Fusarium graminearum DICER-like-dependent sRNAs are required for the suppression of host immune genes and full virulence

In filamentous fungi, gene silencing by RNA interference (RNAi) shapes many biological processes, including pathogenicity. Recently, fungal small RNAs (sRNAs) have been shown to act as effectors that disrupt gene activity in interacting plant hosts, thereby undermining their defence responses. We show here that the devastating mycotoxin-producing ascomycete Fusarium graminearum (Fg) utilizes DICER-like (DCL)-dependent sRNAs to target defence genes in two Poaceae hosts, barley (Hordeum vulgare, Hv) and Brachypodium distachyon (Bd). We identified 104 Fg-sRNAs with sequence homology to host genes that were repressed during interactions of Fg and Hv, while they accumulated in plants infected by the DCL double knock-out (dKO) mutant PH1-dcl1/2. The strength of target gene expression correlated with the abundance of the corresponding Fg-sRNA. Specifically, the abundance of three tRNA-derived fragments (tRFs) targeting immunity-related Ethylene overproducer 1-like 1 (HvEOL1) and three Poaceae orthologues of Arabidopsis thaliana BRI1-associated receptor kinase 1 (HvBAK1, HvSERK2 and BdSERK2) was dependent on fungal DCL. Additionally, RNA-ligase-mediated Rapid Amplification of cDNA Ends (RLM-RACE) identified infection-specific degradation products for the three barley gene transcripts, consistent with the possibility that tRFs contribute to fungal virulence via targeted gene silencing.


Selection of sRNAs with sequence homology to plant genes
We looked for interaction-related fungal sRNAs that potentially could interfere with plant gene expression by sequence-specific silencing. To this end, a previously published sRNA sequencing data set of Fg sRNAs from an axenic IFA65 culture [10] was analysed for sRNAs with sequence complementarity to barley genes. In order to identify a wide range of potential targets, we applied only two selection criteria, namely i. size (21-24 nt) and ii. a minimal number of reads (at least 400 reads in the dataset). From a total of 35,997,924 raw reads, 5,462,596 (comprising 589,943 unique sequences) had a length of 21-24 nt. From the unique sequences, 1,987 had at least 400 reads. Since the IFA65 genome has not been sequenced, we used the published genome information of Fg strain PH1 (genome assembly ASM24013v3 from International Gibberella zeae Genomics Consortium: GCA_000240135.3) for further analysis. The majority of the 1,987 unique sRNAs mapped to rRNA (64.4%) and intergenic regions (21.6%), while 3.7% and 2.4% mapped to protein coding genes and tRNAs, respectively, and 7.8% did not perfectly match the reference genome (S1 Fig). According to the TAPIR algorithm, the 1,987 sequences overall matched mRNAs of 2,492 genes (Hordeum vulgare IBSC PGSB v2 reference genome; [38]) sufficiently close according to the refined target prediction criteria suggested by Srivastava et al. [39]. GO-enrichment analysis revealed an enrichment in functions of nucleotide binding, motor activity and kinase activity and processes such as transport and localization (S2 Fig). Most of the 14,156 transcripts of the 2,492 target genes, which we nominated as potential sRNA targets, showed partially homologous sequences to more than one sRNA accounting for a total of 17,275 unique pairs of potential target gene-sRNA combinations. Target prediction results are presented with only one transcript (splice variant) for every combination (S1 Table). Of note, merely 101 out of the 1,987 sRNAs had no predicted target among the total number of 248,391 plant mRNAs in the IBSC_PGSB_v2 annotation.

Barley immune genes accumulate to higher levels in PH1-dcl1/2-infected leaves
From the set of 2,492 barley genes with partial sequence homology to Fg sRNAs, we selected 16 genes for further analysis, based on an educated guess that they are potentially involved in biotic stress reactions during plant-fungal interaction (Table 1). When tested with RT-qPCR, we found eight genes, being targeted by a total of 104 unique Fg sRNAs, significantly higher expressed (Student's t-test, paired, � p<0.1, �� p<0.05, ��� p<0.01) in leaves infected with PH1-dcl1/2 vs. PH1 (Fig 2). Among these genes are three that encode proteins involved in the regulation of either ethylene (ET) (Ethylene overproducer 1-like 1, HvEOL1) or auxin responses (Auxin response transcription factors HvARF10 and HvARF19) and three kinases, of which Somatic embryogenesis receptor-like kinase 2 (HvSERK2) and BRI1-associated receptor kinase 1 (HvBAK1) are likely involved in recognition of microbe-associated molecular patterns (MAMPs). Moreover, genes encoding the plastid kinase 2-Phosphoglycolate phosphatase 2 (HvPGLP2), Resurrection 1 (HvRST1, with a rather elusive function in cuticle formation and embryo development), and the histone-lysine N-methyltransferase Su(var)3-9-related protein 5 (HvSUVR5, involved in transcriptional gene silencing) were also strongly expressed.
The first column gives the name of the respective gene abbreviated as in Fig 1 and in full. The second column shows the respective accession. In the third column selected GO terms are shown. The fourth and fifth columns give the accession and abbreviated name of the closest homologue in At.  Fig 3A). In accordance with the results obtained with strain PH1, HvEOL1 was also significantly (p = 0.029, Student's t-test (Δct), one sided, paired) downregulated in response to IFA65 infection compared to mock controls treated with 0.02% Tween20 (Fig 3B). In contrast, however, when leaves were sprayed with dsRNA-dcl1/2 prior to inoculation with IFA65, HvEOL1 transcripts accumulated (p = 0.055, Student's t-test (Δct), one sided, paired) in comparison with the inoculated leaves sprayed with TE buffer (Fig 3C).
Fungal sRNAs targeting HvBAK1, HvEOL1, and HvSERK2 mRNAs are less abundant in PH1-dcl1/2 vs. PH1 To detect the abundance of specific Fg-sRNAs, originally identified by sequencing of axenic IFA65 mycelium, in PH1-infected plant tissue, we performed reverse transcription stem-loop qPCR [41].  (Table 2 and S2 Table). These two sRNAs show high sequence similarities among each other. To identify their origin, they were aligned to the genomic sequence of strain PH1 (GCA_900044135.1). We found that they match the gene Fg_CS3005_tRNA-Gly-GCC-1-9 encoding tRNA-Gly for the anticodon GCC. Of note, a larger cluster of 27 overlapping tRNA-derived fragments (tRFs) with more than 50 reads matching the tRNA-Gly gene sequence were detected (S6 Fig). To assess differential accumulation of tRFs from the Fg_CS3005_tRNA-Gly-GCC-1-9 cluster in leaves infected with PH1 vs. PH1-dcl1/2, sRNAs were reverse transcribed using hairpin-priming followed by qPCR amplification [41]. For this analysis, we chose Fg-sRNA-321, the most abundant tRF from this cluster, along with Fg-sRNA-1921, which targets all three GOIs and an additional tRF (Fg-sRNA-6717), which targets HvEOL1 and HvBAK1 (see Table 2) to assess the sensitivity of the assay. In the initial IFA65 dataset the Fg-sRNA-321 had a read count of 2,106, Fg-sRNA-1921 had 416 and Fg-sRNA-6717 had 86 from a total of more than 5 million reads (S7 Fig). This equals 386 reads per million (rpm) for Fg-sRNA-321, while in average unique reads had only 1.7 rpm. Using TAPIR [42], we also calculated the target score values for all three tRFs, which is a measure for the similarity between sRNA and target. A high value refers to more dissimilarities. Mismatches (MMs) increase the score by one point and G-U pairs by 0.5 points. These values are doubled if the respective MMs and G-U pairs are located between the second and 12 th nt of the sRNA (5'-3') because a high similarity in the seed region of the sRNA is especially important for RNAi [43]. Fg-sRNA-321 has a score of 4.5 for HvBAK1 and HvEOL1, Fg-sRNA-1921 has a score of 4, 3.5 and 6 for HvBAK1, HvEOL1 and HvSERK2, respectively and Fg-sRNA-6717 has a score of 5.5 and 4.5 with HvBAK1 and HvEOL1. In plants other than Arabidopsis, such as wheat and rice, a score cut off at 4 or 6 points lead to a precision of 82% or 62% and a recall of known interactions of 39% or 58% respectively according to Srivastava et al. [39]. All three fungal tRFs were detected in infected leaves, while they could not be found in uninfected leaves (Fig 4). Significantly lower amounts of Fg-sRNA-1921 (59%), Fg-sRNA-321 (56%), and Fg-sRNA-6717 (60%) were detected in PH1-dcl1/2 vs. PH1-infected leaves (Fig 4), showing that their biogenesis is DCL-dependent.

Fg-sRNA-321 and Fg-sRNA-1921 also match SERK2 in Brachypodium distachyon Bd21-3
Next, we assessed the possibility that Fg-sRNA-321, Fg-sRNA-1921 and Fg-sRNA-6717 also have sequence homologies in At and the model grass Bd. Target prediction with the TAPIR algorithm using the optimised parameters for At (score = 4; mfe = 0.7), could not detect potential targets in At ecotype Col-0. In contrast, these three tRFs matched the sequence of Brachypodium somatic embryogenesis receptor-like kinase 2 (BdSERK2) in Bd21-3 with a score of 3.5, 3 and 5.5, respectively (Table 2). We examined the expression pattern of BdSERK2 in response to leaf infection: BdSERK2 is relatively weakly expressed in uninfected plants and is not further suppressed after inoculation with PH1, whereas it strongly accumulated in PH1-dcl1/2 vs.

PLOS ONE
PH1-infected Bd21-3 ( Fig 5). This finding further supports the possibility that the control of SERK2 expression via RNAi pathways by Fg is evolutionary conserved in cereals.

RLM-RACE shows infection specific degradation products of HvBAK1, HvEOL1 and HvSERK2
We assessed the sRNA-mediated cleavage of HvBAK1, HvEOL1, and HvSERK2 mRNAs, using a modified RNA-ligase-mediated Rapid Amplification of cDNA Ends (RLM-RACE) assay. Control samples were prepared both from uninfected tissue and from infected tissue without the reverse transcription step (no-RT control) and PCR products were visualized on an EtBr-Agarose gel. In these no-RT controls no amplification was visible.
For each gene more than one infection-specific product was amplified (blue and red arrows), which could not be amplified from the uninfected sample (Fig 6D-6F). We excised three bands (red arrows) of the expected size for a Fg-sRNA-1921 guided cleavage of HvBAK1 ( Fig 6D) and one band for HvEOL1 ( Fig 6E) and HvSERK2 ( Fig 6F) and cloned them into the pGEM-T easy vector system. According to the IBSC_PGSB_v2 assembly, HvBAK1 has splice variants, which could produce cleavage products of different lengths while for HvSERK2 and HvEOL1 there are no introns between sRNA target site and primer. From each band, five colonies were picked and for 23 of these extracted plasmids sequences were obtained. 16 sequences perfectly matched the reference genome, four with one MM and one with four MMs. Two sequences did not match the reference sufficiently enough to be aligned over the full length. The observed cleavage products are close to but do not match the canonical slice site between the 10 th and 11 th nt of Fg-sRNA-1921 and Fg-sRNA-321 (Fig 6A-6C).

Total sRNAs predicted to target a gene in barley are correlated with the derepression strength
Not all potential targets of Fg-sRNAs are downregulated nor do all potential targets show a reaccumulation upon infection with PH1-dcl1/2 (see Fig 3). To address this bias we conducted a more focused target prediction exclusively for the 16 genes already tested by RT-qPCR. This allowed a much more thorough search, where targets for all sRNAs with at least two reads were predicted. From these 136,825 unique sRNAs (axenic, 21-24 nt length, > = 2 reads) representing 4,997,312 reads of the total of 5,439,472 reads 21-24 nt in length, 5,052 have potential target sequences in the 16 mRNA sequences selected for further investigation in the Hordeum vulgare cv. GP assembly GCA_902500625. An additional filter step was employed to select for sRNAs with a maximum of one MM to the PH1 assemblies GCA_000240135.3 and GCA_900044135.1. Subsequently, sRNAs with up to one MM to Fg-rRNAs were removed leaving a total of 1,212 sRNAs with 1,311 potential sRNA-mRNA interactions representing 85,531 reads in the analysis.
To establish a correlation of the observed resurgence of potential target genes and targeting sRNAs, we analysed the DCL-dependent expression change using ΔΔΔct values. To compare the expression of a GOI in two samples, the difference between the ct-values for a reference gene and the GOI can be determined (Δct) and to calculate the expression difference between the control and treated sample the difference between the Δct values (ΔΔct) is calculated. We further defined the ΔΔΔct value as the difference between the ΔΔct values for a GOI in PH1 and PH1-dcl1/2-infected samples. From this follows a gene with a negative ΔΔΔct value shows a higher transcript accumulation during the infection with a fungal strain with compromised DCL function and the stronger the accumulation the lower this ΔΔΔct value is. We found a negative correlation between the ΔΔΔct value and the number of total sRNAs targeting a GOI (Fig 7). This correlation becomes more significant if a lower score cut-off for the target prediction is chosen until the cut-off of four. The most significant correlation is for all predicted interactions with a score equal or below four with a p-value of 0.011 (t-test) (Fig 7B). The pvalue for a correlation with a cut-off of five (Fig 7C) is 0.033 (t-test) and six (Fig 7D) is 0.094 (t-test), while a score cut-off of 3 leads to a situation, where there are no predicted sRNA interactions for all genes except for three (Fig 7A).

Discussion
We show here that full virulence of the ascomycete fungus Fusarium graminearum on graminaceous leaves depends on the activity of fungal DCLs. The dKO mutant PH1-dcl1/2 is less virulent on barley and the two single KO mutants IFA65-dcl1 and IFA65-dcl2 also are less virulent on Brachypodium. These results are consistent with our previous studies showing that knock-down or SIGS-mediated silencing of Fusarium DCLs and other components of the RNAi machinery reduced the virulence of the fungus on barley [8,44]. DCL enzymes are key components of the fungal RNAi machinery required for the biogenesis of sRNAs directing silencing of sequence-complementary endogenous and foreign genes [45]. The latter case involves DCL-dependent pathogen-derived sRNAs that target plant defense genes to increase virulence as shown for Botrytis cinerea [19,21], Puccinia striiformis [23] and Magnaporthe oryzae [46].
In the present work we found potential host target genes for fungal small RNAs (Fg-sRNAs) that were differentially regulated in response to plant infection with Fg wt vs. Fg DCL KO mutants, and the same effect was confirmed when DCLs were silenced by SIGS. This suggests a scenario in which impaired DCL function resulting in reduced fungal RNAi activity ultimately leads to de-repression of host target genes. Of note, target gene de-repression was also observed when the transcript was not significantly downregulated by the wt fungus during infection. This could be explained by a mutually neutralizing effect in which Fg-sRNAs continuously target genes for silencing, while concurrent plant immune responses are a trigger for up-regulation. Thus, one can speculate that these described effects reflect an abrogation of host-favouring upregulation by host immunity vs. pathogen-favouring downregulation by sRNA effectors.
We identified three tRFs predicted to target BdSERK2, HvBAK1, HvEOL1 and HvSERK2. Unexpectedly, these tRFs are partially DCL-dependent, with a reduced abundance by more than 50% during infections with the dKO mutant PH1-dcl1/2 vs. wt PH1 based on fungal biomass. Current knowledge of tRFs in fungi and oomycetes suggests that their silencing activity is independent of DCL, as shown for Sclerotinia sclerotiorum [47] and Phytophthora infestans, where the production is partially dependent on AGO [48]. Furthermore, analysis of tRFs in Cryptococcus spp. revealed a RNAi-independent generation of tRFs and possible compensatory effects in an RNAi-deficient genotype [49]. Interestingly however, the tRFs Fg-sRNA-321, Fg-sRNA-1921 and Fg-sRNA-6717 are neither 5'-or 3' tRNA halves nor do they belong to any of the described tRF-1, tRF-2, tRF-4 or tRF-5 classes [50] applied by the tRFtarget database for animals, yeast (Schizosaccharomyces pombe) and the bacterium Rhodobacter sphaeroides [51]. When following the classification of the tsRBase used for all eukaryotic kingdoms and bacteria [52], the three tRFs are classified as internal tRFs based on the origin within the mature tRNA. Interestingly, there are tRFs found in Phytophthora sojae starting in the anticodon loop and ending in the T loop of mature tRNAs [53], which resembles the Fg-sRNA tRFs (S8 Fig). We observed several infection-specific degradation products of the predicted host target genes HvBAK1, HvEOL1 and HvSERK2 for tRFs Fg-sRNA-321, Fg-sRNA-1921 and Fg-sRNA-6717. However, cleavage occurred outside the canonical miRNA cleavage site as defined by Mallory et al. [43], though these genes are partially silenced during infection and silencing is apparently abolished upon infection with the DCL dKO mutant. While the canonical cleavage site for miRNA-directed cleavage in At is well defined, the tRF-directed cleavage observed by 5' RACE of transposable elements in At [54] and of defence-related genes during the infection of black pepper (Piper nigrum) with Phytophthora capsici [55] was found outside of the canonical cleavage site. Additionally, the identification of sRNA-directed cleavage sites in barley often leads to divergent findings. Ferdous et al. [56] predicted~400 target genes for 11 presumably drought responsive miRNAs and found cleavage products for 15 targets overlapping the respective miRNAs alignment through degradome sequencing in the two barley cultivars Golden Promise (GP) and Pallas. From these confirmed targets, 13 were cleaved at the canonical 10 th -11 th nt site, one was cleaved at 19 th -20 th nt, and one at the 5 th -6 th nt. Hackenberg et al. [57] predicted 97 target genes of drought responsive miRNAs in GP and identified eight targets through degradome sequencing, which were all cleaved outside of the 10 th -11 th nt site. Thus, both studies suggest the presence of non-canonical miRNA directed cleavage. Of note, both studies relied on the same degradome sequencing dataset from GP, while Ferdous et al. also observed non-canonical cleavage in an independent Pallas dataset. Moreover, in a study performed by Curaba et al. [58] 96 target genes of GP for miRNAs involved in seed development and germination were identified by degradome sequencing and only 16 targets were cleaved exclusively at the 10 th -11 th nt site, while the other targets were sporadically cleaved with an offset (24) and 56 were cleaved in majority in a non-canonical site. Finally, Deng et al. [59] identify in the barley cultivar Morex 65 target genes of 39 miRNAs, and for only 32% of the identified targets the canonical 10 th -11 th nt cleavage product was the major degradome product. Together these studies highlight the challenges in the identification of cleavage sites of sRNAs in barley and cleavage sites of tRFs in plants. The absence of canonical cleavage products for tRFs does therefore not exclude the tRF-directed cleavage of HvBAK1, HvEOL1 and HvSERK2.
We found that 22 Fg-sRNAs target HvEOL1, a putative negative regulator of ET biosynthesis in barley. In Arabidopsis thaliana the EOL1 homolog AtETO1 acts together with AtEOL1 and AtETO1-like 2 (EOL2) in directing the ubiquitination and subsequent degradation of type-2 1-aminocyclopropane-1-carboxylate synthase (ACS) proteins (e.g. ET overproducer 2 (ETO2)) [40,60]. ET is a gaseous plant hormone that plays an important role in regulating plant growth and development, and is critical for pathogen interaction and abiotic stresses [61]. Generally, ET acts synergistically with jasmonate (JA) in the defence response against necrotrophic pathogens and this ET/JA response has antagonistic effects on salicylic acid (SA) signalling against biotrophic pathogens. Yet in low amounts JA and SA act synergistically [62,63]. Therefore, controlling both ET biosynthesis and ET signalling is crucial for plants. Towards this, plants have evolved complex mechanisms that allow tight regulation of ET pathways e.g. at the level of (i) ET production mainly by regulating ACS gene family members, (ii) ET perception through constitutive triple response 1 (CTR1)-mediated inhibition of positive regulator ET insensitive 2 (EIN2) [64,65], and (iii) expression of ET-responsive TFs (e.g. ET response factor 1 (ERF1)) via EBF-mediated degradation of ET insensitive 3 (EIN3) [66] (S4 Fig). According to the anticipated role of ET in the plant response to necrotrophic pathogens, such as Fg, targeting negative regulators of ET synthesis such as HvEOL1 would be detrimental to Fg colonization. Of note, our findings are consistent with previous results demonstrating that Fg exploits ET signalling to enhance colonization of Arabidopsis, wheat and barley [67], supposedly through an increase in DON-induced cell death through ET. These findings further challenge the role of ET in defence against necrotrophic pathogens. Strikingly, the authors showed that in Arabidopsis ET overproducing mutants (ETO1 and ETO2) and a negative regulator of ET signalling (CTR1) are more susceptible to Fg, while At mutants in ET perception (ETR1) and signalling (EIN2 and EIN3) are resistant. These findings were confirmed by the direct application of ET during the infection of wheat and barley, which lead to increased susceptibility to Fg. Based on these findings, we suggest that negative regulators of ET are efficient targets for sRNA-directed manipulation of host immunity by Fg.
The bacterial pathogen Pseudomonas syringae secretes two effector molecules, AvrPto and AvrPtoB, into host plants. These effectors interact with the receptor-like kinase BRI1-associated receptor kinase 1 (BAK1), also known as SERK3, thereby preventing the recognition of various MAMPs through the association of BAK1 with pattern recognition receptors (PRRs) such as flagellin-sensitive 2 (FLS2) and Ef-Tu receptor (EFR) [68]. We observed FgDCLdependent silencing of the cereal BAK1 homologs HvBAK1, HvSERK2 and BdSERK2. While these genes have a higher similarity to AtSERK2 than to AtSERK3 (AtBAK1), they still are among the closest homologs to AtBAK1 found in cereals (S9 Fig). It is tempting to speculate that further experiments will uncover additional hubs that are targeted both by protein and sRNA effectors.

Conclusion
Our data show that in the necrotrophic ascomycete Fusarium graminearum gene silencing by RNAi shapes its ability to cause disease, which is consistent with earlier results on the significance of the RNAi machinery in Fg [8,35]. Pathogenicity relies on DICER-like (DCL)dependent sRNAs that were identified as potential candidates for fungal effectors targeting defence genes in two Poaceae hosts, barley and Brachypodium. We identified Fg-sRNAs with sequence homology to host genes that were down-regulated by Fg during plant colonisation, while they were expressed above their level in healthy plants after infection with a DCL dKO mutant. In PH1-dcl1/2 vs. PH1 the strength of target gene accumulation correlated with the abundance of the corresponding Fg-sRNA. Our data hint to the possibility that three DCLdependent tRFs with sequence homology to immunity-related Ethylene overproducer 1-like 1 (HvEOL1) and three Poaceae orthologues of Arabidopsis thaliana BRI1-associated receptor kinase 1 (HvBAK1, HvSERK2 and BdSERK2) contribute to fungal virulence via targeted gene silencing.
For infection of barley (Hordeum vulgare cv. Golden Promise, GP) and Brachypodium distachyon (Bd21-3), plants were grown in a 16 h photoperiod at 20˚C/18˚C day/night and 60% relative humidity in soil (Fruhstorfer Type LD80, Hawita). Ten detached second leaves were transferred into square Petri plates containing 1% water-agar. GP leaves were drop-inoculated with 3 μl of 1.5 x 10 5 conidia ml -1 conidia suspension. Bd21-3 leaves were drop-inoculated on two spots with 10 μl of 1 x 10 4 conidia ml -1 conidia suspension. Infection strength was measured with the PlantCV v2 software package (https://plantcv.danforthcenter.org/) (Hv) by training a machine learning algorithm to recognize necrotic lesions or by ImageJ (Bd). For gene expression analysis, a suspension of 5 x 10 4 Fg conidia ml -1 was used and leaves were either inoculated on 3 spots with 20 μl (Hv) or on 2 spots with 10 μl (Bd), respectively and experiments were evaluated 5 dpi or 4 dpi for strain PH1 on Bd.

Fungal transcript analysis
Gene expression analysis was performed using reverse transcription quantitative PCR (RT-qPCR). RNA extraction was performed with GENEzol reagent (Geneaid) following the manufacturer's instructions. DNA was digested with DNAse I (Thermo Scientific) according to manufacturer protocol and remaining RNA was used for cDNA synthesis using qScript™ cDNA kit (Quantabio). For RT-qPCR, 10 ng of cDNA was used as template in the QuantStudio 5 Real-Time PCR system (Applied Biosystems). Amplifications were performed with 5 μl of SYBR 1 green JumpStart Taq ReadyMix (Sigma-Aldrich) with 5 pmol oligonucleotides. Each sample had three technical repetitions. After an initial activation step at 95˚C for 5 min, 40 cycles (95˚C for 30 sec, 60˚C for 30 sec, 72˚C for 30 sec) were performed followed by a melt curve analysis (60˚C-95˚C, 0.075˚C/s). Ct values were determined with the QuantStudio design and analysis software supplied with the instrument. Transcript levels were determined via the 2 -ΔΔCt method [71] by normalizing the amount of target transcript to the amount of the reference transcript Elongation factor 1-alpha (EF1-a, FGSG_08811) gene (S2 Table).

Spray application of dsRNA
Second leaves of 2 to 3-week-old GP were detached and transferred to square Petri plates containing 1% water agar. dsRNA was diluted in 500 μl water to a final concentration of 20 ng μl -1 . As control, Tris-EDTA (TE) buffer was diluted in 500 μl water corresponding to the amount used for dilution of the dsRNA. Typical RNA concentration after elution was 500 ng μl -1 , with 400 μM Tris-HCL and 40 μM EDTA in the final dilution. Each plate containing 10 detached leaves was evenly sprayed with either dsRNAs or TE buffer with 500 μl, and subsequently kept at room temperature [10]. Two days after spraying, leaves were drop-inoculated with three 20 μl drops of Fg suspension containing 5 × 10 4 conidia ml −1 . After inoculation, plates were closed and incubated for five days at room temperature.

Target prediction for sRNAs
RNA was purified and enriched for sRNAs from fungal axenic culture (PRJNA749737) using the mirVana miRNA Isolation Kit (Life Technologies). Indexed sRNA libraries were constructed from these sRNA fractions with the NEBNext Multiplex Small RNA Library Prep Set for Illumina (New England Biolabs) according to the manufacturer's instructions. Reads were trimmed with the cutadapt tool v2.1 [74] by removing adapters and retaining reads with a length of 21-24 nt and quality checked with the fastQC tool v0.11.9 (http://www. bioinformatics.babraham.ac.uk/projects/fastqc/). For S1 Fig reads were aligned to the Fg reference genome (GCF_000240135.3_ASM24013v3) with bowtie2 [75] following a sensitive alignment policy (-D 100, -R 10, -L 19). The aligned reads were assigned to the additional attribute "gene_biotype" with htseq-count [76] according to the latest assembly (ftp://ftp. ensemblgenomes.org/pub/release-44/fungi/gff3/fungi_ascomycota3_collection/fusarium_ graminearum_gca_000240135). Remaining reads were collapsed with the fastx toolkit v0.0.14 [77] and reads with at least 400 reads were targeted against the IBSC_PGSB_v2 cDNA annotation with the plant miRNA target prediction algorithm TAPIR [42], following the optimized parameters according to Srivastava et al. [39]. The results of the target prediction were further analysed with RStudio [78] and the package biomaRt [79] to find targets associated with stress and immunity associated Gene ontology (GO) terms in the database "plants_mart" from plants.ensembl.org hosted by the EBI (European Bioinformatics Institute) and the Wellcome Trust Sanger Institute. The same method was used for the identification of target genes in B. distachyon (GCA_000005505.4) and A. thaliana (Araport11).

Stemloop-RT-qPCR of sRNAs
RNA was extracted and genomic DNA was digested as described for the transcript analyses. The sequences of sRNAs found in axenic fungal culture were used to design specific stem loop (SL) primers matching the sRNA over 6 nt at the 3'end. For the primer design, the tool of Adhikari et al. [80] was used. SL-primers were diluted to 10 pM and folded in a cycler (95˚C for 15 min, 90˚C 5 min, 85˚C 5 min, 80˚C 5 min, 75˚C 1 h, 68˚C 1 h, 65˚C 1 h, 62˚C 1 h, 60˚C 3 h). These primers were used for cDNA synthesis (Thermo Scientific RevertAid RT Reverse Transcription Kit) according to manufacturer's instruction with an annealing step at 16˚C instead of 25˚C and were used in multiplex to target respective fungal sRNAs and barley miR-NAs Hvu-mir159 and Hvu-mir168 as references. To obtain amplification efficiencies, a mix from all RNAs was diluted in a four step dilution series with a factor of ten and reverse transcribed. Reactions were set up with the highest concentration of 15 ng μl -1 and the lowest of 15 pg μl -1 cDNA. All sRNA amplifications showed an efficiency of 80-82% and an R 2 between 1 and 0.997 except for Fg-sRNA-6717 with an efficiency of 66.4%. For RT-qPCR, 1.5 μl of 3 ng μl -1 cDNA was used as template in the QuantStudio 5 Real-Time PCR system (Applied Biosystems). Amplifications were performed with 5 μl of SYBR 1 green JumpStart Taq ReadyMix (Sigma-Aldrich) with 1.5 pmol or 3 pmol oligonucleotides. Each sample had three technical repetitions. As forward primer the unused nucleotides of the remaining sequence of the sRNA were used, which were extended to achieve optimal melting temperature, and as reverse primer the universal stem loop primer developed by Chen et al. [41] was used. Relative abundance of the sRNAs was calculated with the ΔΔct-method with incorporation of amplification efficiencies. sRNAs were normalized against the reference miRNAs Hvu-mir-159a and 168-5p and after this against the fungal biomass measured as EF1-α against HvUBQ (HORVU1Hr1G023660).

Statistics
To assess the differential expression of genes via RT-qPCRs the Δct values were compared via a one or two sided paired Students t-test. Disease symptoms were either compared via Students t-test if the data showed a normal distribution in Shapiro-Wilk test or via a Wilcoxon rank sum test.

RLM-RACE
RNA from GP barley infected with Fg-IFA65 at 5 dpi and an uninfected control was extracted with the Isolate II plant miRNA kit (Bioline). 1 μg of RNA (>200 nt) of infected, uninfected and a mix of both samples for a-RT-control were assembled. 1 μl of the 5'RACE Adapter [0.3 μg/μl], 1 μl of the 10x Reaction Buffer, 1 μl of 1mg/μl BSA, 0.5 μl of T4 RNA Ligase [10U/ μl] (Thermo Scientific) and DEPC-treated water up to 10 μl were prepared and incubated at 37˚C for 60 min. Subsequently, the whole reaction was used for reverse transcription (Rever-tAid Reverse Transcriptase, Thermo Scientific). 10 μl ligation reaction, 1 μl Random Hexamer [100pmol/μl], 4 μl 5x Reaction Buffer, 0.5 μl RiboLock RNase Inhibitor (Thermo Scientific), 2 μl dNTP Mix [10 mM] and 1 μl RevertAid Reverse Transcriptase (or water (-RT control)) and 1.5 μl water were mixed and run for 10 min at 25˚C, 60 min at 42˚C and 10 min at 70˚C. Then, a nested hot-start touch-down PCR for each target gene was performed. The primer sequences for the outer (first) and inner (second) PCR are shown in S1 Table.

Analysis of target genes and targeting sRNAs
After the initial target prediction an additional target prediction for the newly released cultivar specific genome (GCA_902500625) of barley cv. Golden Promise (GP) was conducted. Adapters were removed and reads were collapsed as described before for the target prediction. All sRNA sequences were read with SeqinR v3.6-1 [81] and stored in a list of SeqFastadna objects. To identify the homologous genes to the already identified targets in GP, the cDNA library was blasted with the command-line blast application (Nucleotide-Nucleotide BLAST 2.6.0+) [82] against the identified target sequences from the IBSC_PGSB_v2 cDNA library with percent identity of 90 and a query coverage of 55% as cut-off values. All sRNAs with at least two reads were written to a file in chunks of 2000 each and ran against each individual target gene with TAPIR via the system2 function in R [83] in the RStudio software. Results were collected, stored in a data.frame, and further analysed with R. sRNAs identified to target a gene of interest (GOI) were written to a fasta file with SeqinR and blasted against the rRNAs from the assemblies GCA_900044135.1 (Fg-PH1), GCA_000240135.3 (Fg-PH1) and the Fusarium rRNAs from the RNAcentral fungal ncRNA dataset (ftp://ftp.ebi.ac.uk/pub/databases/ RNAcentral/current_release/sequences/by-database/ensembl_fungi.fasta (12/Sep/2020)) with the options wordsize = 4, perc_identity = 95, qcov_hsp_perc = 95. All sRNAs matching rRNAs were removed. Thereafter, sRNAs were compared to the Fg assemblies GCA_900044135.1 (Fg-PH1) and GCA_000240135.3 (Fg-PH1) with the same blast strategy and only perfectly matching sRNAs were retained.
To derive the relative expression of a GOI between two samples the following formula is used.
We further defined the ΔΔΔct value as the difference between the ΔΔct values for a GOI in PH1 and PH1-dcl1/2-infected samples.
This enables the calculation of the re-accumulation between the two samples as follows.
The sum of all reads and the corresponding ΔΔΔct-value were plotted with ggplot2 [84] and a linear regression was added to the plot. To allow a log2-transformation of the plots genes with zero targeting reads were set to one targeting read. The plots were arranged using ggpubr v.0.4.0 [85].

GO enrichment analysis
Gene ontology (GO) enrichment analysis was performed via the AgriGO v.2.0 analysis toolkit [86] with the standard parameters singular enrichment analysis (SEA).