the cellular compartments where proteins were localized (Additional file 6).

An important aspect of the data was that, for many genes, transcript abundance varied over time points, and a number of genes were only up- or down-regulated at one or two time points (Figure 4). Two approaches were used to identify patterns of gene expression. First; the ASCA-gene analysis methodology revealed that most of the total variability in the data was related to time-associated changes 34. According to ASCA, 289 probes were selected as differentially expressed, 172 of which were at statistically significant levels (Additional file 7). Moreover, the time-associated variation could be divided into two main variability patterns. One pattern (accounting for 20% of the variation) represented genes whose expression levels changed significantly at 24 hpi from their levels at 6 hpi and then recovered to values similar to the starting values (or to even greater values in the opposite direction) at 72 hpi. However, the major pattern (80% of the time-associated variation) indicated a strong gene expression change between 6 and 24 hpi followed by preservation of expression levels at 72 hpi. This indicates that the strongest response to infection occurred at 6 to 24 hpi, and the majority of genes maintained their change for up to 3 days with a smaller percentage reverting to initial values.

The second approach, maSigPro analysis, indicated that 317 probes were differentially expressed throughout time, (adjusted p value < 0.05 and R² of the model fit >= 0.6; Additional file 8). The results of both approaches were combined into 433 probes that were then filtered using more stringent conditions to provide a unique result. The union rather than the intersection of the two approaches was taken because the two methods reveal different aspects of the data and are thus complementary. The ASCA-gene methodology focuses on shared gene expression changes to find important genes, while maSigPro treats genes independently and evaluates significant time dependent-changes. Although ASCA-gene methodology may miss some genes whose expression pattern is rare but significant, these will be captured by the gene-wise maSigPro approach. Alternatively, maSigPro can miss genes with less pronounced changes, which can be recovered by ASCA-gene if their profile is abundant within the dataset. The use of both approaches together resulted in the identification of 437 differentially expressed genes 312 of which with acceptable p-values that could be divided into 4 clusters according to their expression patterns (Figure 5, Additional file 9). The criterion for this division is as follows. From the ASCA analysis we obtained the main patterns of variation: Cluster pattern A indicates a strong change in expression between 6 hpi and 24 hpi, which is then maintained at time 72 hpi. Cluster B pattern is comprised of genes differentially expressed at 6 hpi as compared to either 24 hpi or 72 hpi. For each pattern, the correlation of the mean value of each gene at each time point with the profiles indicated by ASCA-gene was calculated; subsequently genes were divided into 4 clusters depending on whether expression levels changed in positive or negative directions. In this analysis, genes cannot be classified simply as induced or repressed, because this depends on the time points considered; for example, genes in cluster pattern C are repressed at 24 hpi and then induced at 72 hpi.

Based on the assumption that altered gene profiles during plant-microbe interactions can be correlated with symptoms, gene ontology and annotation, we believe that Xcc represents a typical example of how the bacterial pathogen can manipulate the host systems in its favour as elucidated previously in different studies 34113536. Information on all of the specific transcripts discussed in the subsequent paragraphs is given in Table 1. Cernadas et al. inoculated 'Pera' sweet orange with either Xcc, which causes typical canker symptoms on this citrus type, or Xanthomonas axonopodis pv. aurantifolii pathotype C (Xaa), which only produces symptoms on Mexican lime, followed by a detailed transcriptional analysis for the sweet orange plants 36. Although the analyses done in that study cannot be directly compared with our study because of differences in methodology, some generalizations are noted below.

The distribution of functions within the significantly expressed genes in Xcc infected kumquats indicates that the highest number of transcripts (~30%) was associated with response to stress, electron transport, and/or oxidative stress (as shown in Figure 3), an indication of an early regulatory changes in the plant immune system by Xcc. Earlier studies, such as that of Cernadas et al., have come to the same perception 36. Each identified cluster was subjected to functional analysis by either studying the distribution of GO terms or performing enrichment analysis to see if there were functional categories that were significantly represented. A total of 137 genes, which makes up more than 30% of the genes that were significantly expressed, were down-regulated in the interval between 6 hpi and 24 hpi,. Most of them were grouped in Clusters A and C (Figure 6). The expression levels of the genes in both of these clusters reached a minimal expression level at 24 hpi followed by either a minor (Cluster A) or major (Cluster C) recovery by 72 hpi (Table 1). For instance, the expression of the thioredoxin f gene homologue (KLLFI3-F09) that belongs to cluster A reached its maximum level of expression (+1.8 fold) by 72 hpi after slight decrease at 24 hpi.The lipoxygenase gene homologue (KSLFII1-F07) that belongs to cluster C was 1.5 fold down-regulated at 6 hpi followed by a 3 fold increase in expression when compared to the its expression level at 6 hpi sample. Genes in Cluster A and C were frequently related to oxidative stress response. Most of the activity for genes in these clusters is located in the mitochondria, the cell membrane and the chloroplast (Figure 6). Cluster B was the largest cluster and included 235 genes with up-regulated expression levels between 6 hpi to 24 hpi followed by sustained expression until 72 hpi (Figure 5, Figure 7A). Cluster D contained 61 members that had a low steady expression up to 24 hpi, and were subsequently upregulated (Figure 5, Figure 7B). This cluster includes genes, such as the glycosyltransferase-like gene (KSLFI7-F12), that mediate the transfer of glycosyl residues from activated nucleotide sugars to acceptor molecules (aglycones), a key mechanism in determining the diversity, activity and chemical complexity of plant natural products. In plants, UGTs (uridine diphosphate sugar glycosyl transferases) generally use UDP-glucose and occasionally UDP-xylose for glucosylation of phenylpropanoid aglycones. Albrecht and Bowman 37 proposed using UGTs and other glycosyltransferases as prospective genetic engineering candidates due to their important role in resistance and tolerance to citrus tristeza virus (CTV) as well as citrus huanglongbing (HLB) in trifoliate oranges (Poncirus trifoliata L. Raf.). Phenolics are mainly synthesized in plants via the phenylpropanoid pathway and are incorporated into many important compounds including plant hormones, secondary metabolites involved in stress, defence responses, and xenobiotics such as herbicides 38. In addition, phenylpropanoid pathway intermediates, for example p-coumaric acid, caffeic acid, ferulic acid and sinapic acid, and pathway derivatives, including flavonoid aglycones and glycosides, exhibit antimicrobial activity 3940.

Fortunella margarita (Lour.) Swingle (Nagami kumquat) plants were used in all of the experiments described in this study. Plants were approximately 2 years old at the time of the experiment and were maintained in the quarantine greenhouse facility at the Division of Plant Industry, Florida Department of Agriculture (Gainesville, FL, USA) under controlled conditions. Leaves from a set of six kumquat plants were infiltrated with bacterial cultures according to Lund et al. 75. The bacterial strain used was Xanthomonas citri subsp. citri A; Miami X04-59 (Xcc). The inoculum was adjusted to 5 × 10⁸ cfu/ml. A similar set of plants was mock-inoculated using sterile tap water as controls. Leaves from the two sets of plants were used in subsequent experiments.

The kumquat microarray chip was developed and printed at the University of Florida (Gainesville, FL, USA). The array included ESTs chosen from 4 previously constructed Nagami kumquat forward and reverse leaf subtraction cDNA libraries. The cDNA libraries were constructed using RNA extracted from leaf tissue collected at different intervals post inoculation with Xcc (see below) and pooled into early and late library sets to capture a wide spectrum of differentially expressed transcripts 24. Random DNA sequencing was performed from 5'and 3' ends of randomly selected clones using universal primers, generating sequence information from 2788 and 1655 clones from the early and late leaf subtraction libraries respectively.

The initial dataset was reduced to a total of 2304 transcripts that were selected according to sequence alignment similarities with proteins in the Genbank database. Sequences were selected based on quality and length. The dataset included 2254 kumquat ESTs comprising 738 contigs and 1516 singletons, in addition to 50 cDNA control elements. Each probe was printed in 3 locations on the array using the Omnigrid Microarrayer (Gene Machines, San Carlos, CA, USA) so that all clones had 3 technical replicates on each slide, generating a total of 6912 spots. Post-printing slide processing was performed as described in Heller et al. (1997) 76 with some modifications. In brief, a combination of sequential baking and UV crosslinking was implemented where slides were baked for 80 min at 80°C in a drying oven without vacuum. The slides were then washed twice in 0.1% SDS for 5 minutes each to remove any unbound DNA.

To identify genes that are considered to be differentially expressed in kumquat, a time-course experiment was designed utilizing the kumquat/Xcc pathosystem. Six independent Xcc or mock-inoculated kumquat plants were used; each plant was considered an independent biological replicate. Since citrus canker is a non-systemic disease, 6-10 leaves per treated kumquat plant were independently infiltrated using 5 × 10⁸ cfu/ml Xcc. All RNA samples isolated from healthy mock- or Xcc inoculated (infected) leaves were processed independently. It is unlikely that differential gene expression observed was caused by the pressure infiltration inoculation method used, since this factor was normalized by treating the mock inoculated plants in the exact same way as infected plants.

Individual leaves were harvested from the inoculated and mock-inoculated plants at specific time-points post-inoculation (pi) according to designated conditions for each experiment; there were 3 time points for the microarray experiment and 5 for subsequent real-time PCR assays. For the microarray experiment, the three time points (6 hpi, 24 hpi, and 72 hpi) were chosen based upon the internal bacterial populations previously detected at these times following inoculation and the knowledge that kumquat leaves abscised 3-5 days after inoculation. In addition, previous experiments revealed that there were some transcripts differentially expressed as early as 30 min post-inoculation with Xcc. Finally, the RNA yield and the abundance of cellular transcripts decreased as the leaves approached total PCD, as has been shown previously by others 53. Time points of 0 and 120 hpi were added for the real time PCR assays. The healthy mock-inoculated and Xcc-inoculated leaf samples were immediately frozen in liquid nitrogen. For each respective time point, total RNA was extracted using RNeasy columns (QIAGEN, Valencia, CA, U.S.A.) according to the manufacturer's protocol. RNA purity, concentration and quality were assessed using a spectrophotometer and a BioAnalyzer 2100 (Agilent Technologies, Palo Alto, CA).

Prior to slide hybridization with probe, slides were pre-hybridized in a solution containing 5× SSC, 0.1% SDS and 1% bovine serum albumin at 42°C for 45 min to eliminate nonspecific binding of the probe to the slide. Slides were washed using MilliQ RNase free water, then isopropanol, once each, and air-dried. Slides were maintained at the hybridization temperature until loaded with probe. cDNA labeling was performed using the Genisphere (Hatfield, PA, USA) 3DNA Array50^® Expression Array Detection Kit according to the manufacturer's protocol for total RNA. For each time-point 125 μg of DNA-free total RNA isolated from an independent plant (biological replicate) per slide was reverse transcribed for each of the mock inoculated and the infected leaf samples separately using Ambion reverse transcriptase (Ambion; LaJolla, CA) in the presence of Genisphere dT primers. Two-step hybridization was performed as follows. The first hybridization, carried out at 48°C overnight, contained 10 μl of the concentrated cDNA (heat denatured probe) made using either the Cy5-RT primer capture sequence or the Cy3-RT primer capture sequence, in Genisphere 2× formamide-based hybridization buffer. Three successive post hybridization washes were performed, first in 2X SSC, 0.2% SDS at 55°C for 10 min, then 2× SSC for 10 min and finally in 0.2× SSC for 10 min at room temperature. The second hybridization (for addition of the dye) was performed using 2.5 μl of either Cy3 or Cy5 dendrimer, 2 μl of high-end differential buffer and 58 μl of hybridization buffer. For each time point, three mock-inoculated samples from three individual plants were labeled with Cy-3 and the Xcc infected samples were labeled with Cy-5 and both were hybridized to the same array. Post-hybridization washes were conducted as performed earlier following the primary hybridization, with the addition of 0.1 ml dithiothreitol (DTT) into the first and second wash solutions to reduce oxidation of fluorescent dyes.

Control measures such as the detection sensitivity level were determined using internal control probes and non-specific control elements. Human genomic DNA, the green fluorescent protein gene, and the lambda control template DNA fragment were included as negative controls. Additionally, cDNAs previously implicated in pathogen defence such as PR1 and NPR1 from Arabidopsis, NDR1 from citrus, were printed 3 times on the array to test the ability of the microarray method to detect changes in gene expression. These were considered to be specific positive controls. In addition, the microarray ratio for each gene analyzed was normalized against the microarray ratio obtained for 18 S.

Agilent's Feature Extraction Software (Agilent Corp., Palo Alto, CA) was used to analyze the microarray data. Data were uploaded into the statistical platform R 77 for statistical analysis and the Limma package was used for pre-processing. Data were Lowes transformed followed by scaling between arrays 78. Two array slides were chosen for each time-point experiment based on the consistency of the signal across the replicates. The fold difference in expression was computed as: 2٨average ratio (2 to the power of the average ratio). cDNAs with an average ratio of 1.0 or higher were considered differentially expressed, which represents a 1.5 fold or higher difference in expression. Statistical analysis was performed using two different approaches. Time-dependent gene expression changes were analyzed by the maSigPro methodology 79. Data were subsequently subjected to ASCA-gene analysis that combines ANOVA and multivariate methods to identify main and secondary patterns of gene expression associated with different experimental factors 34. Statistical analysis identified a number of selected genes that were further grouped into clusters.

Functional information about the ESTs represented in the array was obtained by Blast2GO analysis using default parameters 29. Blast2GO uses Blast and an elaborated annotation algorithm to assign Gene Ontology (GO), Enzyme Code and InterPro functional labels to a set of uncharacterized sequences 27. The functional characterization of these clusters was done by applying the Functional Enrichment (FE) method included in Blast2GO which implements the Gossip algorithm. FE methods assess which functional categories are over-represented within a group of genes in relation to a broader list, in this case the whole kumquat array. Finally, the major induced transcriptional changes considered functional classes as a whole were studied with the PCA-maSigFun method 80. This method combines Principal Component Analysis and maSigPro to characterize the "expression profiles" associated with cellular functions. Sequence data from this work have been deposited in the NCBI Genbank database libraries (http://www.ncbi.nlm.nih.gov/GenBank/index.html), using the BankIt dbEST database, and accession numbers were obtained. [Genbank: GW687757to GW690680]. (See Additional File 11).

Kumquat leaves were infiltrated with Xcc (5 × 10⁸ cfu per milliliter), then total RNA was isolated from inoculated leaves 0,6,24,72, 120 hpi for both the microarray and the quantitative real-time PCR experiment as previously stated in the 'experimental plan'.

Total RNA was isolated separately for each respective time point using the TRIzol reagent (Invitrogen, Carlsbad,CA, U.S.A.) according to manufacturer's instructions. RNA samples were further purified using the RNeasy Plant Mini Kit (Qiagen, Valencia, CA) including DNaseI TURBO DNase (Ambion, Austin, TX, U.S.A.) treatment according to the manufacturer's instructions. RNA quality and quantity was then assessed using microspectrophotometry (Nanodrop Technologies, Inc., Wilmington, DE). cDNA was synthesized from 1 μg of total RNA using Applied Biosystems (High Capacity cDNA Reverse Transcription Kit PN 4368813, 4374966). A TaqMan gene expression assay was then used to validate the transcript accumulation levels of a specific subset of genes from the kumquat microarrays. Reactions were performed in the ABI Prism7900 HT sequence detector (Applied Biosystems, Foster City, CA, U.S.A.). Primers for qRT-PCR were designed using the Primer Express 3.0 software (Applied Biosystems), and data were normalized using a Taqman ribosomal RNA control, in addition to the kumquat 18S ribosomal gene that served as an internal control where each real-time PCR reaction was done in parallel with the 18S primers. For internal controls, a number of genes, for example actin showed inconsistencies. The 18s surprisingly showed coherency throughout the interaction, and this was noticed during the microarrays and was subsequently confirmed with the Realtime PCR (RT-qPCR) 18S expression curve. qRT-PCR was carried out at 50°C for 2 minutes, 95°C for 10 minutes, followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute.