Melon is an important horticultural crop and is rapidly becoming a popular research model for cucurbits, thanks to the increasing availability of genomic resources. Several diseases affect melon production, but molecular investigations of plant-pathogen interactions in melon are still rare . Recently, a melon array resource has been developed, allowing the transcriptomic analysis of several physiological and pathological conditions . However, the melon genome has not been fully sequenced, so alternative transcriptomic approaches allowing novel gene discovery are still important. The cDNA-AFLP method is particularly appropriate for gene expression studies in non-model species and it also allows identification of both plant and pathogen genes expressed during infection, an important pre-requisite for the identification of pathogenicity and virulence factors and thus for the identification of targeted control strategies [44–48, 20]. Our study provides the first large-scale investigation of gene expression changes that occur when melon is infected with FOM, the most important melon pathogen, and is the first to compare compatible and incompatible interactions in the same genetic background (Charentais Fom-2).
Analysis of F. oxysporum f. sp. melonis colonization in melon stems
Because few researchers have investigated FOM infections in melon, the site and timing of recognition is currently unknown, which makes difficult to propose suitable time points for molecular analysis. We therefore began this investigation by characterizing the infection process in melon plants inoculated with avirulent FOM race 1 and virulent race 1,2. Disease progression was monitored using the same approach (plating stem fragments at different time points) that has been successful in tomato . Colonization followed a similar trend to that reported for F. oxysporum f. sp. lycopersici in tomato , i.e. the fungus distribution was discontinuous in all combinations from 2-8 dpi, then continuous from 14-21 dpi with distinct patterns in the incompatible and compatible combinations. From 14 dpi onwards, symptoms became obvious in the compatible interaction as generally reported in the literature [49, 11, 20]. Whereas the two virulent strains fully colonize the stem, colonization by the avirulent strain is reduced, and at 18 and 21 dpi the height reached in stems is significantly lower than that reached at 2 and 4 dpi. These findings suggest that the plant may attack the invading pathogen and reduce its vitality. The data were confirmed by real-time PCR, indicating a progressive reduction in the amount of fungus present at later time points in the incompatible interaction (data not shown). Di Pietro and colleagues  found that, having reached the xylem, the fungus remains exclusively within the vessels using them to colonize the host rapidly, mainly through the production of microconidia rather than mycelia which, in turn, progressively grows inside the xylem inducing vessel clogging. In contrast to this prominent microconidia model [20, 23], studies using GFP-labeled F. oxysporum have shown that neither conidiophores nor microconidia are found in Arabidopsis or tomato xylem [21, 22]. The response to infection may be affected by inoculum concentration, the age of the plant, the duration of exposure to the inoculum, and the type of substrate for plant growth (e. g. sand or soil) [11, 27]. The assessment time points may also play an important role in the picture that emerges of the host/pathogen genetic responses. Nevertheless, differences in the infection process are likely to occur among different formae speciales and between different experimental designs.
Genetic elements of host colonization and pathogenicity
Most transcriptomics studies involving F. oxysporum have focused on the interactions that occur in the xylem, and these studies suggest that the main resistance responses occur within or along the vessels. In this context, genes that are expressed solely in planta and not in artificial culture are the most interesting because they are likely virulence factors (Table 2 and Additional File 2). We identified 195 genes that were expressed in planta, 72 of which (~37%) were not expressed under artificial culture conditions and therefore represent putative virulence factors. Interestingly, only 11 out of 218 genes in cotton plants infected with F. oxysporum f. sp. vasinfectum were expressed specifically in planta . The group of putative virulence factors identified in our analysis included plant cell wall degrading enzymes (CWDEs), represented by five transcripts encoding pectate lyases, endo-1,4 beta xylanases and endo-1,4 beta glucanases, possibly activated by interaction with the host. Among these transcripts, an endo-1,4-beta-xylanase 2 precursor is the only sequence peculiar to race 1, induced in the incompatible interaction, while the other four TDFs are specific to the race 1,2 strains. Like most fungi, F. oxysporum secretes CWDEs during either penetration or colonization . Although the inactivation of individual CWDE- or protease-encoding genes might not have a detectable impact on virulence (reviewed in ), possibly because of functional redundancy, their activity is crucial in the process of fungal colonization. Active fungal growth is also documented by the specific in planta expression of several genes related to carbohydrate and lipid metabolism, among them a squalene synthase involved in sterol biosynthesis. Sterols facilitate normal membrane function by controlling their fluidity, but they have also been implicated as ligands for nuclear receptors directly affecting transcription and signal transduction pathways . Other examples include genes for cytoskeleton components (hypothetical protein similar to actin binding protein and dynamin-2) and a chitin synthase gene. Class V chitin synthase is a pathogenicity determinant in F. oxysporum and a mediator of protection against plant defense compounds .
Three other in planta-specific TDFs seem particularly important in terms of virulence. These represent genes encoding homologs of an avenacinase, a fumonisin 16p, and a siderophore iron transporter (mirB). There is increasing evidence that mycotoxin production may enhance pathogen virulence, especially fumonisins and some trichothecenes . Fumonisin enhances the ability of F. graminearum to cause wheat head blight, one of the most important wheat diseases in the world . It has been reported that mycotoxin production can be induced in fungi following the perception of the oxidative burst produced by the plant in response to infection, and could enhance pathogenicity by reducing the oxidative status of the fungal cell. Interestingly, the gene encoding the fungal toxin fumonisin was strongly and specifically expressed in planta only by virulent strains.
Avenacina is a hydrolytic enzyme that can degrade the oat saponin avenacine, and was first recognized as an essential pathogenicity factor in the take-all fungus Gaeumannomyces graminis var. avenae. Saponins, glycosides with soap-like properties that disrupt membranes, are a class of phytoanticipins. The role of saponin detoxification remains controversial in other plant-pathogen interactions [54, 55]. However, the saponin-degrading tomatinase from F. oxysporum f. sp. lycopersici has recently been confirmed as a virulence factor in tomato, by targeted disruption and over-expression of the corresponding gene . In melon, we found that the avenacinase transcript is not only expressed specifically in planta, but is also differentially expressed between the two 1,2 strains, with higher levels produced by ISPaVe1018. To our knowledge, this is the first evidence to support a role for saponin-detoxifying enzymes in FOM infection. The siderophore iron transporter mirB gene may also represent a virulence factor because siderophores (low-molecular-mass iron chelators that facilitate iron uptake and storage) are crucial for fungal pathogenicity in both animals and plants, and also maintain plant-fungal symbioses . The final group of FOM genes expressed only in planta includes several involved in transport and intracellular trafficking, and three related to signal transduction, with similarity to a calnexin involved in calcium-regulated protein folding , a phosphoserine phosphatase and a MADS box protein.
Although expressed both in planta and in vitro, a peroxisomal biogenesis factor PEX11 and an arginase coding gene are also worth mentioning (Table 2). Peroxisomes are single-membrane-bound organelles which, in filamentous fungi, are involved in the β-oxidation of fatty acids, peroxide detoxification and the occlusion of septal pores . Peroxisomal function and fatty acid metabolism are required for fungal virulence. In F. oxysporum, four different Pex genes (PEX1, PEX10, PEX12 and PEX26) were identified as potential pathogenicity genes in a recent insertional mutagenesis screen, and the requirement for full pathogenicity was verified for two of them (PEX12 and PEX26) by complementation with the intact genes . Arginase regulates the production of nitric oxide (NO), which is induced in a jasmonate-dependent manner in response to wounding and is strongly implicated in the activation of disease resistance genes [60, 31]. In microorganisms, arginase activity has been correlated with pathogenicity  and was shown to act as a bacterial survival mechanism by downregulating host nitric oxide production . Other transcripts expressed by FOM in planta, specifically or otherwise, are involved in ubiquitinylation and protein degradation, both of which are necessary for pathogenicity in F. oxysporum f. sp. lycopersici , and in different aspects of fungal metabolism.
Differentially expressed genes among F. oxysporum f. sp. melonis strains in vitro
One major problem in FOM diagnosis is the identification of isolates at the race level. In melon-growing areas, the introduction of races that can overcome resistance in cultivated genotypes may result in dramatic crop losses. At present, no molecular tools are available to replace the time-consuming race-determination tests. We identified a number of transcripts with differential expression profiles between the two races (Additional File 4). Although differences in gene expression cannot be used directly as genetic markers of race identity, TDFs could be used as 'fingerprints' for this purpose. In addition, the differential virulence of the two 1,2 strains demonstrated by the host colonization pattern, could also be fingerprinted using TDFs that are differentially expressed between ISPaVe1018 and ISPaVe1083. Unfortunately, most TDFs in this category either matched hypothetical protein sequences in public databases or did not generate hits at all, and therefore do not allow speculation about the possible metabolic differences between the two races or between the two strains of FOM race 1,2.
Large-scale transcriptional changes underlie disease development
Transcriptional changes associated with resistance responses occur within the first 2 dpi, and are maintained with few changes thereafter (Figure 4). However, only 11 melon transcripts (Cluster A) are specific for the incompatible interaction. The largest group of modulated genes (Cluster D) is expressed in a non-specific manner, with variable modulation throughout the experiment, in both the incompatible and compatible interactions. The establishment of compatibility is characterized by a slightly delayed but progressive increase of the number of genes involved, underlying the significant metabolic disturbances that might be associated with symptom development. The majority of these changes are included in Cluster D and are thus non-specific up to 8 dpi, but are followed by a sudden wave of susceptibility-specific transcriptional changes at 21 dpi, almost completely conserved between the virulent strains ISPaVe1018 and ISPaVe1083.
Although the precocity of the resistance response is expected, the small number of genes involved is unexpected. Incompatible interactions commonly involve large-scale transcriptional reprogramming toward defense, which is generally more intense and rapid than in corresponding compatible interactions [63, 64, 28]. However, vascular diseases may represent a peculiar situation, in which symptom development and consequent damage could depend not only on the pathogenetic activity of the fungus but also the strength and timing of the host response. This was indicated by pioneering research in which delayed formation of tyloses in susceptible genotypes eventually contributes to vessel clogging . In agreement with the above, our data suggest that more striking changes in gene expression accompany disease and symptom development than resistance, thus resistance might depend more on the ability to tolerate the infection, avoiding reactions.
Transcriptional changes in the compatible interaction
Although cluster analysis and functional annotation of the identified melon transcripts provided an overview of the transcriptional changes occurring in infected melon plants, the limited availability of sequence data made it difficult to draw firm conclusions about the molecular events occurring in infected plants. Of the 115 TDFs identified in Cluster C, 41 did not match any database sequences or matched sequences that have yet to be annotated. The remaining 74 TDFs encode well-known components of disease resistance responses and related signal transduction cascades, such as calmodulin and calmodulin-binding proteins, transcription factors, a 12-oxophytodienoate reductase, and a 13S-lipoxygenase involved in jasmonic acid biosynthesis , and enzymes involved in the biosynthesis of secondary metabolites acting as antimicrobial compounds, or in a general stress responses, such as xanthine dehydrogenase and betaine aldehyde dehydrogenase [66, 67]. Genes encoding pathogenesis-related proteins such as endochitinase, beta-1,3 glucanase and a type I proteinase inhibitor-like protein were also specifically modulated in the compatible interaction. Altogether, transcripts related to the defense, response to stimulus and secondary metabolism categories accounted for ~25% of modulated TDFs in Cluster C. These findings further support the hypothesis that a delayed defense response might indeed be responsible for symptom development.
Cluster C also contained genes potentially involved in the establishment of susceptibility, such as those related to auxin accumulation. Several reports indicate that an increase in auxin levels in the cell can contribute to disease susceptibility  and that a similar increase can be induced by pathogens in order to facilitate colonization. TDFs with homology to an indole-3-acetic acid-amino synthetase and to an IAA-type protein Q75GK0  are specifically induced in the compatible interaction. However, other genes in Cluster D that induce auxin signaling (e.g. auxin response factor 6 and AUX1-like auxin transport protein) are repressed by both virulent strains, but induced by the avirulent strain at 21 dpi. The overall picture is therefore complex and suggests that the compatible interaction mainly involves transcriptional changes that are otherwise typical of effective resistance responses. It is tempting to speculate that the recessive resistance identified in Asian accessions might be related to the lack of a plant reaction and thus to better tolerance of the infection process .
Transcriptional changes in the incompatible interaction
Resistance responses are generally characterized by rapid and extensive reprogramming of transcriptional activity, especially in race-specific interactions. However, that resistance of Charentais Fom-2 to FOM race 1, although mediated by a single R-type resistance gene, is not complete, since the fungus can always be reisolated from the stem of Charentais Fom-2 plants. In our model system we noted surprisingly few transcriptional changes specifically associated with the incompatible interaction (11 TDFs, Cluster A). These included a calmodulin-related protein, stably upregulated from 4 dpi onwards, which plays an important role in the transduction of calcium signaling and could be involved in the resistant response (as already demonstrated in other species ) and a catalase that may protect the pathogen from the strong oxidative burst associated with resistance.
Additional information about the resistance response could be found by analyzing genes that are modulated in both interactions (Cluster D) but with a peculiar pattern (e.g. earlier or more strongly in the incompatible interaction), or those showing a reverse modulation of gene expression in one or the other interactions (which might reveal specific requirement in each interaction). An interesting finding is that many of the genes modulated earlier or more strongly during resistance are classified as signal transduction-related genes (Additional File 1). These include several transcripts involved in calcium signaling (calcium-transporting ATPase, calcium dependent protein kinase and calcium-binding protein), transcription factors, kinases and a homolog of NDR1 (non race-specific disease resistance 1). This last gene is induced at 2 and 4 dpi following infection with race 1 and only at 8 dpi following infection with race 1,2. NDR1 was originally identified in Arabidopsis as a factor required for resistance to both bacterial and fungal pathogens  and it is known to mediate resistance controlled by R genes of the nucleotide binding site leucine rich repeats (NBS-LRR) class, which is distinct from the Toll/interleukin receptor (TIR) class . Fom-2 in Charentais Fom-2 plants is indeed a non-TIR R gene, although with a peculiar structure that lacks the typical N-terminal coiled-coil domain . Therefore, it seems plausible that its action might require NDR1.
Other genes modulated earlier in the establishment of resistance include two adenosylhomocysteinases, an aspartokinase-homoserine dehydrogenase and a serine carboxypeptidase. An Arabidopsis adenosylhomocysteinase encoded by the gene HOG1 (homology dependent gene silencing 1) is required for DNA-methylation gene silencing . The involvement of RNA-silencing machinery in plant innate immunity has recently been demonstrated not only against viruses but also bacterial and fungal pathogens, including Verticillium in Arabidopsis [74, 75]. The same transcript increases in aphid-infested sorghum plants . The potential involvement of these candidate genes in Charentais Fom-2- controlled resistance could be the object of future investigations.
Most genes in Cluster D are not differentially modulated in the incompatible and compatible interactions. Interesting examples include TDFs related to ACC oxidases. These enzymes participate in the last step of ethylene biosynthesis and are involved in the response to stress and to pathogens, but are also implicated in senescence, necrosis and disease development. Ethylene has been associated with both wilting and resistance against vascular diseases . We detected four transcripts with similarity to ACC oxidases. These showed variable expression profiles, but there was no difference between compatible and incompatible interactions, which suggests that ethylene might be involved in both susceptibility and resistance. In melon, different ACC oxidase genes are induced differentially during development and pathogen infection . The same variable modulation has been detected for other transcripts possibly related to jasmonate biosynthesis, such as allene oxide synthase and the 13S-lipoxygenase mentioned above.