Secretome analysis reveals effector candidates associated with broad host range necrotrophy in the fungal plant pathogen Sclerotinia sclerotiorum

Background The white mold fungus Sclerotinia sclerotiorum is a devastating necrotrophic plant pathogen with a remarkably broad host range. The interaction of necrotrophs with their hosts is more complex than initially thought, and still poorly understood. Results We combined bioinformatics approaches to determine the repertoire of S. sclerotiorum effector candidates and conducted detailed sequence and expression analyses on selected candidates. We identified 486 S. sclerotiorum secreted protein genes expressed in planta, many of which have no predicted enzymatic activity and may be involved in the interaction between the fungus and its hosts. We focused on those showing (i) protein domains and motifs found in known fungal effectors, (ii) signatures of positive selection, (iii) recent gene duplication, or (iv) being S. sclerotiorum-specific. We identified 78 effector candidates based on these properties. We analyzed the expression pattern of 16 representative effector candidate genes on four host plants and revealed diverse expression patterns. Conclusions These results reveal diverse predicted functions and expression patterns in the repertoire of S. sclerotiorum effector candidates. They will facilitate the functional analysis of fungal pathogenicity determinants and should prove useful in the search for plant quantitative disease resistance components active against the white mold. Electronic supplementary material The online version of this article (doi:10.1186/1471-2164-15-336) contains supplementary material, which is available to authorized users.


Background
The white mold fungus Sclerotinia sclerotiorum (Lib.) de Bary is a cosmopolitan necrotrophic pathogen infecting over 400 plant species. It is among the most devastating pathogens of soybean, rapeseed and sunflower, causing several hundred million dollar losses annually at the preand postharvest stages [1]. S. sclerotiorum host range is remarkably broad, with fruit and vegetable productions also being severely impacted [2]. S. sclerotiorum and its close relative the grey mould fungus Botrytis cinerea are among the few fungal pathogens considered as typical necrotrophs. As such, they derive energy to complete their life cycle mostly from dead plant cells, as opposed to biotrophs that feed on living plant cells.
There is now ample evidence that biotrophic and hemibiotrophic fungi secrete specialized effector proteins manipulating host cell physiology to obtain nutrients, suppress plant defense and ultimately promote infection [3]. Effectors may also trigger plant defense responses, leading to plant resistance, when recognized directly or indirectly by the plant immune system in a gene-for-gene relationship. This results from a co-evolutionary arms race between pathogen effectors, their plant targets, and components of the plant immune system [4]. Necrotrophs have long been considered as less adapted, secreting mostly degrading enzymes and toxins that unspecifically trigger programmed cell death (PCD) and dismantle plant cells.
However, host specific necrotrophs such as Cochliobolus victoriae secrete effector proteins translocated into plant cells that interact with specific corresponding host proteins to facilitate disease progression [5,6]. This involves the activation of plant PCD instead of its suppression as in the case of infection by biotrophic pathogens. S. sclerotiorum also produces the non-proteic pathogenicity determinant oxalic acid. This molecule induces the synthesis of reactive oxygen species (ROS) and triggers plant PCD late during infection, but has the opposite effect, suppressing ROS burst and PCD induction, at the early stages of infection [7]. The SSITL secreted integrin-like protein of S. sclerotiorum promotes virulence and delays the activation of plant defense responses, supporting the view that S. sclerotiorum secretes effectors to finely manipulate plant physiology [8]. In addition, enzymes secreted by necrotrophs can act as virulence factors independently of their catalytic activity [9]. Effector repertoires vary considerably, notably according to pathogens lifestyle [10], and it is becoming clear that interactions between necrotrophs and their host plants are considerably more complex and subtle than previously considered. What is the effector candidate repertoire associated with broad host range necrotrophy remains unclear. As a first step towards elucidating the molecular bases for colonization by S. sclerotiorum, its repertoire of effector candidates needs to be determined.
The recent release of genome sequences for a number of plant pathogenic fungi facilitated the search for effector candidates (ECs) at the genome level [11]. Nevertheless, considering that pathogen effector repertoires are typically lineage-specific, the identification of effectors remains challenging [4]. The analysis of S. sclerotiorum genome sequence uncovered sets of genes associated with the manipulation of redox status, including enzymes of OA biosynthesis, the degradation of plant cell wall, and 603 secreted proteins with other functions [12]. Known hallmarks of fungal effectors such as the presence of signal peptides and absence of transmembrane domains, small size and amino-acid composition generally produce lists of hundreds of potential effectors in a given pathogen. Therefore, more sophisticated approaches are required to pinpoint the most relevant ECs for the promotion of infection in S. sclerotiorum secretome.
A limited number of known fungal effector families show conservation at the sequence level or similar predicted functions. This is notably the case for the toxin and cell death elicitor proteins of the Necrosis and ethylene-inducing Like Proteins (NLPs), the ceratoplatanin, cyanovirin-N homology (CVNH) and ECP2 families [13][14][15][16]. The growing number of characterized fungal effectors suggests conservation at the biochemical function level in the overall effector repertoire of several fungal pathogens. The ability to bind chitin or other cell wall oligosaccharides, masking the presence of the pathogen or dampening damage-induced plant responses, is a feature common to effectors from multiple fungal pathogens [17][18][19][20]. Fungal effectors harboring a protease inhibitor activity are also common [21][22][23][24]. The biochemical activity of a few other fungal effectors such as M. oryzae Fungalysin metalloprotease AvrPita [25], U. maydis chorismate mutase cmu1 [26] and peroxidase inhibitor PEP1 [27] may also be part of the arsenal of effector functions in multiple fungal lineages. This hypothesis suggests that thorough annotation of protein domains and prediction of biochemical function of secreted proteins may prove useful to identify novel effectors in S. sclerotiorum.
However, a majority of effectors do not show significant similarity to known sequences in other organisms nor obvious protein domains. Yet other genomic characteristics may help identify EC genes. The rapid evolution of effector genes allows the fungi to overcome selection pressures induced by resistant plant cultivars. A high ratio of non-synonymous over synonymous substitutions (Ka/Ks) in alleles from related strains is a frequently used proxy for inferring fast gene evolution and the action of positive selection [28]. This approach has been used to reveal ECs in several filamentous plant pathogen lineages [29][30][31][32][33][34]. Positive selection has been detected in B. cinerea genome [35] suggesting that it may be used to mine S. sclerotiorum genome for ECs. Second, gene duplication is another hallmark of several known fungal effector genes, such as the ToxB host specific toxin of Pyrenophora tritici-repentis [36,37]. Third, genomic regions with high repeat and transposable element content are enriched in effector genes in several lineages of plant pathogens [37][38][39] suggesting that genome architecture analysis can assist in the search for EC genes. Finally, effectors can alter host cell function by mimicking plant peptides [40]. These ECs likely elude functional annotation on the basis of primary amino-acid sequence, but may be revealed using three-dimensional structure prediction.
As opposed to Oomycete pathogen genomes in which many effector genes can be identified through conserved sequence motifs [29,41], the use of conserved sequence motifs, such as the Y/F/WxC motif [42], has proven limited in revealing fungal effectors across lineages. However, the presence of a signal peptide directing protein secretion and gene expression in planta are relatively universal properties of effectors that can be exploited as first filters to narrow down the list of effector candidates in fungal genomes.
S. sclerotiorum effector proteins would be useful as probes to search for resistance components in plants and to design strategies for inhibiting infections by this devastating but poorly characterized pathogen. In this study, we report a diverse repertoire of S. sclerotiorum effector candidates revealed by an in depth analysis of its predicted secretome. We combined refined secretome annotation, phylogeny, selection and gene duplication analyses, and three-dimensional structure prediction to identify 78 ECs. Among those, we highlight a predicted subtilisin inhibitor, a xylanase, a duplicated gene of unknown function and three toxin mimics as high priority candidates for functional studies. We analyzed in planta expression pattern for 16 EC genes and revealed hostblind and host-regulated ECs.

Definition and annotation of S. sclerotiorum secretome
In the original analysis of S. sclerotiorum genome, secreted proteins were predicted using SignalP, TargetP and TMHMM, and annotated using Interproscan [12]. To identify and explore candidate effectors in the genome of S. sclerotiorum, we built a bioinformatics workflow exploiting genomic features typical of known filamentous plant pathogen effectors. We first refined the prediction of secreted proteins combining predictions from SignalP2 and 4 to identify 1070 proteins with a secretion signal. To evaluate the sensitivity of this prediction, we applied the same methodology on a list of 1985 eukaryotic secreted proteins validated experimentally from the SPdb [43]. We retrieved 1971 proteins predicted as secreted, corresponding to a true positive rate of 99.34%. Next, we removed 172 proteins predicted by TMHMM to harbor transmembrane helices, then 153 proteins predicted by GPIsom to harbor glycophosphatidylinositol anchor motifs, that likely represent surface proteins rather than secreted effectors. This resulted in a list of 745 predicted secreted proteins. To increase the likelihood of identifying genes relevant for infection, we selected those for which there is evidence for expression in planta based on publicly available EST and microarrays data. We considered genes expressed, even if not induced, during interaction with rapeseed, sunflower or tomato (see Methods). Although this filtering pipeline likely excluded relevant effector candidates, it resulted in a total of 486 genes encoding predicted secreted proteins expressed in planta (SPEPs, Figure 1a, Additional file 1: Table S1).
Next, we performed three different effector-oriented analyses on the 486 SPEP genes. First, we used Blast2GO, PFAM domain and nuclear localization signal (NLS) searches to annotate 326 SPEPs (Figure 1b). We built a database of known fungal effectors and explored the literature to select 31 S. sclerotiorum ECs among annotated SPEP genes. Second, we defined clusters of orthologous genes (COGs) between S. sclerotiorum and B. cinerea predicted genes using Inparanoid (Figure 1c). A total of 197 SPEP genes grouped in COGs. We aligned S. sclerotiorum and B. cinerea orthologs for these 197 gene pairs and calculated ratios of non synonymous over synonymous substitutions (Ka/Ks) to identify five ECs with signature of positive selection. The 289 SPEP genes with no ortholog in B. cinerea were grouped into clusters based on sequence similarity to identify 29 ECs distributed in 24 families containing genes duplicated in S. sclerotiorum. Finally, we analyzed the taxonomic distribution of SPEP genes across the kingdom Fungi using BlastP searches against a database of 14 complete genomes representative of all major fungal lineages. This identified 70 S. sclerotiorum-specific SPEP genes, most of which had no annotation ( Figure 1d). We used protein structure prediction and pattern and fold recognition searches to identify 17 ECs analogous to known protein fold encoded by S. sclerotiorum-specific SPEP genes.
Using effector-oriented analyses, we identified four lists of ECs, containing a total of 78 EC genes (four being common to two lists). We could not predict any enzymatic activity encoded by 33% of the SPEP genes (160), suggesting that S. sclerotiorum effector repertoire encodes diverse functions that are not restricted to plant cell degrading enzymes. Besides, 59.5% of the SPEP genes (289) did not cluster in B. cinerea COGs, revealing a relatively high degree of divergence from this closely related fungal pathogen.

Sclerotinia effector candidates showing conserved domains
Some fungal effectors show conserved protein domains and biochemical functions that can be identified at the protein sequence level. In a first approach to identify ECs, we used (i) PFAM annotations, (ii) nuclear localization signal (NLS) predictions, and (iii) BlastP searches against known fungal effectors. We found 326 annotated SPEP genes among which we selected 13 as effector candidates based on the presence of specific PFAM domains (Table 1). Indeed, six predicted SPEPs (SS1G_00642, SS1G_02014, SS1G_04786, SS1G_09392, SS1G_12336, SS1G_14184) included chitin-recognition or chitin-binding domains (PF00187, PF03067) and one SPEP (SS1G_12509) contained LysM domains (PF01476). SS1G_07836 was annotated as a concanavalin A-like lectin/glucanase that binds to complex carbohydrates, and harbors a Peptidase A4 (PF01828) with a typical lectin fold. SS1G_08698 contains a Ricin type beta-trefoil lectin domain (PF00652) typical of ricin-like toxins. These predicted SPEPs are relevant ECs considering that chitin-and carbohydrate-binding activity has been reported for several fungal effector families. We also selected three predicted SPEPs (SS1G_01593, SS1G_03282, SS1G_12605) with protease inhibitor domains (PF05922), another activity found in known fungal effectors. We identified two predicted SPEPs (SS1G_ 03611 and SS1G_13935) with a CFEM cystein-rich fungal effector motif (PF05730). Second, we selected another 11 SPEP genes based on the presence of at least one predicted NLS in their sequence ( Table 1). The co-occurrence of a secretion signal and a NLS in encoded proteins suggest that they may be active in the plant nucleus.
Consistent with a nuclear activity, SS1G_01866 and SS1G_05895 also harbored transcription factor domains. Third, we selected 7 predicted SPEPs that were retrieved based on sequence similarity to known effectors or ECs from other plant pathogenic fungi (Table 1). We recovered the two necrosis and ethylene inducing peptides SsNEP1 and SsNEP2 (SS1G_03080 and SS1G_11912) described in [44] sharing~40% identity with various NEP-like proteins, and harboring a characteristic PFAM domain (PF05630). We selected SS1G_00849 for sharing 59% identity with Colletotrichum hingginsianum effector candidate 91 (CHEC91, [45]). There was no protein domain identified in this protein but three-dimensional structure prediction indicates that it is analogous to Alternaria alternata AltA-1 allergen. We selected SS1G_02904 for sharing 35% identity with C. hingginsianum effector candidate 80 (CHEC80, [45]). It features a Cyanovirin-N homology domain (PF13639) that corresponds to a carbohydratebinding module [15,46]. We identified SS1G_08858 showing 25% identity with M. oryzae effector AvrPita (MGG_15730, [25]) and harboring a M35 metalloprotease domain (PF02102). Finally we selected SS1G_10096 as similar (>60% identity) to pathogen-associated molecular patterns (PAMP) with a cerato-platanin domain (PF07249) [14,45,47,48]. These 31 ECs identified based on protein annotations showed that the activity of S. sclerotiorum secretome may not be limited to typical cell death elicitors but is predicted to cover a wide range of functions known for fungal effectors, including chitin binding, proteases and protease inhibitors.

A family of subtilisin-inhibitor effector candidates conserved in Ascomycetes
Effectors with a protease inhibitor activity have been described in fungal plant pathogens with a biotrophic phase of infection [22,49]. We identified three S. sclerotiorum candidate effectors (SS1G_01593, SS1G_03282 and SS1G_ 12605, Table 1) with a serine protease inhibitor I9 domain (PF05922) that corresponds to the propeptide inhibitor domain of subtilisins. To document the taxonomic distribution of SS1G_01593 homologs in fungi, we performed a BlastP search against the predicted proteome of 234 fungal species. Using signal peptide predictions, we identified 99 secreted homologs across 97 species. Secreted homologs of SS1G_01593 are restricted to Ascomycetes and found in all Leotiomycete species considered (Figure 2a). At the sequence level, homologs of SS1G_01593 are clearly distinct from C. fulvum Avr2, U. maydis Pit2, Uromyces fabae RTP1p and M. lini AvrP123, suggesting that SS1G_01593 family represent a distinct class of protease inhibitor effectors. To get insights into SS1G_01593 function and evolution, we predicted the 3D structure of the protein and mapped residue conservation among the 99 secreted homologs on this structure. SS1G_01593 protein is predicted to adopt the structure of subtilisin pro-domain consisting of 4 beta-sheets facing two alphahelices ( Figure 2b). The beta-sheets, interacting with the catalytic domain in subtilisins, are well conserved in fungi, whereas the exposed alpha-helices are more variable, suggesting that the ability to interact with subtilisins is conserved.

Sclerotinia effector candidates showing high Ka/Ks ratios
According to the red queen hypothesis, fungal pathogen effectors are predicted to evolve rapidly. Signatures of natural selection were used to identify effectors and elicitors of immune responses in bacterial, oomycete and fungal plant pathogens [29,35,50,51]. To detect signatures of natural selection in SPEP genes, we calculated Ka/Ks ratios using Yn00 [28] on alignments between S. sclerotiorum and B. cinerea 197 core ortholog pairs. We obtained Ka/Ks values ranging from 0.009 to 6, with a median value of~0.1. Notably 5 genes (2.5%) showed Ka/Ks > 1, suggesting positive selection ( Figure 3a, Table 2). To identify codon sites under positive selection in the five SPEP genes showing Ka/Ks > 1, we used a Bayesian inference approach on alignments of S. sclerotiorum 1980, B. cinerea B05.10 and B.cinerea t4 orthologs. We could not detect sites with Ka/Ks > 1 in SS1G_04551 and SS1G_07158 using this dataset. We detected 7, 50 and 20 sites with Ka/Ks > 1 in SS1G_07749, SS1G_10165 and SS1G_10617 mature proteins respectively. The pvalue for positive selection using M8 and M8e models was 8.2e −2 , 3.8e −4 and 3.0e −1 for SS1G_07749, SS1G_10165 and SS1G_10617 respectively (Table 2). Interestingly, SS1G_07749 encodes a putative xylanase that may function as a virulence factor such as xylanases from other fungi [9,52]. To get insights on the constraints shaping the evolution of SS1G_07749 xylanase, we predicted the 3D structure of SS1G_07749 mature protein and mapped the local Ka/Ks ratios on this structure ( Figure 3b). All seven selected sites correspond to surface-exposed residues with three residing at the surface potentially interacting with plant xylanase inhibitors (R36, D74, Q76), three at the surface of the region corresponding to B. cinerea Xyn11 necrotizing peptide (N174, L189, S200) and one (L43) modifying the surface of the substrate-binding pocket. These findings are consistent with adaptation to host and the possible involvement of SS1G_07749 in virulence.

Sclerotinia effector candidates encoded by recently duplicated genes
Several effectors of filamentous plant pathogens evolved through gene duplication followed by rapid diversification  The predicted 3D structure of SS1G_01593 with residues color-coded based on conservation in fungi. [16,51]. To identify SPEP genes that underwent duplications since the divergence of S. sclerotiorum with B. cinerea, we grouped proteins from their whole predicted proteomes based on sequence similarity using Markov clustering, and selected groups matching all three criterions: (i) containing at least 3 proteins, (ii) containing at least one S. sclerotiorum SPEP gene and (iii) containing more than 50% of proteins from S. sclerotiorum. We obtained 24 such groups including a total of 71 S. sclerotiorum proteins among which 29 S. sclerotiorum SPEP genes ( Figure 4a, Table 3). Groups 051, 170, 175 and 204 contained only secreted proteins and probably carry effector-specific functions. Group 182 and 167 contained SPEP genes identified as ECs based on the presence of CFEM and LysM domains respectively (Table 1). SPEP genes from group 51, 80 and 204 had no conserved domain or homology to proteins of known function. In some pathogen genomes, the expansion of effector gene families is associated with the proximity to transposable elements [4]. To determine the repeat environment of S. sclerotiorum ECs, we annotated repeats in S. sclerotiorum genome, and calculated for all genes the distance to the nearest repeat. Median distance to repeats is 7.89 Kbp across S. sclerotiorum genome and 3.56 Kbp for the 29 ECs that underwent recent duplications. Nine duplicated ECs are located less than 2Kbp apart from a retrotransposon and may have undergone duplication due to transposable element activity (Figure 4b, Table 3). Cluster051 is remarkable for containing only genes coding for secreted proteins, with only one from B. cinerea and 6 from S. sclerotiorum, such as the SPEP gene SS1G_13371 (Table 3). To get support for the duplication of SS1G_13371 gene ancestor in S. sclerotiorum lineage, we constructed a phylogenetic tree of SS1G_13371 homologs. A total of 58 homologs in 25 fungal species could be retrieved from the JGI database covering 238 complete fungal genomes. We selected the 20 closest homologs to build a parsimony tree based on a 90 amino-acid alignment (Figure 4c). For the 6 S. sclerotiorum genes of cluster 051, the phylogeny revealed clustering based on paralogy rather than orthology, suggesting that this family expanded after the separation of the 25 species analyzed. Gene duplication in this family may have allowed increased accumulation of the corresponding protein, neo-functionalization in some paralogs, or differential regulation.

S. sclerotiorum-specific effector candidates with toxin structural analogs
The emergence of virulence is frequently associated with a high rate of mutation, gene gain and gene loss in effector genes. This evolutionary pattern results in a discontinuous taxonomic distribution for effector genes [4,55]. To identify genes showing a discontinuous taxonomic distribution among S. sclerotiorum SPEP genes, we looked for homologs of all 486 S. sclerotiorum SPEP genes in the complete genome of 13 fungal pathogens with a necrotrophic phase of infection covering all major Residues of the 3D model are color-coded according to site-specific Ka/Ks ratios calculated using Bayesian inference with M8 model [28]. Residues with Ka/Ks > 1 are labeled on the structure. A putative Beta-D-Xylopyranose substrate molecule present in the 3b5l_B model, best structural analog of SS1G_07749, is shown as balls and sticks. The side chains of residues forming the predicted substrate binding site predicted by COFACTOR are show as sticks. The interface with plant xylanase inhibitors shown as a yellow dotted line was inferred from [53,54] and the necrotizing peptide region shown as a grey dotted line was inferred from [9]. fungal lineages. As expected, the number of SPEP homologs identified in a given genome decreased with phylogenetic distance to S. sclerotiorum, with 395 SPEPs (81.3%) having homologs in B. cinerea and only 90 (18.5%) having homologs in Cryptococcus neoformans (Figure 5a). Forty-six SPEP genes had homologs in all 14 fungal species monitored, and 70 were S. sclerotiorumspecific, absent in the 13 other fungal species considered (Figure 5b). Among those, only six (8.6%) had weak matches to PFAM domains (0.004 < e-value < 0.74) other than domains of unknown function. We could not obtain information about the function of the 64 remaining SPEP genes based on annotation or homology.
To get insights into the putative function of S. sclerotiorum-specific predicted SPEPs with no annotation, we performed 3D protein structure predictions followed by fold-recognition searches with an aim to identify structural analogs of known function. We focused on 17 SPEP genes encoding mature proteins of at least 50 amino-acids and for which the best BlastP hit is not a B. cinerea protein (Table 4). We closely examined the predicted structures of three SPEPs in comparison with their structural analogs. SS1G_09512 had analogy to the lectin domain of lectinolysin, a toxin forming pores in cell membranes regulated by sugar-binding through a lectin domain [56] (Figure 6a). SS1G_12769 had analogy to the hookworm saposin-like protein Na-SLP-1, a toxin forming pores in membranes through lipid-binding activity [57] ( Figure 6b). SS1G_13235 has analogy to the C-terminal domain of death-associated protein 5 (DAP5), a member of the eukaryotic translation initiation factor eIF4G protein family (Figure 6c). Cleavage of DAP5 by caspases at its C-terminus induces apoptosis [58]. SS1G_07354 also had eIF4G protein as closest analog (Table 4). We propose that these EC are toxin analogs that may have emerged through convergent evolution and could contribute to S. sclerotiorum virulence.

S. sclerotiorum effector candidate genes show diverse patterns of expression in planta
To test whether ECs identified in this work could play a role during interaction with host plants, we monitored the expression pattern after plant inoculation by quantitative RT-PCR for 16 EC genes, representative of the four lists of ECs identified through our bioinformatics analyses. For this, we inoculated four different host and model plants, including Tomato, Nicotiana benthamiana, Arabidopsis thaliana resistant accession Rubezhnoe and susceptible accession Shahdara, and harvested samples at 6, 24 and 48 hours post-inoculation (hpi). S. sclerotiorum mycelium grown in vitro was used as a reference. We included the S. sclerotiorum ubiquitin-conjugating enzyme gene SS1G_11173 as non-induced control gene and the cerato-platanin gene SS1G_10096 as an in planta-induced control [14]. Eight effector candidate genes showed in planta induction >2 on all four host plants tested. One gene (SS1G_07295) did not show in planta induction ≥2 fold on any plant. Using hierarchical clustering, we grouped effector candidate genes into five clusters according to their expression pattern (Figure 7a). Genes in cluster I show strong induction in planta over at least two time points; genes in cluster II showed strong and early  (Figure 7b). At 24 hpi, extensive cell death was visible on leaves of A. thaliana accession Shahdara whereas only limited cell death symptoms were visible on N. benthamiana (Additional file 2: Figure S1), suggesting that the activation of host cell death is not the only determinant of S. sclerotiorum EC induction. We observed a consistent 2-to 4-fold induction between 6 and 24 hpi for SS1G_06213 on all four host plants tested. By contrast, SS1G_08858 was induced >4-fold at 6 hpi on N. benthamiana, at 24 hpi on tomato, and at 48 hpi on A. thaliana (Figure 7c). This result suggests that S. sclerotiorum possess effector genes that are regulated independently of the host being colonized and others that are differentially regulated in a host-dependent manner. Furthermore, SS1G_13371 was induced >2 fold during infection of A. thaliana resistant accession Rubezhnoe, but not during infection of the susceptible accession Shahdara. Conversely, SS1G_00849 was induced >8 fold during infection of the susceptible accession Shahdara, but only~4 fold during infection of the resistant accession Rubezhnoe (Figure 7d). This data points towards a versatile repertoire of effector candidates the expression of which can be modulated according to the nature of the host plant being colonized.

Discussion
In their global analysis of S. sclerotiorum genome, Amselem et al. [12] identified 603 genes encoding non-CAZYme, non-peptidase secreted proteins. These secretome genes did not appear significantly enriched in genes induced in planta. In this study, we combined multiple bioinformatics approaches to identify a total of 745 predicted secreted proteins, among which 486 with experimental evidence for expression in planta (SPEPs). The predicted SPEPs include SsNEP1 and SsITL1 that have proposed to be S. sclerotiorum virulence factors [8,44]. Since we have chosen to focus the search for effector candidates on these 486 SPEP genes, we have deliberately ignored genes expressed in planta for which experimental evidence is lacking, and enzymes that contribute to the biosynthesis of secondary metabolites as virulence determinants. It is therefore expected that the diversity of S. sclerotiorum virulence factors exceeds that of the candidate effectors presented here. Sequence similarity to known fungal effectors is a powerful method to uncover effector families conserved across species [16], that allowed us to identify S. sclerotiorum homologs of B. cinerea NEP1 [44] and Spl1 [14], M. oryzae MGG_15370 and C. hingginsianum CHEC91 and CHEC80. To complement this approach, we used PFAM domain and NLS motif searches to reveal additional effector candidates. We identified putative chitin-binding proteins, putative protease inhibitors, cystein-rich proteins and putative nuclear localized proteins. Effectors with chitin binding activity such as C. fulvum Ecp6, M. oryzae Slp1 and M. graminicola Mg3LysM function in suppressing plant immunity [18,20,59]. Similarly, SsITL1 (SS1G_14133) integrin-like secreted protein suppresses plant jasmonic acid and ethylene signaling pathways and enhances susceptibility [8]. These findings suggest that S. sclerotiorum secretes proteins able to suppress plant immunity. The comparative analysis of Fusarium graminearum secretome and genomes of other ascomycetes revealed a high level of conservation with only 25 F. graminearum specific out of 574 secreted proteins [60]. The taxonomic distribution of S. sclerotiorum SPEP homologues analyzed in this work supports the conservation of more than 50% of SPEP genes across ascomycetes. As proposed by Brown et al. [60], these core SPEP genes may support S. sclerotiorum epiphytic growth and highlight important distinctions between multiple phases in infection by this fungus [61]. Nevertheless it also revealed 70 SPEP genes (14%) specific to S. sclerotiorum, many of which are unannotated proteins. The systematic prediction of their 3D structure allowed identifying putative structural analogs of some predicted SPEPs and suggests that they may carry unique functions to assist S. sclerotiorum pathogenicity. It will be interesting to take advantage of these predictions to test the biological function of these effector candidates and confront them to experimentally determined structures. Furthermore, in spite of the limited sequence diversity included in the dataset analyzed here, we were able to detect signatures of positive selection in five S. sclerotiorum SPEP genes (2.5% of genes analyzed). Similar frequency (3.2%, 21 out of 642 genes) has been reported in B. cinerea [35]. In the future, an in depth exploration of sequence diversity in S. sclerotiorum should allow to reveal more sites subjected to selection. SPEP genes for which positive selection has been detected encode cell wall degrading enzymes, including SS1G_07749 encoding a putative xylanase. This protein is related to B. cinerea Xyn11 considered as a Pathogen Associated Molecular Pattern (PAMP) [9]. The detection of positive selection in SS1G_07749 is therefore consistent with the hypothesis that PAMPs may be characterized by signatures of positive selection in a background of strong negative selection [50]. It may therefore be hypothesized that a subset of S. sclerotiorum critical secreted enzymes are engaged in an evolutionary arms race with plant pattern recognition receptors, driving opposing forces of natural selection S. sclerotiorum effector genes. Since plant inhibitors are known for many fungal cell wall degrading enzymes, it is also possible that an evolutionary arms race with plant inhibitors drives the evolution of some S. sclerotiorum effector candidates [53,62,63]. Remarkably, we also identified 14.4% of species-specific SPEP genes. The extent to which evolutionary constraints imposed by a broad host range contributes to diversification in the effector candidate repertoire of S. sclerotiorum remains to be determined. Detailed functional analysis of effector gene alleles and their plant targets will be needed to address this question.
Although S. sclerotiorum is considered as a typical necrotroph, there is evidence that it colonizes plant tissues through multiple phases involving important transcriptional and physiological reprogramming [61]. Consistent with this model, the phytotoxin oxalic acid dampens plant immune responses at the initial stages of infection and later enhances programmed cell death [7]. In this work, we report the sequential transcriptional activation of S. sclerotiorum candidate effector genes. A >2 fold induction was measured at 6 hpi for several effector candidate genes, whereas no necrotic symptoms are visible at this time, except on A. thaliana Shahdara accession. This suggests that the sequential secretion of effectors is required for the efficient induction of host cell death by S. sclerotiorum, or that some secreted proteins could contribute to S. sclerotiorum virulence independently of host cell death activation. By comparing the expression pattern of selected Blumeria graminis f. sp. hordei genes grown on barley and on A. thaliana, Hacquard et al. [64] concluded that very divergent hosts do not significantly alter the fungal gene expression program. The expression pattern of some S. sclerotiorum ECs is indeed independent on the host plant being colonized (e.g. SS1G_06213). Nevertheless, other ECs showed differential regulation in a host-dependent manner (e.g. SS1G_08858, SS1G_13371). The transcriptional activation of distinct set of effectors depending on the host being colonized has also been reported for the generalist root endophyte Piriformospora indica in barley and A. thaliana [65]. The growing number of transcriptomic studies on various pathosystems should help determine whether host-dependent modulation of effector gene expression differs according to pathogens lifestyle or host range. We speculate that the white mold fungus benefits from a versatile repertoire of secreted proteins with diverse functions, evolution and expression patterns, to successfully infect a wide range of host plants. A systematic characterization of S. sclerotiorum transcriptome on multiple hosts and the functional analysis of differentially regulated effector genes should prove useful to decipher the molecular determinants of quantitative disease resistance and host range in this fungal pathogen.

8-angstrom resolution
C-score is a confidence score for estimating the quality of models predicted by I-TASSER ranging from −5 (low confidence) to 2 (high confidence). TM score is a quality score for the superimposition of 3D models calculated with TM-align with values in [0;1]. A TM-score >0.5 generally corresponds to the same fold in SCOP/CATH.

Conclusions
In this work, we explored systematically the diversity of candidate virulence genes in the necrotrophic fungal pathogen S. sclerotiorum using in silico structure and evolution analyses. We report the identification of 486 S. sclerotiorum secreted proteins expressed in planta, including 78 ECs. We have analyzed in planta expression for a representative subset of 16 ECs, highlighting diverse predicted functions and expression patterns. This study reveals that besides plant degrading enzymes, S. sclerotiorum genome encodes numerous predicted secreted proteins that may be involved in the interaction between the fungus and its host plants. It will facilitate future investigation on their relevance in the infection process and sheds new light on the underestimated complexity of host colonization by necrotrophic plant pathogens.

Secretome prediction and annotation
We used complete genome and predicted proteomes of Sclerotinia sclerotiorum strain 1980 v.2, Botrytis cinerea strain b05.10 v.1 and strain t4 v.1 described in [12]. The presence of secretion signals was predicted with SignalP v.2 and v.4 [66,67], transmembrane helices and GPI anchor sequence were predicted with TMHMM [68] and GPIsom [69] respectively. For the identification of genes expressed in planta, microarrays data for gene induction fold at 2 days post inoculation on sunflower cotyledons and Expressed Sequence Tags (ESTs) from [12] were used. ESTs were assigned to the S. sclerotiorum predicted transcript giving the lowest e-value in a BLASTN search. Genes were considered expressed in planta when either (i) showing induction fold ≥1 in during sunflower infection in microarrays data or (ii) being assigned at least one EST in either infection cushion, infected B. napus or infected tomato library. S. sclerotiorum predicted proteins were annotated using Blast2GO [70], PFAM [71] and NLStradamus [72]. Predicted proteins shorter than 40 amino-acids were excluded from the analysis. PFAM domains were annotated using HMMER3 searches against the PFAM 26.0 database [71]. We defined non-annotated predicted SPEPs as having no hit to PFAM_A with e-value <0.1. For the identification proteins similar to known fungal effectors, BlastP searches against a local database of 191 effectors with an e-value cutoff of 1e −3 .

Definition of ortholog clusters and natural selection analysis
Core ortholog groups (COGs) between S. sclerotiorum 1980 and B. cinerea b05.10 or B. cinerea t4 proteomes were identified using Inparanoid7 [73] with the following parameters: score cutoff 40 bits; sequence overlap cutoff 0.5; group merging cutoff 0.5; scoring matrix BLO-SUM62. COGs in which a length difference >10 aminoacids existed between S. sclerotiorum and B. cinerea were discarded. Pairwise ortholog alignments were generated using the needleall program from the EMBOSS package using the following parameters: gapopen 50.0; gapextend 0.2; minscore 100.0; aformat3 MARKX3. Needleall output files were parsed into .axt alignments used as input in Ka/Ks calculator2 [74]. Ka/Ks ratios were  calculated for all COG pairs using Yn00 method [28]. The identification of codon sites under positive selection was achieved through Bayesian inference using the Selecton2.2 server [75] with the "Positive selection enabled (M8, beta + w > =1)" evolutionary model with 8 categories, on alignments of S. sclerotiorum 1980, B. cinerea b05.10 and B. cinerea t4 orthologs.

Protein structure modeling and analysis
Protein structure modeling was performed with the I-TASSER server [76] and rendered using UCSF Chimera [77]. Site-specific alignment consensus and Ka/Ks ratios were mapped onto protein models using the 'define attribute' function in UCSF Chimera. Moving average over a 3 amino-acid window of the percentage consensus in a 99 homologous protein alignment was used to characterize conservation in SS1G_01593 family. Structural analogs were identified using the TM-align program in I-TASSER.

Taxonomic distribution and phylogenetic analyses
Fungal taxonomy trees are based on [16]. The presence of SPEP homologs in 234 fungal species was assessed using BlastP searches against JGI fungi Gene Catalog Proteins [78] with an e-value cutoff of 1e −5 without low complexity filter. Among retrieved homologs, proteins that had no signal peptide detected with SignalP4 or SignalP2 were discarded. For the global analysis of taxonomic distribution of SPEP genes, the predicted proteomes of Neurospora crassa, Magnaporthe oryzae, Verticilium dahlia, Fusarium oxysporum, Stagonospora nodorum, Pyrenophora tritici-repentis, Alternaria brassicicola, Leptosphaeria maculans, Mycosphaerella graminicola, Aspergillus flavus, Cryptococcus neoformans and Rhizopus oryzae were used in local BlastP searches with e-value cutoff 1. For each SPEP gene, BlastP scores for all hits in a given species were summed up, and SPEP genes were considered as absent if total score is <2. Phylogenetic trees for SS1G_13371 family was generated using the parsimony method with 100 bootstrap replicates with the Extended Majority rule, as implemented in the protpars and consense programs of the Phylip 3.67 package [79].
Sequence-based clustering and genome distribution of duplicated genes S. sclerotiorum 1980 and B. cinerea b05.10 proteins were clustered based on sequence similarity by Markov clustering using the orthoMCL function in Biolayout 3D [80]. A self BlastP search on the combined S. sclerotiorum 1980 and B. cinerea b05.10 complete proteomes with e-value cutoff 1e −30 was used as input for orthoMCL. Repeats and transposable elements were identified using RepeatMasker on S. sclerotiorum 1980 supercontigs with the cross_match method at slow speed and "Fungi" as a DNA source. Genomic distances and genome architecture heatmaps were generated according to [81].

Plant and fungus cultivation, inoculation procedure
Arabidopsis thaliana accession Shahdara and Rubezhnoe-1 were grown in Jiffy pots for four weeks at 22°C with cycles of 9 hours of light per 24 hours. Tomato (Solanum lycopersicum cv. Heinz) were grown for six weeks in pots containing disinfected soil in a greenhouse at 23°C with cycles of~14 hours of light per 24 hours. Nicotiana benthamiana plants were grown for four weeks at 21°C with cycles of 16 hours of light per 24 hours. S. sclerotiorum strain S55 was first grown for 4 days on PDA plates at 25°C in the dark. Fifty mL of liquid PDB medium were inoculated with 3 agar plugs of PDA cultures and incubated for 4 days at 25°C in the dark, with 150 rpm shaking. Three independent inoculation experiments were performed in which fully grown plant leaves were cut and placed right side up on a wet paper towel in large petri dishes. Mycelium was washed twice in PDB, filtered on Miracloth (Calbiochem, CA), and spread over whole leaf surfaces. Inoculated leaves were incubated for up to 3 days at 25°C with 14 hours of light per 24 hours.

Effector candidate gene expression by quantitative RT-PCR
Plant leaves were harvested immediately and 6, 24 and 48 hours after inoculation, and ground in liquid nitrogen. Total RNA was extracted using a Nucleospin RNA II kit (Machery-Nagel) according to manufacturer's instructions. RNAs were analyzed and quantified on an Agilent 2100 Bioanalyzer. The first-strand cDNA was synthesized using TRT reverse transcriptase (Roche) according to manufacturer's instructions. Real-time PCR reactions included 3.5 μL of SYBR green mix (Roche), 1 μL of 5 μM primers (Additional file 3: