Time-resolved dual root-microbe transcriptomics reveals early induced Nicotiana benthamiana genes and conserved infection-promoting Phytophthora palmivora effectors

Background Plant-pathogenic oomycetes are responsible for economically important losses on crops worldwide. Phytophthora palmivora, a broad-host-range tropical relative of the potato late blight pathogen, causes rotting diseases in many important tropical crops including papaya, cocoa, oil palm, black pepper, rubber, coconut, durian, mango, cassava and citrus. Transcriptomics have helped to identify repertoires of host-translocated microbial effector proteins which counteract defenses and reprogram the host in support of infection. As such, these studies have helped understanding of how pathogens cause diseases. Despite the importance of P. palmivora diseases, genetic resources to allow for disease resistance breeding and identification of microbial effectors are scarce. Results We employed the model plant N. benthamiana to study the P. palmivora root infections at the cellular and molecular level. Time-resolved dual transcriptomics revealed different pathogen and host transcriptome dynamics. De novo assembly of P. palmivora transcriptome and semi-automated prediction and annotation of the secretome enabled robust identification of conserved infection-promoting effectors. We show that one of them, REX3, suppresses plant secretion processes. In a survey for early transcriptionally activated plant genes we identified a N. benthamiana gene specifically induced at infected root tips that encodes a peptide with danger-associated molecular features. Conclusions These results constitute a major advance in our understanding of P. palmivora diseases and establish extensive resources for P. palmivora pathogenomics, effector-aided resistance breeding and the generation of induced resistance to Phytophthora root infections. Furthermore, our approach to find infection relevant secreted genes is transferable to other pathogen-host interactions and not restricted to plants.


Background
Phytophthora is a genus of plant-pathogenic oomycetes responsible for economically important losses on crops worldwide, as well as damage to natural ecosystems [1]. Phytophthora infestans is the causal agent of tomato and potato late blight in temperate climates and contributed to major crop losses during the Great Irish Famine [2]. Phytophthora palmivora, a broad-host-range tropical relative of P. infestans originating from south-eastern Asia [3] but now present worldwide due to international trade [4] causes root, bud and fruit rotting diseases in many important tropical crops such as papaya, cocoa, oil palm, black pepper, rubber, coconut, durian, mango, cassava and citrus [5][6][7][8]. In addition, P. palmivora infects roots and leaves of several model plant species such as Medicago truncatula [9], Hordeum vulgare [10] and Arabidopsis thaliana [11]. Despite its economic impact and widespread distribution, nothing is known about the molecular basis underlying its ability to infect many unrelated host species and the root responses associated with an infection.
P. palmivora has a hemibiotrophic lifestyle. Similar to other Phytophthora species, its asexual life cycle in plants is characterised by adhesion of mobile zoospores to the host tissue, encystment and germ tube formation [12]. Entry into the plant is achieved via surface appressoria and is followed by establishment of an apoplastic hyphal network. During this biotrophic stage P. palmivora projects haustoria into plant cells. These contribute to acquisition of nutrients and release virulence proteins known as effectors [13]. This is followed by a necrotrophic stage characterised by host tissue necrosis and the production of numerous sporangia which release zoospores [14].
Sequencing of Phytophthora genomes and transcriptomes has revealed repertoires of effector proteins that counteract plant defenses and reprogram the host in support of infection. Secretome predictions and subsequent evolutionary and functional studies have helped to understand how these pathogens cause diseases [15,16]. Oomycete effectors are secreted into the apoplast of infected plants. Some of them act inside plant cells and conserved RXLR or LFLAK amino acid motifs in their N-terminal parts have been associated with their translocation from the microbe into the host cell [17,18]. The LFLAK motif is present in Crinkler (CRN) effectors, named after a crinkling and necrosis phenotype caused by some CRN proteins when expressed in plants [19]. RXLR effectors are usually short proteins with little similarity to conserved functional domains in their C-termini.
They localise to diverse subcellular compartments and associate with plant target proteins with key roles during infection [20].
Recent studies on bacterial and oomycete plant pathogens identified subsets of effectors that are conserved among a large number of strains. These so-called core effectors are responsible for a substantial contribution to virulence and thus cannot be mutated or lost by the pathogen without a significant decrease in virulence [21]. Thus, core effectors constitute highly valuable targets for identification of resistant germplasm and subsequent breeding disease-resistant crops [21][22][23].To date, the occurrence of such core effectors in oomycetes has largely been reported from plant pathogens with narrow economical host range such as Hyaloperonospora arabidopsidis, Phytophthora sojae [24] and P. infestans [25].
Plants have evolved a cell autonomous surveillance system to defend themselves against invading microbes [26]. Surface exposed pattern recognition receptors (PRRs) recognize conserved microbeassociated molecular patterns (MAMPs) released during infection, such as the Phytophthora transglutaminase peptide pep-13 [27,28]. In addition, plants are also able to recognize self-derived so-called damage-associated molecular patterns (DAMPs). These include intracellular peptides that get released in the apoplast upon wounding, such as systemins [29] and secreted plant peptides precursors with DAMP features that get processed in the apoplast [30][31][32]. Pathogen recognition initiates basal defense responses which include activation of structural and biochemical barriers, the MAMP-triggered immunity (MTI) [26]. Plant pathogens are able to overcome MTI by secreting effectors that suppress or compromise MTI responses, thereby facilitating effector-triggered susceptibility (ETS). In response, plants have evolved disease resistance proteins to detect pathogen effectors or effector-mediated modification of host processes, leading to effector-triggered immunity (ETI) [26]. Phytophthora genes encoding effectors which trigger a resistance response in host plants carrying the cognate disease resistance gene are often termed avirulence (AVR) genes.
Cross-species transfer of PRRs and disease resistance genes against conserved MAMPs or AVR proteins has been successfully employed to engineer resistant crops [33,34].
Host cell responses to oomycete infections have mainly been studied in aboveground tissues and notably involve subcellular rearrangements of the infected cells, including remodelling of the cytoskeleton [14, 35,36] and focal accumulation of secretory vesicles [37,38], which contribute to defense by delivering antimicrobial compounds to the extrahaustorial matrix [39,40]. Endocytic vesicles accumulate around oomycete haustoria [41] and the plant-specific small GTPase RAB5 is recruited at the extrahaustorial membrane during Arabidopsis infection by obligate biotrophs [42].
In this study, we employ the model plant N. benthamiana [50] to study root infection by P. palmivora. Dual transcriptomics and de novo assembly of the P. palmivora transcriptome allowed us to define pathogen and plant genes expressed during the interaction. We identified major shifts in pathogen gene expression dynamics associated with lifestyle changes which, interestingly, are not mirrored by dramatic shifts in plant gene expression patterns. We characterised two conserved RXLR effectors, REX2 and REX3 that promote root infection upon expression in plants. We furthermore show that REX3 was able to interfere with host secretion. By studying host transcriptional changes upon infection we identified a gene encoding a secreted peptide precursor with potential DAMP motifs whose promoter was specifically activated at root tip infection sites.
Hence, our work establishes a major resource for root-pathogen interactions, showcases examples of how to exploit these data, and provides inroads for effector-aided resistance breeding in tropical crops.

Phytophthora palmivora exerts a hemibiotrophic lifestyle in Nicotiana benthamiana roots
To describe the infection development of the root pathogen P. palmivora we investigated the infection dynamics of hydroponically grown N. benthamiana plants root-inoculated with P. palmivora LILI-YKDEL [9] zoospores ( Figure 1). Disease development was followed on the aerial parts (Figure 1a) since infected roots did not display visible disease symptoms. The plants looked healthy for up to 3 days (Symptom Extent Stage 1, SES 1). Disease progression in the aerial parts then resulted in a shrunken, brown hypocotyl and wilting of the oldest leaves (SES 2). This was rapidly followed by brown coloration and tissue shrinkage of the stem (SES 3) up to the apex (SES 4). Infected plants eventually died within 8 to 10 days (SES 5), indicating that N. benthamiana is susceptible to root infection by P. palmivora (Figure 1a) . We next characterised the P. palmivora -N. benthamiana interaction on the microscopic level using the fluorescently labeled isolate LILI-YKDEL (Figure 1b-h). Infection events were observed at 3 hours after inoculation (hai). Zoospores were primarily attracted to root tips, where they encysted and germinated. Appressoria were differentiated at this stage and, when infection of the first cell had already occurred, an infection vesicle and subjacent nascent hyphae were also observed ( Figure   1b). Haustoria were visible from 6 hai -24 hai, indicative of biotrophic growth (Figure 1c-e). At 18 hai, P. palmivora hyphae grew parallel to the cell files in the root cortex, forming a clear colonisation front between infected and non-infected tissues. In addition, extraradical hyphal growth was observed near the root tip (Figure 1d). First sporangia occurred at 30 hai (Figure 1f).
Consistent with the symptoms observed on aerial parts, hypocotyl colonization occurred between 30 hai and 48 hai (Figure 1g). Finally, the presence of empty or germinating sporangia at 72 hai suggests possible secondary infections (Figure 1h). Therefore, P. palmivora asexual life cycle completes within 72 hai in N. benthamiana roots.
We supported our microscopic studies with biomass quantification based on transcript levels of the P. palmivora 40S ribosomal protein S3A (WS21) (Figure 1i). We further characterized the different stages observed microscopically by quantifying expression of the P. infestans orthologs of Hmp1 (haustorium-specific membrane protein) [51] (Figure 1j) and the cell-cycle regulator Cdc14 [52] ( Figure 1k). Hmp1 transcripts peaked between 3 hai and 6 hai and then decreased at later stages.
By contrast, Cdc14 transcripts increased at late time points (48 hai and 72 hai). Taken together, these results further support the conclusion that P. palmivora exerts a hemibiotrophic lifestyle in N. benthamiana roots.

De novo assembly of P. palmivora transcriptome from mixed samples
We performed dual sequencing permitting de novo assembly of a P. palmivora transcriptome as well as an assessment of transcriptional changes in both, host and pathogen over time. We extracted RNA from infected and uninfected N. benthamiana roots at six time points matching the key steps of the interaction identified by microscopy: 6 hai, 18 hai, 24 hai, 30 hai, 48 hai and 72 hai and an axenically grown P. palmivora sample containing mycelia and zoospores (MZ). Using Illumina HiSeq 2500 paired-end sequencing we obtained a relatively uniform read depth of 50-60 M reads per sample (Table S1). To cover all possible transcripts we reconstructed the P. palmivora transcriptome de novo, combining ex planta and in planta root samples as well as 76 nt Illumina paired-end reads from infected N. benthamiana leaf samples (more than 515 M reads, Figure 2a, Table S1).
Following standard adaptor trimming and read quality control, we applied a two-step filtering procedure (Figure 2a) to separate pathogen reads from plant host reads. First we mapped the pooled read dataset to the N. benthamiana reference genome and collected unmapped read pairs. Recovered reads were subsequently mapped to the N. benthamiana transcriptome [53]. Reads not mapped to either host plant genome or transcriptome were used to run assemblies. Short reads (<60 nt) were filtered out to produce transcripts of better quality and coherence. Final de novo Trinity assemblies were run from 190 M pre-processed, properly paired and cleaned reads (Table S1). This yielded 57'579 'trinity genes' corresponding to 100'303 transcripts with an average backwards alignment rate of 76%, indicative of an overall acceptable representation of reads and therefore reasonably good assembly quality [54]. 9'491 trinity genes (20'045 transcripts including all isoforms) were removed by additional checks for residual plant contamination, resulting in a final P. palmivora transcriptome of 48'089 trinity genes corresponding to 80'258 transcripts ( Table 1).
We further selected 13'997 trinity genes (corresponding to 27106 transcripts) having the best expression support (Supplementary Dataset 1).
Interestingly, the remaining 35 BUSCO genes were consistently missing from all analysed Phytophthora genomes and transcriptomes (Supplementary Table 2). These results suggest that our P. palmivora (LILI) transcriptome assembly actually contained 87% of BUSCO genes occurring in Phytophthora Hence, our assembly shows acceptable quality and integrity and can be used as a reference for further studies.

P. palmivora secretome prediction and annotation identifies a set of effector candidate genes
Pathogen-secreted effectors and hydrolytic enzymes are hallmarks of Phytophthora infection [56].
Therefore, we probed our P. palmivora transcriptome for transcripts encoding secreted proteins. A TransDecoder-based search for candidate open reading frames (ORFs) [57] identified 123'528 ORFs from predicted trinity genes (isoforms included). We then analyzed the predicted ORFs using an automated pipeline for secretome prediction (Figure 2b) building on existing tools [58][59][60]. The pipeline was designed to predict signal peptides and cellular localisation with thresholds specific for oomycete sequences [61,62] and to exclude proteins with internal transmembrane domains and/or an endoplasmic reticulum (ER) retention signal. We identified 4'163 ORFs encoding putative secreted proteins.
Partial translated ORFs which were not predicted as secreted were subjected to an additional analysis (M-slicer) (Figure 2b) and resubmitted to the secretome prediction pipeline. This improved procedure allowed us to rescue 611 additional ORFs encoding putative secreted proteins. In total, we identified 4'774 ORFs encoding putative secreted P. palmivora proteins. We further selected a single representative secreted ORF for genes with sufficient expression support (TPM ≥ 1 in 3 or more samples). This yielded 2'028 P. palmivora genes encoding putative secreted proteins

(Supplementary Dataset 5).
To maximise functional annotation of the P. palmivora secretome we used an integrative approach (Figure 2c) tailored to the use of known short motifs characteristic of oomycete secreted proteins.
The pipeline contains three major blocks. The first block integrated all the sequence information, assignment to 2'028 non-redundant genes encoding secreted proteins as well as expression data.
The second block combines results of homology searches, for both full-length alignments (blastn and blastx) and individual functional domains (InterProScan). The third block was designed to survey for known oomycete motifs and domains (such as RXLR, EER, WY for RXLR-effectors; LXLFLAK for crinklers and NLS for effectors in general). The pipeline produced an initial secretome annotation (Figure 2c) which was then manually curated to avoid conflicting annotations. This strategy allowed us to to assign a functional category to 768 (38 %) of predicted secreted proteins (  Figure   2). Notably, P. palmivora AVR3a (PLTG_13552) harbours the K80/I103 configuration, but combined with a terminal valine instead of a tyrosine in PiAVR3a [63]. It thus remains to be tested whether PLTG_13552 is capable of triggering a R3a-mediated hypersensitive response.
Our pipeline only identified 3 genes encoding putative CRN effectors (PLTG_06681, PLTG_02753, PLTG_03744). Crinklers often lack predictable signal peptides, but instead might be translocated into plant cells by an alternative mechanism [64]. An independent survey using HMM-prediction without prior signal peptide prediction revealed a total of 15 CRN motif-containing proteins.
The P. palmivora secretome also contained a substantial number of apoplastic effectors ( Table 2).
We identified 28 genes encoding extracellular protease inhibitors, including extracellular serine protease inhibitors (EPI) with up to five recognisable Kazal domains, several cystatins and cysteine protease inhibitors (EPICs) (Supplementary Dataset 5). PLTG_05646 encodes a cathepsin protease inhibitor domain followed by a cysteine protease and an ML domain (PF02221, MD-2related lipid recognition domain). We also identified 28 proteins with small cysteine rich (SCR) signatures, 18 of them being encoded in full-length ORFs, but only six where the mature peptide is shorter than 100 aa. Longer SCRs can harbour tandem arrangements (PLTG_08623). In one case an SCR is linked to a N-terminal PAN/APPLE domain, which is common for carbohydrate-binding proteins [66].
Taken together, de novo transcriptome assembly followed by multistep prediction of ORF encoding potentially secreted proteins and a semi-automated annotation procedure allowed us to identify all major classes of effectors characteristic to oomycetes as well as P. palmivora-specific effectors with previously unreported domain arrangements. Our data suggest that P. palmivora's infection strategy relies on a diverse set of extracellular proteins many of which do not match to previously characterised effectors.  In order to highlight dynamic expression changes of P. palmivora genes during infection, we performed fuzzy clustering of P. palmivora DEGs (Figure 4) to lower sensitivity to noisy expression signals and to distinguish between expression profiles, even if they partially overlapped [68]. We identified 12 expression clusters falling into four main groups according to their temporal expression level maximum (Figure 4a).

REX3 impairs plant secretion processes
Suppression of defense component secretion has previously been found to be targeted by at least two mechanisms [48,49]. We thus investigated the ability of the infection promoting REX2 and

The TIPTOP promoter is activated at root tip infection sites
When screening our data for plant promoters responding early to P. palmivora attack we found Niben101Scf03747g00005, encoding a small secreted protein containing two repeats of a conserved SGPS-GxGH motif known from pathogen-associated molecular pattern (PAMP)-induced peptides  infection (Figure 8c). P. palmivora-triggered TIPTOP promoter activation was strongest adjacent to invasive hyphae as revealed by GFP confocal fluorescence microscopy ( Figure 8d). In addition, the TIPTOP promoter was not activated by abiotic stresses (cold, heat and 1 M sodium chloride) and wounding, but weak activation was observed in root tips in response to flagellin (flg22) treatment ( Figure S14). PlantPAN 2 [71] analysis of the TIPTOP promoter sequence identified various transcription factor binding motifs (Table S5). Taken together, these results suggest that TIPTOP is a root tip specific P. palmivora-induced promoter.

Dual transcriptomics and de novo assembly enables functional studies of unsequenced genomes
Dual transcriptomics captures simultaneous changes in host and pathogen transcriptomes [72,73] when physical separation of two interacting organisms is unfeasible. The diversity of plant pathogens often results in the absence of microbial reference genomes. This is particularly relevant for obligate biotrophic plant pathogens which cannot be cultivated separately from their host. Our established viable alternative, a de novo assembly of a plant pathogen transcriptome from separated mixed reads followed by an semi-automated annotation is thus applicable to a broader community.
Taking advantage of the availability of the host reference genome, we separated P. palmivora reads from the mixed samples and combined them with reads from the ex planta samples to create a single de novo assembly for the pathogen transcriptome. Assembly completeness in terms of gene content might be assessed based on evolutionary expectations, so that recovery of conserved genes serves as a proxy measure for the overall completeness (CEGMA [74] and BUSCO [55]). Our P. palmivora de novo assembly had sufficient read support (on average 76% reads mapping back), so we further probed it for the presence of BUSCOs. Since there is no specific oomycete set, we checked presence of 429 eukaryotic BUSCO genes and found 326 of them (76%). Lack of some BUSCO genes in our assembly might result from the fact that originally BUSCO sets were developed to estimate completeness of genomic assemblies and did not require expression evidence [55]. To verify this, we extended the same completeness analysis to existing Phytophthora genomes and transcriptomes and found that transcriptomes in general indeed contained fewer BUSCOs. Moreover, we found 35 eukaryotic BUSCO genes consistently missing from Phytophthora genomic assemblies. Therefore, a BUSCObased completeness test for transcriptomes should be applied with caution within the Phytophthora genus, considering adjustments for expression support and uneven distribution of eukaryotic singlecopy orthologs. We propose that with an ever-growing body of oomycete genomic and transcriptomic data a specific set of benchmarking orthologs needs to be created to support de novo assemblies and facilitate studies of these economically relevant non-model plant pathogens [75].
So far, dual transcriptomics has only been used with limited time resolution and sequencing depth in plant-pathogenic oomycete studies [76,77]. Our study encompasses the full range of P. palmivora sequential lifestyle transitions occurring in N. benthamiana root, allowing reconstruction of comprehensive transcriptional landscape in both interacting organisms. We found three major waves of P. palmivora gene expression peaks that correlate with its major lifestyle transitions: 1) early infection and biotrophic growth inside host tissues; 2) switch to necrotrophy; 3) late necrotrophy and sporulation. Similar transcription dynamics following switches of life styles were previously described for the hemibiotrophic pathogens Colletotrichum higginsianum [78], Phytophthora parasitica [79] during Arabidopsis root infection and P. sojae upon infection of soybean leaves [80], though the exact timing of infection was different.
Interestingly, the N. benthamiana transcriptional response to infection does not mirror the observed significant shifts in infection stage specific P. palmivora gene expression. Instead it is characterized by steady induction and repression. High-resolution transcriptomics were applied to A. thaliana leaves challenged with Botrytis cinerea to untangle the successive steps of host response to infection [81]. However, in the absence of pathogen expression data, it is not possible to correlate these changes with changes in the pathogen transcriptome. It is likely that pathogen expression patterns are not useable to infer a link to corresponding plant responses.
The response of N. benthamiana roots to P. palmivora is characterised by an upregulation of genes associated with hormone physiology, notably ethylene through activation of ethylene response transcription factors (ERFs) and ACC synthase. Ethylene is involved in N. benthamiana resistance to P. infestans [82]. We also observed an induction of two PIN-like auxin efflux carriers.
Suppression of the auxin response was associated with increased A. thaliana susceptibility to P. cinnamomi disease and was stimulated by phosphite-mediated resistance [83]. Interestingly, phosphite was also required for defense against P. palmivora [11]. We found upregulation of chitinases and endopeptidase inhibitors, such as Kunitz-type trypsin inhibitors, which are often induced by oomycete and fungal pathogens [84][85][86]. Induction of genes encoding O-glycosyl hydrolases is associated with cell wall remodelling while phenylalanine ammonia lyases (PAL) contribute to cell wall reinforcement by activation of lignin biosynthesis [87,88]. Upregulation of the trehalose biosynthesis pathway is associated with membrane stabilisation [89] and partially mitigates toxic effects of oxidative stress [90]. Upregulation of several enzymes of the mevalonate pathway suggest modulation of the biosynthesis of isoprenoids such as defense-associated phytoalexins as well as sterols. In particular, transcriptional repression of genes encoding sterol 4-alpha-methyl-oxidase 2-1 and C5 sterol desaturases suggest an attenuation of the brassinosteroid synthesis, while repression of genes with homology to sterol methyl transferase 2 point to a repression of the beta-sitosterol/stigmasterol branch. Conversely, induction of terpenoid synthases/epi aristolochene synthases points to a selective induction of the sesquiterpenes which contain defense associated phytoalexins such as capsidiol [91,92]. Finally, the N. benthamiana response to P. palmivora also includes upregulation of genes encoding late embryogenesis abundant (LEA) proteins as well as heat shock proteins. LEA proteins have been associated with the drought response [93,94] and upregulation of genes associated with water deprivation upon Phytophthora infection has been previously reported [77]. Conversely, downregulated genes were mostly associated with photosynthesis, cellulose biosynthesis and cell division. These results were consistent with previous reports [95,96].

Analysing partial transcripts improved the predicted P. palmivora secretome
To study P. palmivora secreted proteins we developed a prediction and annotation pipeline tailored for signal peptide prediction based on ORFs derived from de novo assembled transcripts. Often a six-frame translation is utilised to identify candidate ORFs (Stothard, 2000; Lévesque et al 2010).
However, we use a TransDecoder approach which enriches for the most likely translated ORF by integrating homology based matches to known Pfam domains and Phytophthora proteins.
Compared to six-frame translation, this approach can result in partial ORFs which may lead to a mis-prediction of translation start sites and therefore signal peptides. So, we implemented a refinement step in our secretome prediction pipeline to rescue partial ORFs by finding the next likely translation start position and following the secretome prediction steps. This procedure allowed us to rescue an additional 611 ORFs including several which likely encode RXLR effectors, elicitins and cell wall degrading enzymes thus highlighting the importance of this additional step.

Effector-guided resistance breeding potential
We identified two RXLR effectors that show high sequence conservation among P. palmivora isolates worldwide, suggesting they may represent core effectors that cannot be lost or mutated without a fitness cost for the pathogen [21]. As such, these effectors constitute valuable candidates to accelerate cloning of disease resistance (R) genes and effector-assisted deployment of resistance.
This strategy has been used against P. infestans [22].
Our approach identified a potential AVR3a homolog in P. palmivora (PLTG_13552). The P. infestans AVR3a KI allele confers avirulence to P. infestans isolates on R3a-expressing potatoes while the AVR3a EM allele is not being recognised [63]. It will be interesting to study whether potato R3a or engineered R3a derivatives with a broader recognition spectrum [97,98] can be exploited to generate resistance towards P. palmivora in economically relevant transformable host plants.
Additionally, P. palmivora proteins also harbour pep13-type MAMP motifs present in four transglutaminases and several nlp20-containing NLPs. While the pep13 plant receptor remains to be found, the receptor like kinase RLP23 has recently been identified as nlp20 receptor [99] with the potential to confer resistance even when transferred into other plant species. Introduction of RLP23 into P. palmivora host plants may thus be another strategy to engineer resistant crops.

The P. palmivora effector REX3 inhibits plant secretion pathways
We found that REX3 interferes with host secretion, a common strategy of bacterial and oomycete pathogens [49,69]. Rerouting of the host late endocytic trafficking to the extrahaustorial membrane [41,100] and accumulation of the small GTPase RAB5 around haustoria [42] is well documented.
Given that REX3 is almost invariant in P. palmivora it is likely that REX3 targets components of the secretory pathway which are conserved among diverse host species. Of the four functionally tested RXLR effectors the two most conserved ones (REX2, REX3) amongst P. palmivora isolates both conferred increased susceptibility. REX2 and REX3 therefore represent important targets for disease resistance breeding in tropical crops. It is possible that isolate-specific variants of REX1 and REX4 may provide a colonisation benefit only in hosts other than N. benthamiana.

P. palmivora triggers expression of danger-associated molecular pattern peptides
Upon P. palmivora root infection 2'886 N. benthamiana genes were up and 3'704 genes downregulated. Compared to previously studied root transcriptomes of responses to broad-host-range Phytophthora species [95,101] our data permitted the identification of early induced genes such as TIPTOP, a P. palmivora-responsive root tip promoter. An exciting future perspective is its exploitation for induced early resistance against Phytophthora root infections. This promoter also provides inroads to dissect early host cell responses to P. palmivora, when employed in combination with a cell sorting approach to generate samples enriched for infected cells.
The TIPTOP gene encodes a peptide with similarities to DAMP peptides [102]. The occurrence of two tandem repeats of a conserved sGPSPGxGH motif in the TIPTOP protein is reminiscent of the SGPS/GxGH motifs of PIP and PIPLs peptides [32,103] and the closest Arabidopsis homologs of TIPTOP, PIP2 and PIP3, are implied in responses to biotic stress.

Hou and coworkers showed that the PIP1 peptide is induced by pathogen elicitors and amplifies
A. thaliana immune response by binding to the receptor-like kinase 7 (RLK7) [32]. Analysis of in silico data showed that PIP2 and PIP3 were activated upon A. thaliana infection by Botrytis cinerea or P. infestans [103]. By contrast to PIP1 and PIP2, the TIPTOP promoter is inactive under control conditions, suggesting it may undergo a different transcriptional regulation than the previously characterized Arabidopsis peptides.

Conclusions
Dual transcriptomics represent a successful approach to identify transcriptionally regulated effectors as well as plant genes implicated in the root infection process. We found conserved MAMPs and effectors with similarity to known AVR proteins such as AVR3a which may harbour the potential for disease resistance engineering. We characterised two conserved RXLR effectors conferring enhanced susceptibility to root infection and confirmed interference with host secretion as a P.
palmivora pathogenicity mechanism. Furthermore, the P. palmivora inducible TIPTOP promoter and the PIP2,3-like peptide are promising leads for engineering P. palmivora resistance. In summary, our findings provide a rich resource for researchers studying oomycete plant interactions.

P. palmivora Butler isolate LILI (reference P16830) was initially isolated from oil palm in
Colombia [70], and maintained in the P. palmivora collection at the Sainsbury Laboratory (Cambridge, UK). Transgenic P. palmivora LILI strain expressing KDEL-YFP [9] and tdTomato [70] have been previously described. Phytophthora growth conditions and the production of zoospores have been described elsewhere [10].   [108]. Three biological replicates of the entire experiment were performed. Gene expression was normalized with respect to constitutively expressed internal controls, quantified and plotted using R software.

Plasmid construction
The vector pTrafficLights was derived from pK7WGF2 (Plant System Biology, Gent University, Belgium). A cassette containing the signal peptide sequence of Nicotiana tabacum pathogenesisrelated protein 1 (PR-1; GenBank accession X06930.1) fused in frame with the green fluorescent protein (GFP) was obtained by PCR using primers SP-F/SP-R (Table S3) and ligated into pK7WGF2 using SpeI and EcoRI restriction enzymes. The AtUBQ10pro::DsRed cassette was amplified from pK7WGIGW2(II)-RedRoot (Wageningen University, Netherlands) using primers RedRoot-F/RedRoot-R (Table S3) and ligated into pK7WGF2 using XbaI and BamHI restriction enzymes.

Confocal microscopy
Confocal laser scanning microscopy images were obtained with a Leica SP8 laser-scanning confocal microscope equipped with a 63x 1.2 numerical aperture (NA) objective (Leica, Germany).
A white-light laser was used for excitation at 488 nm for GFP visualization, at 514 nm for YFP visualization and at 543 nm for the visualization of tdTomato. Pictures were analysed with ImageJ software (http://imagej.nih.gov/ij/) and plugin BioFormats.

Library preparation and sequencing
N. benthamiana and P. palmivora mRNAs were purified using Poly(A) selection from total RNA sample, and then fragmented. The raw fastq data are accessible at http://www.ncbi.nlm.nih.gov/sra/ with accession number SRP096022.

De novo transcriptome assembly
In order to capture the full complexity of the P. palmivora transcriptome we pooled all the samples potentially containing reads from P. palmivora (Figure 2): eight mixed (plant-pathogen, combining leaf and root infections), one exclusively mycelium and one mixed mycelium-zoospores sample.
Initial read quality assessment was done with FastQC (Babraham Bioinformatics, Cambridge, UK).
Unmapped reads (with both mates unmapped) were collected with samtools (samtools view -b -f 12 -F 256), converted to fastq with bedtools and processed further. To estimate the level of residual contamination by plant and potentially bacterial reads, the resulting set of reads was subjected to FastQ Screen against the UniVec database, all bacterial and archaeal sequences obtained from RefSeq database, all viral sequences obtained from RefSeq database, N. benthamiana genome (v1.01), and subset 16 oomycete species (mostly Phytophthora species). Since the above test revealed substantial residual contamination by N. benthamiana reads, an additional round of bowtie2 alignment directly to N. benthamiana transcriptome [53] was performed followed by FastQ Screen. Reads, not aligned to N. benthamiana genome and transcriptome were further subjected to quality control using Trimmomatic (minimum read length = 60). The quality parameters for the library were assessed using FastQC. The total of ~190 M filtered reads were subjected to de novo assembly with Trinity (trinity v2.1.1) on a high-RAM server with minimal k-mer coverage = 2 and k-mer length = 25. In silico read normalization was used due to the large number of input reads, in order to improve assembly efficiency and to reduce run times [57]. The resulting assembly was additionally checked for plant contamination using blastn search against plant division of NCBI RefSeq genomic database. Trinity genes having significant sequence similarity (e-value threshold ≤10-5) to plant sequences were removed from the resulting transcriptome. The final version of assembly included trinity genes with sufficient read support.

De novo assembly statistics and integrity assessment
General statistics of the assembly were determined using the 'TrinityStats.pl' script provided with Trinity release and independently using Transrate (http://hibberdlab.com/transrate/) and Detonate (http://deweylab.biostat.wisc.edu/detonate/) tools. Assembly completeness was estimated using the eukaryotic set of BUSCO profiles (v1) [55]. BUSCO analysis was performed for the full transcriptome assembly and for the reduced assembly, obtained after retaining only the longest isoform per trinity gene. BUSCO genes missing from the assembly were annotated with InterProsScan based on the amino acid sequences emitted from the corresponding hmm profile ('hmmemit' function from hmmer package, http://hmmer.org/). Overall expression support per assembled transcript was performed after transcript abundance estimation. Trinity genes with TPM ≥ 1 in at least 3 samples were considered further.

Protein prediction and annotation
ORFs were predicted using TransDecoder software [57]. At the first step ORFs longer than 100 aa were extracted. The top 500 longest ORFs were used for training a Markov model for coding sequences, candidate coding regions were identified based on log-likelihood score. Additionally all the ORFs having homology to protein domains from the Pfam database and/or P. sojae, P. parasitica, P. infestans and P. ramorum protein sequences downloaded from Uniprot database (accession numbers: UP000005238, UP000006643, UP000002640, UP000018817) were also retained (blastp parameters: max_target_seqs 1 -evalue 1e-5).

Secretome prediction
For the automatic secretome prediction a custom script was written, employing steps taken for P. infestans secretome identification [16]. Predicted proteins were subsequently submitted to SignalP 2.0 (Prediction = 'Signal peptide'), SignalP 3.0 (Prediction = 'Signal peptide', Y max score ≥ 0.5, D score ≥ 0.5, S probability ≥ 0.9), TargetP (Location = 'Secreted') [112] and TMHMM (ORFs with transmembrane domains after predicted signal peptide cleavage site were removed) [113]. Finally, all proteins with terminal 'KDEL' or 'HDEL' motifs were also removed, as these motifs are known to be ER-retention signals [114]. Exact duplicated sequences and substrings of longer ORFs were removed to construct non-redundant set of putative secreted proteins. Taking into account possible fragmentation of de novo assembled transcripts a custom python script (M-slicer) was developed to rescue partial proteins with mis-predicted CDS coordinates. It takes as an input all the partial translated ORFs, which were not predicted to be secreted initially and creates a sliced sequence by finding the position of the next methionine. The M-sliced proteins were subjected to the same filtering step as was done with the initial secretome. The same script, omitting the M-slicer refinment, was used to systematically predict N. benthamiana genes encoding putative secreted proteins.

Secretome annotation
To annotate putative secreted proteins a complex approach was used, combing several lines of evidence: 1) blastp search against GenBank NR database with e-value ≤10-6; 2) InterProScan   [118]. For P. palmivora quantification was performed on 'trinity gene' level. For within-sample normalisation TPMs were calculated. Between-sample normalisation was done using trimmed means approach (TMM) [119]. TMM-normalised TPMs were reported for both P. palmivora and N.
benthamiana. PCA-analysis was performed on the log-transformed TPM values and visualized in R with the help of "ggplot2" [120] and "pheatmap" [121] packages. Overlap between groups of genes identified in the PCA analysis was visualised with "Vennerable" package [122]. Differentially expressed genes were identified with edgeR Bioconductor package [119] following pair-wise comparisons between all the samples. The dispersion parameter was estimated from the data with the estimateDisp function on reduced datasets: for P. palmivora we combined close time points (based on PCA analysis) and treated them as pseudo-replicates; for N benthamiana common dispersion was estimated based on 6 uninfected plant samples, treating them as replicates. The resulting common dispersion values of 0.15 and 0.1 were used for P. palmivora and N. benthamiana analysis, respectively. Most differentially expressed genes (log2(fold change) ≥ 2 and p-value ≤ 10-3 were used to perform hierarchical clustering of samples. Heatmaps for the most differentially expressed genes were generated using R "cluster" [123], "Biobase" [124] and "qvalue" packages.      Statistical significance has been assessed using one-way ANOVA and Tukey's HSD test (P < 0.05).    was detected in the cytoplasm but was excluded from the nucleus. Scale bar is 10 µm.   Figure S1 -BUSCO genes missing from available Phytophthora genomes and transcriptomes.            Table S1 -Sequencing and mapping statistics in RNA-seq samples containing P. palmivora Table S2 -BUSCO genes missing from genomes and transcriptomes of Phytophthora genus.

Supporting information
BUSCO genes were Annotated using InterProScan based on sequences emitted from HMM profiles.