Evidence for Small RNAs Homologous to Effector-Encoding Genes and Transposable Elements in the Oomycete Phytophthora infestans

Phytophthora infestans is the oomycete pathogen responsible for the devastating late blight disease on potato and tomato. There is presently an intense research focus on the role(s) of effectors in promoting late blight disease development. However, little is known about how they are regulated, or how diversity in their expression may be generated among different isolates. Here we present data from investigation of RNA silencing processes, characterized by non-coding small RNA molecules (sRNA) of 19–40 nt. From deep sequencing of sRNAs we have identified sRNAs matching numerous RxLR and Crinkler (CRN) effector protein genes in two isolates differing in pathogenicity. Effector gene-derived sRNAs were present in both isolates, but exhibited marked differences in abundance, especially for CRN effectors. Small RNAs in P. infestans grouped into three clear size classes of 21, 25/26 and 32 nt. Small RNAs from all size classes mapped to RxLR effector genes, but notably 21 nt sRNAs were the predominant size class mapping to CRN effector genes. Some effector genes, such as PiAvr3a, to which sRNAs were found, also exhibited differences in transcript accumulation between the two isolates. The P. infestans genome is rich in transposable elements, and the majority of sRNAs of all size classes mapped to these sequences, predominantly to long terminal repeat (LTR) retrotransposons. RNA silencing of Dicer and Argonaute genes provided evidence that generation of 21 nt sRNAs is Dicer-dependent, while accumulation of longer sRNAs was impacted by silencing of Argonaute genes. Additionally, we identified six microRNA (miRNA) candidates from our sequencing data, their precursor sequences from the genome sequence, and target mRNAs. These miRNA candidates have features characteristic of both plant and metazoan miRNAs.


Introduction
Small noncoding RNAs (sRNAs) of 20-40 nucleotides (nt) have emerged as key players in both transcriptional and posttranscriptional gene regulation in an array of organisms [1]. They also play important roles in genome stability, chromatin organization, and virus resistance [2]. Since their discovery, diverse classes of sRNAs have been described, and new types of sRNAs and pathways continue to be reported. The most commonly referred to are microRNAs (miRNAs), small interfering RNAs (siRNAs), and Piwi-interacting RNAs (piRNAs), based on their biogenesis and Argonaute (AGO) protein-binding partners [3,4]. The biogenesis and function of sRNAs typically involves a core set of proteins, namely Dicer or Dicer-like (DCR or DCL; double stranded (ds)RNA ribonuclease), Argonaute (AGO; RNA binding, slicer ribonuclease), and RNA-dependent RNA polymerase (RdR; secondary sRNA production).
Oomycetes, also known as water molds, are fungus-like heterotrophs which are ubiquitous in marine, freshwater and terrestrial environments [5]. This group of organisms belongs to the lineage of biflagellate Stramenopiles within the kingdom Chromista [6,7]. Brown algae and diatoms are also grouped within the Stramenopiles, and thus oomycetes share an evolutionary history with photosynthetic organisms [8]. Oomycetes are also distantly related to the Apicomplexa, such as Plasmodium and Toxoplasma parasites [9].
Phytophthora infestans is a destructive plant pathogen best known for its role in the mid-19 th century Irish potato famine. This oomycete plant pathogen remains a threat to food security due to its impact on the world's fourth most important food crop, potato.
It also infects tomato, another globally significant crop. The lifecycle of P. infestans involves both asexual and sexual reproduction pathways ( Figure S1). Spore dispersal is mainly via asexual sporangia that can germinate directly within a few hours after landing on potato or tomato foliage. Alternatively, sporangia can release motile biflagellate zoospores that are attracted to potential plant hosts and new infection sites. Both propagule dispersal mechanisms lead to rapidly spreading disease epidemics. Further, this plant pathogen secretes numerous proteins (effectors) that are important for colonization and promotion of disease (pathogenicity) on its host plants. The diverse effector repertoire of P. infestans assists pathogen adaptation to a changing host environment, leading to frequent breakdown of introduced resistance in crop plants [10,11]. The most studied oomycete effectors are the RxLR and Crinkler (CRN) groups that are delivered to the host cytoplasm and are characterized by conserved peptide motifs required for translocation into the host [12,13].
The genome size of P. infestans is two-to four-fold greater than those of the other sequenced plant pathogenic Phytophthora species, P. sojae, P. ramorum [14], and P. capsici [15]. One reason for the increased size of the P. infestans genome (240 Mb) is the large number of transposable elements (TEs), chiefly retrotransposons, and repeated sequences. The P. infestans genome is organized into gene-rich regions that contain relatively little repetitive DNA, and gene-poor regions that are enriched for repetitive DNA that includes a high proportion of transposons [14]. The repeat-rich regions are also enriched for effector-encoding genes. It has been proposed that repeats may be associated with pathogenicity effector family expansion and genome reorganizations [14,16,17].
Gene silencing is known to occur in oomycetes, and this has been best studied in P. infestans, for which it has been exploited to determine the role(s) of specific genes in the lifecycle or during infection [18][19][20]. However, despite being the most intensely studied oomycete, the molecular mechanisms underlying gene silencing in P. infestans are not as well characterized as in other model eukaryotes. Expression of sense and antisense gene constructs, followed by nuclear run-on assays, suggests the involvement of transcriptional gene silencing processes [21]. This is most likely mediated through heterochomatin formation that relies on histone modifications such as deacetylation and methylation, but does not involve cytosine methylation [21,22]. The formation of heterochromatin may spread for short distances (hundreds of bp) away from the silenced region. In contrast, treatment of protoplasts with dsRNA results in transient RNA silencing, and was proposed to be initiated via the posttranscriptional silencing machinery [23,24].
While much of the focus in understanding RNA silencing has been directed towards mouse, zebrafish, Caenorhabditis elegans, Drosophila melanogaster, Arabidopsis thaliana, and Schizosaccharomyces pombe, comparatively few reports exist from the phylogenetic branch that includes the Stramenopiles and Apicomplexans. RNA silencing pathways in these unicellular eukaryotes have revealed some striking differences to several model organisms, such as atypical DCL proteins and unusual biogenesis of miRNAs [25,26]. To advance the understanding of gene silencing processes operating in P. infestans, we have previously shown that P. infestans possesses the components of canonical gene silencing pathways similar to those of other eukaryotes [27]. However, genes encoding silencing-related proteins such as cytosine methyltransferases, RNA polymerase IV, Drosha and ERI1, present in other organisms, were not identified from its genome sequence.
In this study we used deep sequencing (SOLiD) to characterize classes of sRNAs from two P. infestans isolates with contrasting virulence and pathogenicity. We determined the relative abundance and distribution of sRNAs in the predicted transcriptome, effector genes, and TEs. Of particular interest are questions regarding how P. infestans controls the excessive transposon load that has profoundly shaped its genome, and how traits such as pathogenicity and specific virulence/avirulence may connect to variations in the presence or prevalence of sRNAs. Our data revealed the expected 21 nt sRNAs, but new sRNA size classes of 25/26, 31/32, 35 and 40 nt were also identified. Most commonly, sRNAs were derived from TEs, followed by CRN and RxLR effector genes. Generation of 21 nt sRNAs involves PiDcl1 (Dicerlike), as revealed by analysis in P. infestans lines silenced for PiDcl1. Furthermore miRNA sequences were predicted by combining sRNA sequencing with precursor prediction from the genome sequence.

P. infestans Produces at Least Three Size Classes of sRNA Molecules
Isolate R0 is weakly pathogenic on potato plants lacking any known resistance genes. In contrast, isolate 3928A is highly pathogenic, and exhibits specific (gene-for-gene) virulence on potato plants carrying the R1-R7, and R10-R11 resistance genes [28]. Compared to isolate 3928A, R0 produces relatively few sporangia, which in turn produce relatively few zoospores (and subsequent cysts). From both isolates, sRNA fractions were prepared from cultured non-sporulating mycelia, sporangia, germinating sporangia, and germinating cysts. Sequencing of sRNAs from eight SOLiD libraries generated a combined total of 12.8 million sequence reads for 3928A and 15.3 million for R0 (Table S1). Small RNAs that were derived from tRNAs and rRNAs, representing 2-4% of total sequence reads, were filtered out before mapping to other datasets. To assure that the sRNA sequences were not simply the products of mRNA turnover, they were first mapped to a small set of highly expressed single copy number genes that included (as examples) actin, ß-tubulin, and ubiquitin from P. infestans. Consistent with other studies that have used ABI SOLiD sequencing of sRNAs [26], a large proportion of the remaining sequence reads did not match any of the genomic sequences and were discarded. A high proportion of unmapped sRNAs has been reported previously from SOLiD sequencing experiments and may be attributable to the level of filtering applied [29]. That is, SOLiD reads are reported immediately without filtering, while other sequencing technologies such as 454 and Illumina incorporate a pre-filtering step. Of the sRNAs not derived from tRNAs and rRNAs, 6.6-15.6% mapped to the dataset of the entire genome sequence (core set), and a further 6.2-17.0% mapped to the dataset of unplaced sequence reads from the P. infestans genome database (www.broadinstitute.org) (Table S1; Figure S2A, B). The exact composition of the unplaced sequence reads is not well defined, but is likely to contain highly repetitive sequences that could not be accurately placed in the assembled P. infestans genome. Of the core set, 46.2-87.3% were homologous to the total predicted transposons, 14.5-23.5% mapped to the total predicted mRNAs, 0.1-0.3% to RxLR and 0.1-1.9% to CRN effector genes (Table S1).
Alignment of all sRNA sequences in the core set to the entire P. infestans genome sequence revealed an enrichment for size classes of 21 and 25/26 nt ( Figure S2A). Although not forming a clear peak in the histogram, sRNAs in the range of 31-33 nt were also prevalent. Indeed, sRNAs of 32 nt were the most abundant size class in the R0 isolate. All of the sRNA libraries sequenced displayed a similar size distribution of sRNAs. The exceptions to these were the sporangia and germinating cysts libraries from R0. These showed relatively low abundance of 21 and 25/26 nt sRNAs, and enrichment for 31-33 nt sRNAs (data not shown). The existence of multiple sRNA size classes in P. infestans was in contrast to findings from organisms such as D. melanogaster, Dictyostelium discoideum, Toxoplasma gondii, and Cryptococcus neoformans, which exhibited marked abundance of a single sRNA length [25,[30][31][32]. In the diatom Thalassiosira pseudonana, SOLiD sequencing of sRNA identified enrichment for 22 nt sRNAs, while longer sRNAs were identified by 454 sequencing [26]. Similarly, in members of the ''green lineage'', such as the moss Physcomitrella patens, and higher plants Populus trichocarpa and A. thaliana, more than one size class of sRNA is also prevalent [33][34][35].

Abundant sRNAs Derived from CRN Effector Genes and Long Terminal Repeat (LTR) Retro-transposons
We aimed to determine if mapping of sRNA sequences to the specific sequence datasets of TEs and effector genes (RxLRs, CRNs), could reveal a differential distribution of matching sRNAs. We focused on these three categories since the expression of P. infestans effectors is considered to be the main determinant of pathogenicity [36], and P. infestans has the most transposon-rich genome of the oomycetes [14]. Transcripts from TEs have also been identified previously among P. infestans expressed sequence tags and in microarray hybridizations [14,37,38]. Furthermore, over half of the RxLR effector genes predicted from the P. infestans genome are within 2 kb of transposon-derived sequences, and it has been hypothesized that RNA silencing of transposons may also influence expression of nearby RxLR effectors through formation of heterochromatin [17,39].
An overwhelming majority of 21 nt sequences was the most striking feature of sRNAs derived from CRN genes (67-77% of all CRN-mapping sRNAs; Figure 1C). In comparison, sRNAs mapped to TEs were most prevalent at sizes of 21 (17-22%), 25/26 (35-38%), and 31-33 nt (3-11%) ( Figure 1A). Small RNAs mapped to RxLR effector genes also exhibited enrichment in specific size classes, although the distribution was not as narrow as for CRN effector genes ( Figure 1B, C). Small RNAs mapping to the unplaced sequence reads from the P. infestans genome database (www.broadinstitute.org) revealed a marked enrichment (28%) of 32 nt sRNAs ( Figure S2B) in the R0 isolate, similar to that found for sRNAs mapped to the entire genome ( Figure S2A). The R0 isolate also exhibited a higher proportion of 31-33 nt sRNAs derived from the TE (11%) and RxLR (23%) categories ( Figure 1A, B).
We assessed the comparative abundance of sRNAs derived from each RxLR, CRN, and transposon group in all of the libraries from 3928A and R0 (Table S2). For the majority of transposon or effector gene sequences, the difference in sRNA abundance was no more than two-fold between the two isolates for comparable libraries (for example R0-mycelium versus 3928A-mycelium). The maximum difference for transposon-derived sRNAs was observed in the germinating cysts libraries, where sRNAs from two different novel PiggyBAC transposons were 5-fold to 9-fold more abundant in 3928A compared to R0, respectively. The greatest difference for sRNAs derived from RxLR and CRN effector encoding genes was found between the germinating cyst and sporangia libraries for the two isolates. Here, sRNAs derived from CRN (PITG_18133) were found to be 84-fold more abundant in 3928A. The CRN effectors have been proposed to represent an ancient class of pathogenicity effector proteins that have evolved through recombination among gene family members, and many CRN gene copies exist as pseudogenes or have been subject to transposon insertion [14].
However, little is known about how they are regulated, or how diversity in their expression may be generated among different P. infestans genotypes. Our finding of significant accumulation of sRNAs to this effector class may be the first indication for involvement of silencing processes in CRN gene regulation.
The fact that the majority of sRNAs were homologous to TE sequences was not surprising since the P. infestans genome is predominantly composed of these elements. The P. infestans genome hosts a wide diversity of highly repeated transposons including: Gypsy and Copia LTR retrotransposons, Cryptons, Helitrons, DIRS, hAT, Mutator, Mariner, Dodo, Piggy-BAC, Pogo, and other novel DNA transposons and LTR retrotransposons [14]. To determine if the class of TE resulted in a specific sRNA size distribution or sense/antisense orientation, sRNA sequences were matched to datasets of specific groups of TEs ( Figure S3, S4). The majority of sRNAs in our datasets were derived from LTR retrotransposons ( Figure S3B); these are the most prevalent class of TEs in P. infestans and account for over 30% of the entire genome [14]. LTR retrotransposons were distinguished by an approximately two-fold greater abundance of 21 nt sRNAs, compared to 25/26 nt sRNAs ( Figure S3B). Conversely, Mutator and Mariner DNA transposons ( Figure S3H, I) showed an approximately twofold greater abundance of homologous 25/26 nt sRNAs, compared to 21 nt. However, the most striking example of sRNA size bias was observed for HelENtrons (endonuclease-containing Helitron). The sRNA size distribution for this class of TEs was most similar to that observed for CRN effector genes, with an approximately ten-fold bias towards 21 nt sRNAs, compared to 25/26 nt sRNAs ( Figure S3N).
DCR digestion of long dsRNA yields duplex sRNAs, and typically only the antisense strand (complementary to the target transcript) is incorporated into the AGO-containing RNA-induced silencing complex (RISC), and the sense strand is degraded [40]. Thus, an excess of antisense-aligning sRNAs may be expected when sequencing of sRNAs is performed. To determine if such bias was evident in our sRNA core dataset, we separated the sRNA read sequences into those aligning in sense and antisense directions. For some TE sequences, very few sRNAs were identified, and in these instances it is difficult to draw any clear conclusions about bias in sRNA orientation. For TEs and genes where sRNAs were commonly detected, there was no obvious bias toward antisense sRNAs in our data ( Figure S4). This suggests that process(es) other than the typical dsRNARDCLRAGO pathway may be operating in P. infestans.

Small RNAs are Preferentially Derived from Specific Sites in TEs and Effector-encoding Genes
To determine if sRNAs were derived from specific sequence regions, a selection of TEs and genes encoding RxLR and CRN effectors with a high abundance of homologous sRNAs were analyzed in more detail. Gypsy Pi-1a and a novel Gypsy LTR element called 'Albatross' are the two most abundant TEs and combined, account for 29% of the P. infestans genome [14]. Small RNAs mapped throughout the length of these two TE classes in both sense and antisense orientations in all lifecycle stages ( Figure  S5A, B), with notable 'hotspots' of mapping sRNAs. This distribution is consistent with mapping of sRNAs to LTR transposons in other organisms [30,31,41].
Cryptons, a unique class of transposons that use a tyrosine recombinase for transposition [42], are present in P. infestans and other oomycetes [14]. Crypton6 is approximately 3.2 kb long, and sRNAs were identified that mapped across the entire length of the element at low read depth in both sense and antisense orientations. The exception to this distribution pattern was at the 39 end (at approx. 3 kb) to which abundant antisense sRNAs were identified in all lifecycle stages sampled ( Figure S5C). This specific region corresponds to a tRNA-like infSINEp insertion [43], for which the RNA is predicted to exhibit strong secondary structure ( Figure  S5D).
From our analysis, two closely related CRN genes, PITG_09052 and PITG_09053, exhibited marked foci of mapped sRNAs. These two genes differ by few nucleotides that are distributed across the length of the genes, and these single nucleotide sequence differences greatly influenced the mapping of sRNAs ( Figure   S5E, F). For both of these predicted effector genes, 21 nt sRNAs mapped predominantly to three main sites in the 59 half of the coding region. CRN effectors consist of recombined conserved regions [14]. However, none of the sRNA foci identified here corresponded closely with proposed sites of recombination in the two CRN effector genes.
PITG_23226 encodes a predicted secreted RxLR effector that exhibits elevated transcript accumulation in infected potato leaves [14]. Abundant sRNAs in both sense and antisense orientations were mapped specifically to the 39end of this gene in both isolates and all lifecycle stages ( Figure S5G). Examination of the genomic DNA sequence for PITG_23226 revealed that the 39 end of the gene overlaps with a Copia LTR retrotransposon, leading to probable convergent transcription. The overlap region corresponds to the site from which the majority of the matching sRNAs were derived.

Validation of SOLiD sRNA Sequences by Northern Hybridization
While our SOLiD sequencing has yielded primary insights into the sRNA populations of P. infestans, this technology is known to be biased towards sequences with 39 secondary structure and a higher diversity of nucleotides near the 39 and 59 ends [44]. However, the same study also demonstrated that SOLiD provided a more dispersed sRNA length distribution and the most accurate estimate of sRNA abundance, compared to the two other protocols under study (Illumina versions 1 and 1.5). To confirm the existence of sequenced sRNAs, Northern blot hybridizations were employed for selected TEs and effector genes ( Figure 2; Figure S6). Riboprobes or DNA oligonucleotide probes specific to Copia3-LTR, Satellite2, and Crypton6 detected different size sRNAs for each element, ranging from 21, 26, 32, to 40 nt antisense sRNAs ( Figure 2A; Figure S6V, W). Antisense and sense 35 and 40 nt sRNAs, respectively, were detected for Gypsy Pi-1a ( Figure S6T, U).
Probing for sRNAs specific to CRN effectors PITG_22969 and PITG_18133 revealed the presence of the expected 21 nt antisense and sense sRNAs ( Figure 2A; Figure S6R, S), but only in the highly pathogenic isolate 3928A. This agrees with SOLiD sequencing, which had identified more abundant 21 nt sRNAs in sequenced libraries from 3928A. Among RxLR effector genes analyzed, 28, 30, 32 and 35 nt sense and antisense sRNAs were detected in both isolates for PiAvrblb1 (PITG_21388), PiAvr4 (PITG_07387), PiAvrblb2 (PITG_20300), PiAvr3a (PITG_14371), PITG_14783, and PITG_16240 ( Figure 2A; Figure S6A-H, K, L, O, P). Sense sRNAs of 25 and 28 nt were detected in both isolates for PITG_06308 and PITG_15123, and antisense sRNAs of 26, 28, and 35 nt for PITG_06478, PITG_14736, and PiAvr3b (PITG_18215) ( Figure S6I, J, M, N, Q). Small RNAs of 21 nt were not detected for any of the RxLR effector genes assessed by Northern hybridization in this study. This agrees with our sequencing results which indicated that 25/26, and 30-33 nt sRNAs were the most prevalent sizes for the RxLR effector genes selected for Northern hybridizations.
At least three size classes of 21, 25/26 and 31/32 nt were found by deep sequencing from P. infestans, whereas Northern blot hybridizations detected additional size classes of 35 and 40 nt that were beyond the maximum SOLiD sequence read length in this study. At the time of this study, SOLiD read lengths were comparable with other next-generation sequencing technologies such as Solexa/Illumina. Small RNAs of 21 and 40 nt have been identified previously in P. infestans [24,45]. However, the identification of additional size classes of sRNA (25/26, 32, 35 nt) represents new findings in P. infestans. Small RNAs of 21 nt are highly prevalent in many eukaryotes [31,[46][47][48][49]. There have also been reports of 25/26 nt sRNAs in the distantly related apicomplexan T. gondii, fungi such as Mucor circinelloides, and plants [25,50,51]. These sRNAs have roles in sequence-specific mRNA degradation, systemic silencing, and methylation of homologous DNA. Longer sRNAs have been reported, which also have origins in RNA silencing [52,53]. We will address the possible mechanism of sRNA biogenesis in P. infestans in a later section.
We had hypothesized that RNA silencing of effector-encoding genes, mediated by sRNAs, may contribute to the differences in pathogenicity and specific virulence against potato resistances in the two contrasting isolates analyzed here. Although sRNAs were detected by Northern blot hybridizations for all effector genes tested, differences in relative abundance of sRNAs were found between the two isolates only for some of the RxLR effectorencoding genes. Small RNAs homologous to PiAvrblb2, PiAvr3a and PITG_14783 were found to be more prevalent in isolate R0 ( Figure S6C-F, K, L). While sRNAs derived from PITG_14736, PITG_15123, and PiAvr3b were detected in both isolates, their accumulation appeared to be greater in isolate 3928A ( Figure  S6M, N, Q). To date, PiAvr3a and PiAvrblb2 are the only RxLR effector proteins that have been found to contribute to pathogenicity in P. infestans [45,54,55]. Our discovery of sRNAs specific to PiAvr3a and PiAvrblb2 as more prevalent in the weakly pathogenic R0 isolate ( Figure S7) further supports our hypothesis that RNA silencing may contribute to the variation in pathogenicity in different isolates by down-regulating specific effectors.

Small RNAs may Affect Accumulation of Transcripts Encoding RxLR and CRN Effectors
RxLR and other effector genes typically exhibit marked transcript accumulation during infection of plants [12,14]. To examine if there were differences in the transcript accumulation level of RxLR effectors for which sRNAs were found, real-time (quantitative) reverse transcription polymerase chain reaction (qRT-PCR) was carried out on total RNA preparations from potato leaf samples infected with the R0 and 3928A isolates ( Figure 3; Figure S8). Silencing in P. infestans is due to heterochromatin formation [21] and persists throughout infection of plants [20,45,54]. This signifies that silencing involving sRNAs in vitro will also be detectable during infection. First, genes encoding selected RxLR avirulence effectors were examined, since these are recognized by genetically defined major resistance proteins in potato, and act to trigger disease resistance responses. Additional effectors to which sRNAs were detected were also assessed for transcript accumulation. The PiAvr1 gene was used as an internal control, as this gene is known to be absent from 3928A [28]. As expected, PiAvr1 transcript was clearly detected in R0, while there was no detection of PCR products in 3928A ( Figure 3B). PiAvr3a has been shown to be an essential effector for P. infestans pathogenicity [45,54]. Compared to 3928A, PiAvr3a transcript accumulated to 227-fold lower levels in R0 ( Figure 3A), consistent with sRNAs being readily identified ( Figure S6E, F). In contrast, sRNAs homologous to PiAvr3a were either difficult to detect or undetectable in the highly pathogenic 3928A isolate ( Figure S6E, F), while PiAvr3a transcript accumulated to high levels in this isolate. In previous experiments, silencing of PiAvr3a to transcript levels similar to those found here for R0 led to a loss or reduction in pathogenicity of silenced lines [45,54]. Thus, the weakly pathogenic infection phenotype ( Figure S7), accumulation of sRNAs to PiAvr3a ( Figure S6E, F), and diminished transcript levels for PiAvr3a in isolate R0 ( Figure 3A) are consistent with endogenous RNA silencing of this gene.
Transcripts from PITG_14783, PITG_14736, and PITG_15123 also accumulated to significantly higher levels in 3928A compared to R0 ( Figure 3I; Figure S8A, B). For PITG_14783, this is consistent with greater levels of homologous sRNAs in R0 ( Figure  S6K, L). Transcripts of PITG_14736 accumulated to very low levels in both R0 and 3928A ( Figure S8A), consistent with the identification of sRNAs for this gene in both isolates ( Figure S6M). Indeed, PITG_14736 transcripts could not be detected during any stage of infection for R0. Small RNAs were detected in both isolates for PITG_15123 ( Figure S6N), which is a member of a gene family comprised of three near-identical members [14]. It remains to be determined which PITG_15123 gene family member is actively transcribed, or the source of homologous sRNAs.
The maximum transcript levels of PiAvrblb1, PiAvrblb2 and PITG_06308 were similar in both isolates, although the peak of transcript accumulation was delayed in isolate R0 ( Figure 3C-E), reaching a maximum at 72 hpi. This delay in transcript accumulation for these genes may be due to the slow progression of colonization for the R0 isolate.
Transcripts encoding effectors PiAvr3b, PiAvr4, PITG_23226, PITG_22969, and PITG_18133 accumulated to lower levels in 3928A than R0 ( Figure 3F-H, J; Figure S8C). This was expected for PiAvr4, since virulent isolates possess a truncated form of this gene that is either not expressed, or only expressed at a low level [56]. PITG_22969 and PITG_18133 encode CRN effectors for which sRNAs were more abundant in 3928A (Figure 2A; Figure  S6R, S), consistent with the markedly lower levels of these transcripts in that isolate. Isolate R0 is avirulent on plants with the R3b resistance gene, while isolate 3928A is virulent. The lower levels of transcript accumulation observed here for PiAvr3b are consistent with greater abundance of sRNAs for this gene in 3928A ( Figure S6Q). Although the mechanism by which PiAvr3b is modified in 3928A to avoid detection by R3b has not been described [57], results here suggest that a reduction in transcript level may contribute to its evasion of detection. Such a mechanism has been found in the P. sojae system where the PsAvr3b avirulence effector (sequence unrelated to PiAvr3b) evades detection by the matching Rps3b resistance gene in soybean plants by a combination of sequence variation and reduced transcript accumulation [58].
The RxLR effector PITG_23226 is part of a small gene family comprising three members (including PITG_23014 and PITG_23015) that differ by few nucleotides. The qRT-PCR primer pair used for PITG_23226 will amplify all three genes from cDNA. Considering this, the overall level of transcript accumulation is low for this gene family ( Figure 3H). The overlap of the 39 end of PITG_23226 with a Copia LTR retrotransposon led to an accumulation of sRNAs across this region and may also restrict transcript accumulation for this gene and its paralogs. This is consistent with other findings showing that transcriptional fusion of a SINE to the PiAvr3a gene led to silencing of both the transgene and endogenous PiAvr3a transcripts [45].
Although sRNAs to all of these RxLR effector-encoding genes were identified in both R0 and 3928A, it is obvious that there are additional factors that may contribute to the level of their transcript accumulation. One consideration is that sRNAs may require accumulation to a threshold level to influence gene expression. Alternatively, many RxLR effector genes for which sRNAs were detected are present in the P. infestans genome as gene families comprising very closely related members [14]. From sRNA sequencing and Northern hybridizations, it is difficult to distinguish whether the sRNAs that map to RxLR families may have originated from transcriptionally active or inactive members and which paralog(s) these sRNAs may act on. PiAvrblb2 is one such example [59] and this consideration may explain why sRNAs were readily detected for this effector in R0 ( Figure S6C, D), although little difference in transcript accumulation was observed for this gene between the two isolates ( Figure 3D). That is, PiAvrblb2 transcripts may originate from gene copies that are not influenced by sRNAs.

Small RNAs in P. infestans have Unmodified Termini and Exhibit Preferences for Terminal Nucleotides
Small RNAs often exhibit specific nucleotide preferences or chemical modifications that are characteristic of the pathway involved in their processing and sorting by AGO proteins. This may be observed as a preference for the 59 terminal base, which in numerous organisms has been determined as uracil [31,32,41]. Although the terminal nucleotide of the siRNA has a negligible effect on AGO sorting in flies and mammals, it has a major impact on sorting in plants [4,60]. Analysis of the 59 base of all P. infestans sRNA sequences mapping to TEs showed uracil to be the most prevalent 59 nucleotide compared to other datasets ( Figure 1D-F; Figure S2C, D). However, when the 59 nucleotide of each sRNA size class was examined, the 21 nt sRNAs most commonly had cytosine as the most prevalent 59 base. This observation contrasts with the 59 nucleotide of the 25/26 nt sRNAs, for which uracil is most prevalent ( Figure S9). In D. melanogaster, non-miRNA 21 nt endogenous siRNAs also have a 59 preference for cytosine, while Piwi/Aubergine/Ago3-bound sRNAs of 23-29 nt have a 59 nucleotide preference for uracil [31].
To further study the 59 and 39 ends of sRNAs, we examined specific sRNAs that had been identified from our deep sequencing analysis. The assay used for 59-end analysis of sRNAs involved digestion with terminator exonuclease, a processive 59 to 39 exonuclease that specifically targets and degrades RNA with a 59monophosphate, but not sRNAs with 59-triphosphates. Northern hybridization to identify candidate 21 nt and 32 nt sRNAs to Copia3-LTR and Crypton6, respectively, demonstrated that these sRNAs are prone to degradation by terminator exonuclease, confirming that the 59 ends have monophosphates ( Figure 2B). As a control, the RNA was treated with tobacco acid pyrophosphatase (TAP; that converts 59-triphosphate RNA into 59monophosphate) and sequential reaction with TAP followed by terminator exonuclease. No sRNA degradation was detected after TAP treatment, while the TAP and terminator exonuclease treatment led to complete sRNA degradation. These data indicate that P. infestans sRNAs mapped to Copia3-LTR and Crypton6 are  29-O-methylation of the 39 terminal ribose is an important modification that stabilizes sRNAs such as miRNAs, siRNAs in plants, piRNAs in animals, and siRNAs in D. melanogaster [64]. 29-O-methylation acts as a protective mechanism against 39R59 degradation and 39 uridylation of sRNA in various species. This modification is carried out by sRNA methyltransferase HUA ENHANCER1 (HEN1) which methylates sRNAs in A. thaliana, whereas its ortholog has similar function in other plants, animals and flies. There is no readily identifiable ortholog of HEN1 in P. infestans, although it does possess several predicted RNA methylases. Sodium periodate treatment, followed by b-elimination, was used to examine 39 ends of P. infestans sRNAs, specifically for the presence of hydroxyl (2OH) groups at 29 positions on the terminal ribose of Copia3-LTR sRNAs. If both hydroxyls are present, periodate acts on the hydroxyl groups, oxidizing them to an unstable dialdehyde, subsequently leading to b-elimination of the terminal nucleotide [65]. After the treatment sRNAs will be one nucleotide shorter, leading to a downward mobility shift. Copia3-LTR sRNAs from P. infestans were found to be sensitive to periodate ( Figure 2C). Consistent with observations on the absence of a HEN1 ortholog from P. infestans, these results provide evidence that sRNAs do not have 39 methylation. The examined sRNAs have 59-monophosphate and 39-OH and thus are most likely to be DCL or AGO dependent products [66].

Knock-down of DCL and AGO Reveals their Roles in sRNA Biogenesis
A question that arises from our sequencing of sRNAs from P. infestans pertains to how a single DCL (PiDcl1), a DCL-like helicase (PiRnh5) and RdR (PiRdr1), and four distinct AGO proteins (PiAgo1-5) [27] can lead to such a diverse size repertoire of sRNAs. The involvement of some of these genes in maintaining gene silencing in P. infestans was shown previously when PiDcl1, PiAgo1 and histone deacetylase PiHda1 were transiently silenced by exogenous application of homologous dsRNA [27]. In those experiments, small numbers of lines exhibited significant silencing of the genes encoding components of the silencing pathway. Similar to P. infestans, the diatom T. pseudonana also has an atypical DCL protein and a DCL-like RNA helicase, but differs in having a single AGO protein [26]. However, the diatom genome contains a repetitive DNA content (including transposons) of approximately 2%, compared with 74% in P. infestans [14,67]. Silencing of PiDcl1, PiAgo1-5, and PiRdr1 genes was performed here in stable transformants of P. infestans (88069 isolate background) using inverted repeats to initiate silencing by RNA interference. Transgenic lines were recovered for most of the genes, and for each gene at least one line exhibited significant knock-down of transcript to below wild-type levels ( Figure S10). The level of transcript knock-down ranged from 0.8% of wild-type levels for PiRnh5, to 48% for PiAgo5. The exceptions to this were PiRdr and PiAgo3, where no lines showed evidence of silencing. The low numbers of silenced lines, and the transcript levels, are consistent with experiments using transient silencing [27], and highlight the technically challenging nature of carrying out these experiments in this system. As may be expected for genes involved in RNA silencing, no transgenic lines showed complete silencing. The gene knock-down lines chosen for analysis of sRNAs were D1t7 and D1t8 (PiDcl1), D2t1 (PiRnh5), A1b (PiAgo1/2), A4a (PiAgo4) and A5c (PiAgo5). For further analysis of sRNA biogenesis, samples from the assessments when knock-down of silencing gene transcripts was most apparent were used in Northern hybridizations.
The narrow distribution of sRNA sizes for CRN genes suggests a specific mechanism for generating 21 nt sRNAs. Previously in P. infestans, 21 nt sRNAs have only been detected in experiments where partial silencing of the inf1 gene was observed [24]. Northern hybridizations were performed to detect 21 nt sRNAs to Copia3-LTR and PITG_22969 (CRN) in PiDcl1 and PiRnh5 silenced lines ( Figure 4A, B). Small RNAs of 21 nt from Copia3-LTR were absent only in PiDcl1 line D1t8, consistent with the most pronounced reduction in PiDcl1 transcript. Conversely, in parallel to the reduction in accumulation of 21 nt sRNAs in this line for Copia3-LTR, there is also accumulation of 24/25 nt sRNA. Taken together, these results suggest that, at least some 21 nt sRNAs may originate from processing of longer sRNAs that involves PiDcl1. Alternatively, another uncharacterized component of the silencing pathway is responsible for generating the 24/25 nt sRNAs homologous to Copia3-LTR and this enzyme would have increased substrate abundance in the PiDcl1 silenced line.
Silencing of PiRnh5 (DCL-like RNA helicase) indicates that, apart from PiDcl1, PiRnh5 is also required for generation of 21 nt sRNAs to PITG_22969 (CRN) ( Figure 4B). However, silencing of PiRnh5 had no effect on accumulation of 21 nt sRNAs derived from Copia3-LTR ( Figure 4A). This suggests that PiRnh5 is required for silencing specific classes of genes in P. infestans. Studies in C. elegans have also shown that the helicase domain of DCR is required for accumulation of some, but not all of the endogenous siRNAs [68]. PiDcl1 has only poorly conserved partial helicase domains and no PAZ domain, and it might be speculated that PiRnh5 may act together with PiDcl1 to generate certain populations of 21 nt sRNAs. Furthermore, our results suggest that there are at least two pathways for generating 21 nt sRNAs in P. infestans: one pathway dependent on PiRnh5, and the other independent of PiRnh5. The mechanism by which P. infestans produces discretely sized and DCL-dependent 21 nt sRNA without a PAZ domain remains to be determined. The PAZ domain is considered to accurately size sRNAs during DCR processing [69,70]. Similar to P. infestans, some other unicellular eukaryotes also possess DCL proteins that have no PAZ domain and may also be missing dsRNA binding or helicase domains [25,71].
Small RNAs of 32 nt mapping to Crypton6 or PiAvrblb1 were unaffected by silencing of PiDcl1 and PiRnh5 individually, suggesting that 32 nt sRNAs are generated in DCL-independent pathways ( Figure 4C, D). However, there was a clear effect on 32 nt sRNAs homologous to PiAvrblb1 when PiAgo genes were silenced. Silencing of PiAgo4 or PiAgo5 led to a reduction in the accumulation of 32 nt sRNAs ( Figure 4E, F). Surprisingly, silencing of PiAgo1 resulted in elevated accumulation of 32 nt sRNAs (PiAvrblb1). Taken together, these results suggest that PiAgo4 and PiAgo5 are involved in the accumulation of 32 nt sRNAs, while PiAgo1 negatively impacts on 32 nt sRNA represents R0, and the dark bar represents 3928A. All calculations and statistical analyses were carried out as described in [90]. Error bars represent confidence intervals calculated using three technical replicates for each sample within the qRT-PCR assay. The abundance of mRNA for each gene is shown as a proportion of the actin A (PiactA) transcript on the y-axis of each graph. Amplifications repeated on independent occasions with different starting RNA and cDNA samples resulted in similar transcript accumulation profiles for all genes tested. doi:10.1371/journal.pone.0051399.g003 accumulation. Although the PiAgo proteins in P. infestans do not exhibit similarity to the PIWI class of proteins, the DCLindependent involvement of PiAgo4/5 in 32 nt sRNA accumulation could be speculated as analogous to the activity of PIWI in metazoans [52,53]. Alternatively, an as yet unidentified protein with RNase activity may act to generate 32 nt sRNAs in P. infestans. In S. pombe, an AGO-dependent pathway mediates generation of the siRNAs that are indispensable for formation of pericentromeric heterochromatin, most likely through 39-59 trimming of Ago1-bound priRNAs by the action of the RNA exosome [72].
Although sRNA pathways are generally conserved throughout plant and animal kingdoms, differences do exist between phyla. In species such as A. thaliana, Oryza sativa and M. circinelloides, there are two distinct classes of sRNA: 21 and 24/25 nt [50,51,73,74]. By comparison, 21 nt sRNAs predominate in Drosophila [31]. The different sRNA size classes may have specific roles in RNA silencing. For example, 21 nt sRNAs from A. thaliana are associated with post-transcriptional silencing whereas 24 nt sRNAs are associated with transcriptional silencing [50,75]. Earlier reports have found 21 nt sRNAs associated with partial silencing of the P. infestans inf1 gene from a longer dsRNA precursor, but were undetectable for transcriptionally silenced lines [24]. Based on those findings, and our results here, we can suggest a model for P. infestans where the PiDcl1-derived 21 nt sRNAs are associated with posttranscriptional silencing, while the longer sRNAs of 32 and 25/26 nt are associated with transcriptional silencing ( Figure 5).

Prediction of miRNA Candidates, Precursors, and Targets
miRNAs are abundant and major regulatory molecules that influence many essential cellular functions in a wide variety of organisms by modulating gene expression through translational repression or mRNA degradation. miRNAs have not been described from oomycetes previously, although this class of sRNA has been identified from the distantly related T. pseudonana [26]. We used the SOLiD sRNA sequence data to carry out a survey for putative miRNA candidates and their targets. Six putative miRNAs were predicted ( Figure 6; Table S3). Occurrence of both miRNA and miRNA*(star) sequences are key features of the existence of a miRNA-miRNA* duplex and the biogenesis of miRNAs [76]. In our data we could not detect any miRNA* sequences, probably due to low abundance of star sequences in the different lifecycle stages. The size of precursor miRNA candidates ranged from 55 to 110 bp and the size of the mature miRNA candidates ranged from 21 to 24 nt (Table S3). All of the predicted mature miRNAs were present in both isolates in at least four of the eight sequenced libraries. The origin of the six predicted miRNAs varied to great extent. Two of these candidates originate from intergenic regions, while two more were found in exons, one in a 59 untranslated region (UTR), and another in a 39 UTR. It might be speculated that the origin of predicted miRNA candidates in P. infestans are more plant-like, based on the majority of candidate miRNAs originating from intergenic regions and exons [77]. These findings are echoed by studies in T. pseudonana and P. patens [26,35]. In contrast, in animals the majority of miRNAs originate from introns [78]. We compared the predicted P. infestans miRNAs to the miRBase database to seek homology to known miRNAs across the entire sequence, and specifically in the seed region. No conservation with any known miRNAs was found, suggesting that these may represent a new class of miRNAs, as reported for diatoms and green algae [26].
Target prediction of candidate miRNAs used algorithms that incorporated plant-like and animal-like mapping parameters. TargetAlign prediction showed that P. infestans candidate miRNAs do not map to their targets with perfect complementarity. Plant miRNAs often have targets with perfect or nearly perfect complementarity, leading to target mRNA degradation [79,80]. Conversely, animal miRNAs have relatively low complementarity to their target mRNA, suggesting translation repression [78]. Our results suggest that binding of P. infestans miRNA candidates to target mRNAs could occur in a manner more akin to that for animal miRNAs. Targets for P. infestans miRNA candidates were predominantly predicted in the UTRs and coding regions. The target mRNAs encode proteins involved in biological processes associated with nucleic acid and general metabolism (Table S4). An exception was the PimiRNA4 candidate, for which no targets were found. Studies on animal and plant miRNAs have suggested that recently evolved miRNAs may originate without pre-existing target homology and acquire targets in the course of evolution [78]. It remains possible that this may be the case for PimiRNA4.

Conclusions
Similar to other eukaryotes, P. infestans produces endogenous non-coding sRNA molecules. Despite the marked differences in pathogenicity and specific virulence phenotypes for the two isolates from which sRNAs were sequenced, and the differences between the developmental stages, the sRNA populations were Crypton6 32 nt sRNAs in PiDcl1 (D1t7, D1t8) and PiRnh5 (D2t1) silenced lines. Silencing of these genes had no effect on the accumulation of 32 nt sRNAs. D. PiAvrblb1 32 nt sRNAs in PiDcl1 (D1t7, D1t8) and PiRnh5 (D2t1) silenced lines. Silencing of these genes had no effect on the accumulation of 32 nt sRNAs. E and F. PiAvrblb1 32 nt sRNAs in PiAgo1, 4, or 5 (A1b, A4a, A5c) silenced lines. The accumulation of 32 nt sRNAs is strongly reduced in PiAgo4 or 5 lines, while over-accumulation occurs for PiAgo1. Silenced lines used were PiDcl1 (Dicer-like; D1t8, t7), PiRnh5 broadly similar. Subtle differences were identified, such as an increased abundance of 32 nt sRNAs in isolate R0, and isolatespecific increases in sRNAs derived from specific effector genes. It is likely that these differences may associate with the different infection phenotypes for these two isolates.
We have shown previously that P. infestans possesses DCL and AGO proteins required for RNA silencing [27]. However from the P. infestans RNA silencing pathway comprising single DCL and RdR proteins, and four distinct AGO proteins, it can produce an unexpectedly diverse array of prevalent sRNA sizes. This is in contrast to other systems where a single sRNA size class typically heterochromatic regions or natural antisense transcripts and exogenous dsRNAs will lead to formation of long precursor dsRNAs. PiDcl1 processes these dsRNA into predominantly 21 nt siRNAs, in some instances in association with PiRnh5 (Dicer-like helicase). The 21 nt siRNAs generated by PiDcl1 associate with one of the PiAgo proteins, leading to degradation of target mRNAs. B. Long siRNA (lsiRNA) pathway. lsiRNAs are generated by PiAgo4 and 5 from transposons, coding genes, or overlapping regions of antisense transcription, or dsRNA generated by RNA polymerases. This may be a PiDcl1 independent pathway. Both pathways could be linked via one of the AGOs to heterochromatin formation aided by histone deacetylases and chromodomain proteins. RdR could further amplify the silencing. doi:10.1371/journal.pone.0051399.g005 predominates. Based on former [27,45] and present investigations in P. infestans, we speculate that there are at least two silencing pathways ( Figure 5). One is similar to the classical DCRdependent pathway acting on long dsRNA, processing it into 21 nt siRNAs. The other is an AGO-dependent pathway, that acts on transcripts arising from different sources and generates 25-32 nt long siRNA (lsiRNA). RdR would further amplify the silencing signal either by primer dependent or independent pathways. The siRNA and lsiRNA pathways could also be linked via one of the AGO proteins that might establish silencing via heterochromatin formation.
While the RNA silencing machinery in P. infestans is apparently deployed to silence the TEs that make up most of the genome, there are also many sRNAs homologous to genes encoding disease promoting effector proteins, especially CRN effectors. Endogenous RNA silencing of effector genes may contribute to variability in pathogenicity between different pathogen genotypes, such as for PiAvr3a in isolate R0 in this study. How silencing of these effectors is initiated, why only some effector gene paralogs may be affected, and the influence of nearby transposon sequences will be key questions in future studies directed at further understanding the influence of RNA silencing on pathogenic variation and adaptability in this economically important plant pathogen.

P. infestans Isolates
P. infestans isolates R0, 3928A, and 88069 were maintained at 20uC in darkness on rye agar medium amended with 2% sucrose [81], rifampicin (30 mg/ml) and pimaricin (10 mg/ml). Transgenic lines of P. infestans isolate 88069 were maintained as above, but agar medium was supplemented with geneticin (G418; 10 mg/ml). Isolates R0 and 3928A were used for analysis of sRNAs. R0 is a US1-like genotype [82] of A1 mating type and is weakly pathogenic only on potato containing no known resistances.
3928A is representative of the 'Blue 13' genotype currently dominant in British P. infestans populations [83]. It is highly pathogenic on potato, belongs to A2 mating type, and able to infect potato plants possessing the R1 to R7, R10, and R11 resistance genes [28].

Lifecycle Stages
To prepare mycelium samples for RNA isolation, P. infestans was grown in liquid pea medium [23] for 5 days at 20uC, and collected by gravity filtration. Sporangia were harvested from agar plate cultures as described [84]. Germinated sporangia were prepared by allowing sporangia to germinate in sterile distilled water overnight at 20uC. Germinating cysts were prepared as described previously [84]. This sample also contained some germinating sporangia and zoospores. The different samples were snap frozen in liquid nitrogen, and stored at 270uC until used for RNA extraction.

Pathogenicity Assessment
Susceptible potato cultivar Bintje was used to study late blight disease development for isolates R0 and 3928A. Infections were carried out as described [27]. Leaf samples were collected, snapfrozen in liquid nitrogen and stored at 270uC prior to RNA extraction.

RNA Isolation and cDNA Synthesis
Total RNA for qRT-PCR analysis was extracted from frozen samples using the RNeasy Plant mini kit (QIAGEN) following the manufacturer's protocol. Prior to cDNA synthesis, all RNA samples were DNase treated using the Turbo DNA-free kit (Ambion). Yield and integrity of the RNA was assessed using a NanoDrop Micro Photometer (NanoDrop Technologies), and agarose gel electrophoresis, respectively. First strand cDNA for qRT-PCR detection of P. infestans effector, silencing gene, and endogenous control PiactA transcripts were synthesized from 1 mg of total RNA by combined random hexamer and oligo dT priming using the First Strand cDNA Synthesis Kit (Quanta Biosciences) and following the manufacturer's protocol. Total RNA for sRNA detection was prepared by TRIzol protocol (Invitrogen). The lowmolecular-weight RNA was extracted using an enrichment procedure for sRNAs [85]. For Northern hybridizations, the quality of the extracted RNA was verified by ethidium bromide staining.

SOLiD Sequencing and Computational Analysis
Total RNA for deep sequencing was prepared from the different life cycle stages, described previously, using the mirVana TM miRNA Isolation Kit (Ambion), according to the manufacturer's protocol. The amount and quality of sRNA was estimated using an Agilent Bioanalyzer (Agilent Technologies) using settings recommended by the manufacturer. RNA library preparation and sequencing were performed using a SOLiD (version 2) platform (Applied Biosystems). Library preparation was according to the manufacturer's recommendations, using the sRNA library protocol provided with the SOLiD Total RNASeq kit (Applied Biosystems). SOLiD sequencing generated reads that were 35 bp long. Sequence barcodes were decoded, matching each bead sequence with the identity of the sample. Sequence analysis was performed using the SOLiD System Small RNA Analysis Tool (Applied Biosystems), permitting one mismatch in the seed and a total of three mismatches over the whole read. Adaptor sequences were removed, and all sequences that matched RNAs such as tRNA or rRNA were filtered out. Only sequences of 19 to 33 nt were used in further analyses. The remaining sRNA reads were aligned independently to the total genome sequence, and to separate datasets of P. infestans transposons, RxLR and CRN effectors, all predicted genes, and unplaced genome sequencing reads. Reads matching more than three locations were filtered out, except for when mapping to the whole set of transposons. All sequence datasets were downloaded from the Broad Institute (www.broadinstitute.org). Counting of sRNA lengths and starting base was done using in-house tools.

Detection of sRNA Molecules Homologous to Transposons and Effector Genes
Cloning of effector-encoding genes and TEs for generation of riboprobes was performed as described previously [45]. Oligonucleotide primers designed for effector-encoding genes and TEs sequences are shown in Table S5. For detection of sRNAs, 20-30 mg of the sRNA fraction was separated on 12-15% polyacrylamide Tris-borate-EDTA-urea gels. RNA was transferred to nylon membrane by capillary blotting, pre-hybridized, hybridized, and washed as described [85]. The membranes were exposed in phosphorimager cassettes and scanned with a Molecular Phosphorimager FX (Molecular Dynamics).

Plasmid Construction and Transformation
Phusion DNA polymerase (Finnzymes) was used to amplify P. infestans DNA using primers listed in Table S5. Cloning of amplified DNA into restriction endonuclease sites used standard protocols [86]. Insert orientation and integrity of all cloned DNA fragments were confirmed by DNA sequencing. The background plasmid vector used for all silencing constructs was pFTORA, a version of pSTORA modified by the inclusion of an FseI site (http://oomyceteworld.net/plasmids/plasmids.html). For silencing, PiDcl1, PiRnh5, PiAgo1-5, and PiRdr1 were cloned as inverted repeats into the pFTORA vector using the FseI and SbfI (sense direction), and AscI and SacII (antisense direction) restriction endonuclease sites. Transformations of P. infestans for all constructs were carried out using a modified polyethylene glycol-CaCl 2 -Lipofectin protocol (http://oomyceteworld.net/) [87,88]. All the transformants were maintained on rye-sucrose agar containing 10 mg/l G418.

RNA Secondary Structure Prediction
All RNA secondary structures were predicted using the RNA structure program, Mfold version 2.3 (http://mfold.rna.albany. edu/) [89] with default settings. The folding temperature was reset to 20uC to reflect the temperature used for growth of the P. infestans cultures. The RNA structures with the lowest free energy were adopted as the most likely structures.

SYBR Green qRT-PCR Assays
Oligonucleotide primers were designed and optimized to amplify P. infestans sequences (Table S5) and transcript levels were analyzed as described previously [27]. Transcript abundance in transgenic lines was compared with the level of transcript in calibrator samples (mycelium of wild type isolate 88069). The calibrator samples were assigned the relative value of 1.0, to allow comparisons to be made between lines. All qRT-PCR amplifications were repeated on independent occasions with different cDNA samples. All calculations and statistical analyses were carried out as described in ABI PRISM 7700 Sequence Detection System User Bulletin #2 (Applied Biosystems) [90]. Transcript abundance in wild type R0 and 3928A was assessed as above, except that transcript abundance from effector genes was assessed in samples from infected potato leaves. In each isolate, the abundance of effector gene transcripts in each sample was calculated relative to that of the Actin A (GenBank AAA33749; PITG_15117) transcript, using the difference between the threshold cycle (Ct) values for both genes. The amount of effector gene transcript was expressed as a percentage of the Actin A transcript [27,90]. Actin A was used as an endogenous control since we had previously shown that it exhibits the least variation in transcript accumulation in diverse developmental stages, compared to other transcripts tested [27].

Analysis of sRNA 59 and 39 Ends
For 59 end analysis of P. infestans sRNAs, 5 mg of sRNA-enriched samples were used for each reaction, separately, with terminator exonuclease, tobacco acid pyrophosphatase (TAP), and TAP followed by terminator exonuclease treatments [91]. Small RNA samples were treated with terminator exonuclease (2 U, 30uC, 1 hr), following the manufacturer's recommendations (Epicentre). In a separate treatment, sRNAs were also incubated with TAP (20 U, 37uC, 2 h) to remove b and c phosphates of sRNAs, leaving only the a phosphate group attached (converts 59-triphoshphate RNA into 59-monophosphate). After enzymatic treatment, all samples were extracted with phenol/chloroform and resolved on a 15% polyacrylamide gel. Northern blot analysis was performed using P 32 -labeled probe, to detect the sRNA of interest. Modifications at the 39-OH termini were determined using a belimination assay [65]. This reaction was carried out by treatment of sRNA (20 mg) with sodium periodate (25 mM, 25uC, 10 min), followed by heating to 45uC for 90 min. Treated sRNA were separated on denaturing 15% polyacrylamide gels and membranes probed with 59 end-labeled DNA oligonucleotides.

Prediction of miRNA Candidates and Targets
Prediction of miRNA candidates was performed using the mirDeep2 pipeline [92,93]. Small RNA reads of 21-24 nt that mapped to the P. infestans genome using SHRIMP2 [94] were used for prediction. All miRNA candidates were required to be present in at least two libraries, and have a minimum read count of ten reads for the mature sequence of the miRNA. No miRNA candidates were allowed to map in repetitive regions or to tRNA or rRNA. The folding of putative precursors was calculated using the software RNAfold [95]. The candidate miRNA sequences were also aligned to known miRNA sequences in the miRBase miRNA database [96][97][98][99] (www.mirbase.org). Target prediction for the candidate miRNA was calculated using the program TargetAlign [100]. Figure S1 Disease cycle of potato late blight caused by P. infestans. P. infestans mycelium from infected tubers or from germinating oospores, cysts, or sporangia spreads into shoots produced from infected or healthy tubers, causing discoloration and collapse of the host tissue. When the mycelium reaches the aerial plant parts it produces sporangiophores, which emerge through the stomata of stems and leaves and produce sporangia. When they are dispersed and reach a plant host, depending on the environmental conditions, they may germinate directly and cause new infections. Alternatively, sporangia may release motile zoospores that then encyst and germinate. The germ tube from the sporangium or cyst penetrates directly or via stomata, the mycelium grows intercellularly and intracellular haustoria are formed. Under wet or humid conditions, new sporangiophores emerge from the stomata and new infection cycles can start. In cool, moist weather new sporangia may form within four days. When the two mating types, A1 and A2 are present in the same plant tissue, fertilization can take place (sexual recombination) and oospores may be formed, providing a new source of inoculum to initiate infections. When infected plant debris fall to the ground and decompose, the oospores are released into the soil. The thickwalled oospores can over-winter in soil, in contrast to zoospores, sporangia and mycelia that are considered short-lived. cultivar Bintje (no known resistance genes) relative to the mRNA level in cultured non-sporulating mycelium (M). In each graph, the light grey bar represents R0, and the dark bar represents 3928A. All calculations and statistical analyses were carried out as described in [90]. Error bars represent confidence intervals calculated using three technical replicates for each sample within the qRT-PCR assay. The abundance of mRNA for each gene is shown as a proportion of the actin A (PiactA) transcript on the y-axis of each graph. Amplifications repeated on independent occasions with different starting RNA and cDNA samples resulted in similar transcript accumulation profiles for all genes tested. (TIF) Figure S9 59 nucleotide preferences of sRNAs in P. infestans isolates R0 and 3928A mapped to: A-B. All transposons, C-D. LTR retrotransposon subset, E-F. CRN effector encoding genes, G-H. RxLR effector encoding genes, I-J. All predicted mRNAs (includes RxLR and CRN effector genes). Empty columns in a panel indicate that no sRNAs of the marked size were identified. (TIF) Figure S10 Silencing of genes encoding components of the silencing pathway in P. infestans. Comparison of PiDcl1 (dicer-like) and PiRnh5 (dicer-like helicase), PiAgo (Argonaute1, 4, 5) transcript abundance (qRT-PCR), individually, in wild type (88069; wt) and inverted repeat silenced lines. A. D1t1-D1t14 (PiDcl1), B. D2t1-D2t3 (PiRnh5), C. A1a-A1d (PiAgo1), D. A4a-A4d (PiAgo4), E. A5a-A5c (PiAgo5). All calculations and statistical analyses were carried out as described in [90]. Error bars represent confidence intervals calculated using three technical replicates for each sample within the RT-PCR assay. (TIF)