Role of Dicer-Dependent RNA Interference in Regulating Mycoparasitic Interactions

ABSTRACT Dicer-like proteins (DCLs) play a vital role in RNA interference (RNAi), by cleaving RNA filament into small RNAs. Although DCL-mediated RNAi can regulate interspecific communication between pathogenic/mutualistic organisms and their hosts, its role in mycoparasitic interactions is yet to be investigated. In this study, we deleted dcl genes in the mycoparasitic fungus Clonostachys rosea and characterize the functions of DCL-dependent RNAi in mycoparasitism. Deletion of dcl2 resulted in a mutant with reduced secondary metabolite production, antagonism toward the plant-pathogenic fungus Botrytis cinerea, and reduced ability to control Fusarium foot rot disease on wheat, caused by Fusarium graminearum. Transcriptome sequencing of the in vitro interaction between the C. rosea Δdcl2 strain and B. cinerea or F. graminearum identified the downregulation of genes coding for transcription factors, membrane transporters, hydrolytic enzymes, and secondary metabolites biosynthesis enzymes putatively involved in antagonistic interactions, in comparison with the C. rosea wild-type interaction. A total of 61 putative novel microRNA-like RNAs (milRNAs) were identified in C. rosea, and 11 were downregulated in the Δdcl2 mutant. In addition to putative endogenous gene targets, these milRNAs were predicted to target B. cinerea and F. graminearum virulence factor genes, which showed an increased expression during interaction with the Δdcl2 mutant incapable of producing the targeting milRNAs. In summary, this study constitutes the first step in elucidating the role of RNAi in mycoparasitic interactions, with important implications for biological control of plant diseases, and poses the base for future studies focusing on the role of cross-species RNAi regulating mycoparasitic interactions. IMPORTANCE Small RNAs mediated RNA interference (RNAi) known to regulate several biological processes. Dicer-like endoribonucleases (DCLs) play a vital role in the RNAi pathway by generating sRNAs. In this study, we investigated a role of DCL-mediated RNAi in interference interactions between mycoparasitic fungus Clonostachys rosea and the two fungal pathogens Botrytis cinerea and Fusarium graminearum (here called mycohosts). We found that the dcl mutants were not able to produce 11 sRNAs predicted to finetune the regulatory network of genes known to be involved in production of hydrolytic enzymes, antifungal compounds, and membrane transporters needed for antagonistic action of C. rosea. We also found C. rosea sRNAs putatively targeting known virulence factors in the mycohosts, indicating RNAi-mediated cross-species communication. Our study expanded the understanding of underlying mechanisms of cross-species communication during interference interactions and poses a base for future works studying the role of DCL-based cross-species RNAi in fungal interactions.

S mall RNAs (sRNAs) are a group of noncoding RNAs. They play a central role in gene silencing at the transcriptional level through chromatin modification and at the posttranscriptional level through targeted destruction of mRNAs, also known as RNA interference (RNAi) (1)(2)(3)(4)(5). Dicer-like protein (DCL) plays central role in RNAi by cleaving the double-stranded RNA precursors and single-stranded RNA precursors with hairpin structures to generate sRNAs, often ranging in size from 18 to 40 nucleotides, called small-interfering RNAs (siRNAs) and microRNAs (miRNAs; microRNA-like RNAs [milRNAs] in fungi), respectively. In fungi, the most studied RNAi pathways are mediated by siRNAs and milRNAs and are dependent on DCLs for biogenesis and are thus called Dicer-dependent RNAi. Dicer-independent RNAi, such as that mediated by dicer-independent small interfering RNAs (disiRNAs), has also been identified in the filamentous fungus Neurospora crassa (6).
Small-RNA mediated RNAi is an evolutionarily conserved process of self-defense triggered by a wide variety of exogenous nucleic acids such as invading viruses, transgenes, transposons, and plasmids (7,8). In fungi, a role of sRNA-mediated RNAi pathways in genome defense against the insertion of repetitive transgenes during vegetative growth (quelling) and the sexual phase of the life cycle (meiotic silencing of unpaired DNA [MSUD]) was first reported in N. crassa (9)(10)(11). Since then, RNAi pathways and their role in genome defense against retrotransposon activity have been demonstrated in several fungal species with diverse lifestyles (8,(12)(13)(14)(15)(16)(17)(18)(19)(20). However, in some fungal species, such as Saccharomyces cerevisiae and Ustilago maydis, genes related to the RNAi pathways are absent (21,22). In addition to the role of genome defense against transgenes, the fungal RNAi machinery generates a variety of sRNAs that are involved in the regulation of numerous biological processes through targeted gene silencing (8,23). For instance, sRNAs (mainly milRNAs) are found to be differentially expressed in fungi during different growth phases, developmental stages, and environmental conditions, including those involved in host-pathogen interactions (24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34). Furthermore, sRNAs can move bidirectionally between the species and modulate cellular functions of recipient cells by hijacking their RNAi machinery. Thus, they play an important role in interspecies communication between closely interacting symbiotic organisms, including parasitic and mutualistic interactions (35)(36)(37)(38)(39)(40). However, the role of sRNAs in parasitic fungus-fungus interactions is yet to be investigated.
The filamentous fungus Clonostachys rosea is a ubiquitous soilborne ascomycete with a complex lifestyle as a necrotrophic mycoparasite and saprotroph (41). C. rosea efficiently overgrows and kills its mycohosts such as Botrytis cinerea and Fusarium graminearum (41)(42)(43). During mycoparasitic interactions or exposure to the secreted factors from mycohosts, C. rosea induces expression of genes associated with the production of secondary metabolites, hydrolytic enzymes, and other secreted proteins (43)(44)(45)(46)(47)(48)(49)(50). Furthermore, C. rosea induces expression of genes coding for membrane transporters to efflux the endogenous toxic compounds and exogenous metabolites that may come from interacting organisms during the interspecific interactions (49,51,52). The role of secreted proteins/enzymes, secondary metabolites, and membrane transporters in antibiosis and mycoparasitism in C. rosea is proven (42-44, 50, 53, 54); however, the role of RNAi in regulating the cellular regulatory network during such interactions has not yet been investigated.
The present work aims to (i) characterize the RNAi machinery in C. rosea; (ii) identify milRNAs that are key regulators of genes associated with the antagonistic/mycoparasitic activity in C. rosea, as well as their potential endogenous and cross-species gene targets; and (iii) investigate common or species-specific responses in sRNA-mediated gene regulation in C. rosea against mycohosts. We used the two important plant-pathogenic fungi B. cinerea and F. graminearum as different mycohosts, since they are taxonomically different from each other and represent different disease types on different crops. We hypothesized that (i) sRNAs regulate mycoparasitic interactions in C. rosea at endogenous and cross-species level and that (ii) C. rosea responds with both common and mycohost-specific reactions toward B. cinerea and F. graminearum. To test these hypotheses, we generated gene deletion and complementation strains of genes An in vitro dual culture assay was used to test whether deletion of dcl1 or dcl2 affected the antagonistic ability of C. rosea. No differences in growth rate of F. graminearum or B. cinerea were recorded during in vitro dual plate confrontation with either of the dcl deletion strains, compared to the WT (see Fig. S3A). However, a reduced ability FIG 1 Phenotypic characterizations of C. rosea WT, deletion, and complementation strains. (A) Growth rate of WT, dcl deletion, and complemented strains. Strains were inoculated on PDA medium and incubated at 25°C, and the growth rate was recorded 5 days postinoculation (dpi). Error bars represent standard deviations based on four biological replicates. (B) Conidiation of WT, dcl deletion, and complementation strains on PDA medium 24 dpi. Conidia were harvested in equal volumes of water and were counted using a Bright-Line Haemocytometer according to the instructions of manufacturer. Error bars represent standard deviations based on four biological replicates. (C) Deletion of dcl2 affects pigment production in C. rosea. Strains were inoculated on PDA medium and incubated at 25°C. The experiment was performed in four biological replicates, and photographs of representative plates were taken 16 dpi. (D) Dual culture assay to test antagonistic ability of C. rosea WT, deletion, and complementation strains against B. cinerea. Agar plugs of C. rosea strains (left side in the plate) and B. cinerea (right side in the plate) were inoculated on opposite sides in 9-cmdiameter agar plates, followed by incubation at 25°C. The growth rates (overgrowth) of C. rosea WT, deletion, and complementation strains on B. cinerea were measured from the point of mycelial contact. The experiment was performed in four replicates, and photographs of representative plates were taken 21 dpi of C. rosea strains. An arrowhead indicates the mycelial front of C. rosea strains. (E) In vivo assay to test the biocontrol ability of C. rosea strains against F. graminearum foot rot disease on wheat. Seeds were coated with C. rosea conidia and planted in moist sand together with a F. graminearum agar plug. Seedlings were harvested 21 dpi, and disease symptoms were scored on a scale from 0 to 4. The experiment was performed in five biological replicates with 15 plants in each replicate. Different letters indicate statistically significant differences based on Tukey HSD method at the 95% significance level.
(P , 0.001) to overgrow B. cinerea was observed in Ddcl2 strains compared to the WT (Fig. 1D). The growth rate of Ddcl2 strains displayed 33% reduction on B. cinerea mycelium (overgrowth rate) compared to the growth rate of WT (Fig. 1D). In contrast, overgrowth of F. graminearum was not compromised in either of the deletion strains (see Fig.  S3A). However, a change in F. graminearum color (pigment) was visible at the bottom side of the Ddcl2 mutant-F. graminearum interaction zone (see Fig. S3A). In contrast to in vitro antagonism tests, a bioassay for biocontrol of fusarium foot rot diseases on wheat caused by F. graminearum displayed a significant 56% increase (P = 0.023) of disease severity in wheat seedlings previously seed coated with the Ddcl2 strain compared to seedlings from seeds coated with C. rosea WT (Fig. 1E). However, disease symptoms on seedlings from seeds coated with Ddcl1 strains showed no significant difference compared to the WT.
Analysis of metabolites. The metabolites produced by the WT, dcl deletion, and complementation strains were analyzed by ultrahigh-performance liquid chromatography/mass spectrometry (UHPLC-MS) and UHPLC-tandem MS (UHPLC-MS/MS) (see Table  S2). When analyzing the UHPLC-MS data by principal-component analysis (PCA), the samples from the Ddcl1, Ddcl11, and WT strains grouped separated from each other ( Fig. 2A, left) and, likewise, Ddcl2 and WT samples clustered separately ( Fig. 2A, right). The Ddcl21 samples, however, clustered with the WT samples, indicating restoration of metabolite production in Ddcl21 strains. Two compounds were present in significantly smaller amounts in the Ddcl1 strain, and their production was restored in Ddcl11 strains, along with 15 further compounds (analysis of variance [ANOVA], false discovery rate [FDR] # 0.01; see Fig. S3B and Table S2). Fifty-four metabolites were present in significantly smaller amounts in the Ddcl2 strain compared to the WT; at the same time, their production was restored in the Ddcl21 strain (ANOVA, FDR # 0.01; see Fig. S3B and Table S2). Seventeen of these compounds were tentatively identified or assigned to a compound class by UHPLC-MS, UHPLC-MS/MS, and database mining ( Fig. 2B; see also Fig. S3C). Most of these substances were monomeric or dimeric hexaketides of the sorbicillin type (e.g., sorbicillin, sorbicillinol, oxosorbicillinol, epoxysorbicillinol, and bisvertinolone), whereas three glisoprenins (I, III, and IV) also were identified. The identification of some of these compounds is outlined below.
Sorbicillin was tentatively identified as a compound eluting at 114.7 s with [M1H] 1 m/z 233.118, with two major fragment ions, m/z 95.049 and m/z 165.054, corresponding to bond cleavage on either side of the side chain carbonyl (see Fig. S3C). The ion at m/z 95.049 was diagnostic for all monomeric and dimeric sorbicillin-type compounds containing a hexa-2,4-diene-1-one motif. Fragment ions corresponding to the ion with m/z 165.054 discussed above were important for all monomeric sorbicillin type compounds, and related fragment ions were frequently found with additional loss of CO and/or water, depending on the respective compound structure. The compound eluting at 71.1 s, with [M1H] 1 m/z 249.113, was tentatively identified as sorbicillinol based on such fragment ions (see Fig. S3C), and the two compounds at 58.0 s and 94.5 s, both with [M1H] 1 m/z 265.207, were suggested to be oxosorbicillinol and epoxysorbicillinol, respectively, based on differences in fragment ions (see Fig. S3C). Five compounds in Fig. 2B gave m/z values which, after database mining, suggested that they were vertinolide or hydroxyvertinolide, hexaketides similar to the sorbicillins but with a lactone head-group instead of the aromatic ring or unsaturated cyclohexanone of sorbicillin-type compounds. In MS/MS, however, the vertinolide-type compounds did not yield fragment ions supporting their structures. Instead, MS/MS data suggested that these compounds were novel dihydrosorbicillinols or oxo/epoxy-dihydrosorbicillinol, respectively.
A large number of dimeric compounds of the sorbicillin-type are known (56), and several share the same molecular formula. These substances are dimerized by several different biosynthetic mechanisms, including Diels-Alder cycloaddition, Michael-type addition reactions, and formation of hemi-ketals. The compound eluting at 129.0 s, with [M1H] 1 m/z 513.212 (in accordance with the compound bisvertinolone) gave two  1 . The peak areas shown are average peak areas Â 10 23 with standard deviations in brackets. The heatmap is based on sum-normalized and 10-logaritmized peak areas. Labels in panel A: A, may also be dihydroepoxysorbicillinol; B, proposed to be four different isomers of dihydrosorbicillinol; C, has the same m/z as sorbicillinol but different MS/MS data; D, may also be bisvertinoquinol or isobisvertinol; E, may also be bislongiquinolide or bisorbicillinolide or trichodimerol or trichotetronine; and F, may also be isodihydrobisvertinol.
major fragment ions at m/z 249.111 and m/z 265.107, both [M1H] 1 , corresponding to the constituting monomeric compounds of bisvertinolone, i.e., sorbicillinol and oxosorbicillinol, respectively (see Fig. S3C). This pattern was observed for all putative dimeric sorbicillin-type compounds, i.e., in UHPLC-MS/MS analyses, these compounds fragmented to yield ions of the presumed constituting monomeric compounds, and related ions after loss of CO and/or water (see Fig. S3C). The formation of these fragment ions is possible for dimeric compounds formed by many different mechanisms, and therefore it was difficult to identify these by MS/MS without access to authentic reference compounds or very detailed information about the MS/MS behavior of these compounds. Therefore, several alternative identities are listed in Fig. 2B for some of the dimeric compounds. The polyhydroxy terpenes glisoprenin A, C, and D were identified based on the m/z of their respective [M1H] 1 ions, supported by the m/z of fragment ions (loss of multiple water molecules) detected in UHPLC-MS/MS.
Transcriptome analysis of Clonostachys rosea WT and dcl deletion strains. To gain insights into the molecular mechanisms associated with the altered phenotypes of C. rosea dcl deletion strains, transcriptomes of C. rosea WT, Ddcl1, and Ddcl2 were analyzed by RNA-seq during the interactions with B. cinerea and F. graminearum. An average of 20.5 million clean reads was obtained for each treatment. Since the sequences contained read pairs from both the interacting species, the reads originating from C. rosea or interacting mycohosts were identified by mapping to C. rosea, B. cinerea, or F. graminearum genomes. During the C. rosea-B. cinerea interaction, 24% of reads, on average, were mapped to C. rosea genes, while 58% of reads were assigned to C. rosea in the C. rosea-F. graminearum interaction. Summary data for transcriptome sequencing and mapping are presented in Table S3.
Compared to the C. rosea WT, the analysis identified 126 differentially expressed genes (DEGs; 106 upregulated and 20 downregulated) in the Ddcl1 strain against B. cinerea, while this number was much higher against F. graminearum, where 897 genes (504 upregulated and 393 downregulated) were differentially expressed (see Table S4). Among these, a majority of genes were uniquely expressed in the respective interaction, since only 32 and 3 genes were commonly upregulated and downregulated, respectively, against both the mycohosts (Fig. 3A). The deletion of dcl2 affected the expression pattern of a higher number of genes compared to the deletion of dcl1. In the Ddcl2 strain, in comparison to the WT, totals of 1,894 (251 upregulated and 1643 downregulated) and 1,706 (490 upregulated and 1216 downregulated) genes were differentially expressed against B. cinerea and F. graminearum, respectively (see Table S4). In contrast to the Ddcl1 strain, where a relatively lower proportion of genes (15.7% against B. cinerea; 43.7% against F. graminearum) were downregulated, a higher proportion (87% against B. cinerea,73% against F. graminearum) of DEGs in the Ddcl2 strain were downregulated. Among the upregulated genes in the Ddcl2 strain, 124 genes were commonly upregulated, while 118 genes and 365 genes, respectively, were uniquely upregulated against B. cinerea and F. graminearum. Among downregulated genes, 669 were common, while 973 and 538 genes, respectively, were unique against B. cinerea and F. graminearum (Fig. 3B).
The numbers of DEGs overlapping in Ddcl1 and Ddcl2 strains during the interactions with a common mycohost were determined ( Fig. 3C and D). Among genes that were upregulated in Ddcl1 or Ddcl2 strains against B. cinerea, 61 were common, while 45 (41%) and 190 (76%) were uniquely upregulated in Ddcl1 and Ddcl2 strains, respectively. However, the number of genes downregulated in both mutants against B. cinerea was 12. During contact with F. graminearum, similar numbers of genes were upregulated in the two mutants (246 in the Ddcl1 strain, 230 in the Ddcl2 strain, and 256 in both strains), while the numbers of downregulated genes were greater in the Ddcl2 strain (93 in the Ddcl1 strain, 918 in the Ddcl2 strain, and 296 in both strains) ( Fig. 3C and D).
GO enrichment analysis was performed to evaluate which processes were most affected in the dcl gene deletion mutants. Our results showed that a higher number of GO terms were significantly enriched in C. rosea genes under expressed in the Ddcl2 strain compared to the whole transcriptome. In the molecular function category, we found that terms such as catalytic activity (GO:0003824), hydrolase activity (GO:0016787), and oxidoreductase activity (GO:0016491) were commonly (against both the mycohosts) enriched (P # 0.05) among downregulated genes in the Ddcl2 strain, indicating a role of these genes in mycoparasitism-related functions in C. rosea (Fig. 3E). In contrast, other GO terms were only enriched against one of the two mycohosts. This was the case for the protein catabolism terms peptidase activity (GO:0008233) and proteolysis (GO:0006508), specifically enriched during the Ddcl2 mutant-B. cinerea interaction. Carbohydrate metabolism-related terms such as carbohydrate metabolism process (GO:0005975) and hydrolase activity acting on glycosyl bond (GO:0016798) were characteristic for the Ddcl2 mutant-F. graminearum interaction (Fig. 3E).
DCLs regulate genes with a predicted function during fungus-fungus interactions in Clonostachys rosea. Since the absence of DCL2 affected the production of secondary metabolites, antagonism, and biocontrol of C. rosea, we performed an in-depth analysis of genes with a reported function during interspecific interactions in C. rosea, including membrane transporters, enzymes involved in the biosynthesis of secondary metabolites, and hydrolytic enzymes. In addition, the expression pattern of genes coding for transcription factors and various components of the silencing machinery were analyzed. For each of these categories, there were more upregulated genes than downregulated ones in the Ddcl1 strain. An opposite pattern was evident in the Ddcl2 strain, where the number of upregulated genes in each category tended to be higher than that of downregulated ones, except for ABC transporters (Table 1; see also Table S5A).
(i) Membrane transporters. Deletion of dcl2 affected the expression of 161 major facilitator superfamily (MFS) transporters in C. rosea. Among these, 12 MFS transporters were upregulated, and 64 were downregulated during interaction with F. graminearum, whereas 6 were upregulated, and 99 were downregulated during interaction with B. cinerea (Table 1; see also Table S5A). Interestingly, 10 downregulated and 1 upregulated MFS transporters genes in the Ddcl2 strain showed high sequence similarity ($48% identity) with MFS transporters previously characterized for their involvement in efflux of secondary metabolites (polyketides, quinones, and polyketide/nonribosomal peptide hybrids) that are important for fungal virulence ( Table 2). These included apdF (aspyridones efflux protein in Colletothricum siamense), opS2 (quinone transporter in Aspergillus udagawae), atB (terreic acid efflux protein in F. oxysporum), FUB11 (fusaric acid efflux pump in Lachnellula suecica), FUBT (efflux pump involved in export of fusaric acid in F. culmorum), rdc3 (radicicol efflux pump in F. oxysporum), and aflT (aflatoxin efflux pump in Phialocephala subalpine) (57-60). Furthermore, a homolog of FUS6 (fusarin efflux pump FUS6 in Colletothricum fructicola) was upregulated. However, none of the corresponding gene clusters were present in the genome of C. rosea, suggesting that these MFS transporters constitute resistance proteins activated as a defense against harmful, hitherto-unknown, secondary metabolites. Moreover, 22 MFS transporter genes were previously reported to be induced in C. rosea during the interactions with B. cinerea and F. graminearum (49). Nine of these MFS transporter genes were significantly downregulated in the Ddcl2 strain during the interactions with B. cinerea or F. graminearum (Table 2). In summary the Ddcl2 mutant showed downregulation  of transporters with predicted function in secondary metabolite export and putative detoxification.
In contrast to the expression pattern of MFS transporters, a higher number of ATPbinding cassette (ABC) transporter genes was upregulated in both the deletion strains, but specifically against F. graminearum, where 14 and 10 genes, respectively, were upregulated in the Ddcl1 or Ddcl2 mutant (Table 1). Of 19 ABC transporters that were differentially regulated in Ddcl2, 5 upregulated and 1 downregulated belonged to the multidrug resistance protein (MDR) subfamily, 3 downregulated and 1 upregulated belonged to the multidrug resistance-associated protein (MRP) subfamily, and 4 upregulated and 1 downregulated belonged to pleiotropic drug resistance protein (PDR) subfamily (see Table S5A).
(ii) Secondary metabolite biosynthetic genes. Genes associated with secondary metabolite production are often arranged in biosynthetic gene clusters (BGCs) that consist of genes coding for core enzymes typically nonribosomal peptide synthetase (NRPS), polyketide synthase (PKS), or terpene cyclase, together with genes coding for additional proteins, including modifying enzymes, transporters, and transcription factors (61). We used antiSMASH to predict the biosynthetic gene clusters in C. rosea and identified 33 NRPS BGCs, 29 PKS BGCs, 7 BGCs for terpenes and 7 BCGs for NRPS-PKS hybrids, and 1 BGC for indole and betalactone biosynthesis.
Gene expression analysis of both Ddcl1 and Ddcl2 mutants identified a total of 230 DEGs predicted to be part of BGCs involved in secondary metabolite biosynthesis. Among the BGCs, the core biosynthetic genes in eight NRPS, five PKS, one terpene, and one indole BGCs were differentially regulated in the Ddcl2 mutant against B. cinerea or F. graminearum (Table 2; see also Table S5A). Interestingly, NRPS and PKS BGC core genes showed expression patterns opposite to each other since NRPS BGC core genes were mostly upregulated in the Ddcl2 mutant, whereas PKS BGC core genes were downregulated (Table 2). Among the downregulated core genes of PKS BGCs were the three PKS genes pks22, pks2, and pks12, reported to be part of previously identified BGCs responsible for the production of clonorosein and sorbicillin in C. rosea and T. reesei, respectively ( Fig. 4) (50,62). Sorbicillin is the precursor for sorbicillinol, which is in turn necessary for other sorbicillinoid compounds (63), explaining the low production of these substances by the Ddcl2 mutant.
(iii) Transcription factors. The transcriptome analysis further identified 128 differentially expressed genes predicted to encode transcription factors in the Ddcl1 and Ddcl2 strains (Table 1; see also Table S5A). We identified 11 transcription factors genes that were differentially expressed in the Ddcl1 strain and/or in the Ddcl2 strain and showed .50% sequence identity with genes previously characterized for their role as transcriptional regulators. CRV2G00011734 was upregulated in the Ddcl1 strain and showed identity with the conidiophore development regulator gene abaA (64, 65), whereas CRV2G00016352, CRV2G00019080, and CRV2G00019116, also upregulated, showed identity with the sucrose metabolic gene suc1, shown to be associated with mitotic and meiotic cell division in fission yeast (66). The genes CRV2G00004759, CRV2G00006707, Underlining indicates downregulated genes during Ddcl2 mutant-F. graminearum interactions. Boldfacing and underlining indicates genes that were downregulated against both mycohosts. The gene names for the sorbicillin and fumisoquin gene clusters were assigned by comparison to Trichoderma reesei and Aspergillus fumigatus, respectively (63,73). A minimum query coverage of 80% was required in the comparison, and the maximum E value was fixed at 1 Â 10 7 . and CRV2G00015419, downregulated in the Ddcl2 mutant, showed identity with transcription factor genes fgr27, hap3, and ypr2, shown to be involved in regulating growth and secondary metabolite production (62, 67, 68) ( Table 2). In summary, the Dicer-dependent control of transcription factor gene expression was to a large degree mycohost specific, with no transcription factors differentially expressed against both mycohosts in the Ddcl2 mutant. Moreover, among the identified transcription factors, there were many homologs of genes known to have a role in regulating secondary metabolism and growth.
(v) Genes associated with gene silencing machinery. To investigate an effect of dcl1 and dcl2 deletions on various protein components involved in the gene silencing machinery through chromatin modification in C. rosea, Blast2GO was used to identify genes encoding RNA helicases, chromatin remodeling proteins, histone deacetylases, and histone methyltransferases. We identified 118 genes (excluding DCL, AGO, and RDR), including 67, 23, 18, and 3 genes coding for RNA helicases, chromatin remodeling proteins, histone deacetylases, and histone methyltransferases, respectively (see Table S5B). Deletion of dcl1 did not cause differential expression in the Ddcl1 mutant-B. cinerea interaction, whereas during contact with F. graminearum we detected upregulation of two RNA helicase genes (CRV2G00006603 and CRV2G00007159), one gene coding for a chromatin remodeling protein (CRV2G00016556) and a histone deacetylase gene (CRV2G00012172), while one histone deacetylase gene (CRV2G00012172) was downregulated (Table 2). During the Ddcl2-B. cinerea interaction, one RNA helicase gene (CRV2G00006603) was upregulated, and two RNA helicases (CRV2G00001612 and CRV2G00012613), as well as an RNA-directed RNA polymerase (CRV2G00009762) were downregulated. Conversely, during the Ddcl2 mutant-F. graminearum interaction, two histone deacetylases (CRV2G00012165 and CRV2G00007951), one RNA helicase gene (CRV2G00006603), and one gene coding for an Argonaute protein (CRV2G00000975) were upregulated, whereas one RNA helicase (CRV2G00001612) gene was downregulated (Table 2). In summary, many genes involved in chromatin modification and gene silencing are affected by the deletion of the dcl enzymes, particularly dcl2. Most of these, including an Argonaute protein, are upregulated, possibly due to the diminished presence of regulating sRNAs in the mutants.
Analysis of sRNAs characteristics in the Clonostachys rosea WT and the dcl deletion strains. To investigate the effect of sRNAs on transcriptional regulation in C. rosea, sRNA libraries from C. rosea WT, Ddcl1, and Ddcl2 strains interacting with B. cinerea or F. graminearum were sequenced. The sequencing produced 16 million reads per sample on average. Between 61 and 72% of these read pairs were composed of nonstructural RNAs, including rRNA, tRNA, snoRNA, and snRNA, and were excluded from the further analysis. The remaining subset of reads that were 18 to 32 nucleotides (nt) long were used for alignment to the genomes of C. rosea, B. cinerea, and F. graminearum. A summary of sRNA characteristics and their alignment to the respective genome is presented in Table S6A in the supplemental material. sRNAs mapping exclusively to the C. rosea, B. cinerea, or F. graminearum genome (unique sRNAs) were selected for further analysis. On average 42% of sRNA reads from C. rosea-B. cinerea interactions were aligned uniquely to one of the two organisms, while this percentage was only 18% for C. rosea-F. graminearum interactions. This is plausible because C. rosea is evolutionarily closer to F. graminearum (both belong to order Hypocreales) than to B. cinerea.
We compared the characteristics of sRNAs produced in the Ddcl1 and Ddcl2 mutants to those of the WT. The analysis of length distribution showed a significant reduction in sRNAs with a size of 19 to 22 nt in the Ddcl2 compared to the WT, while no difference in sRNA abundance was found between the Ddcl1 and WT strains (Fig. 5A). The analysis of the 59 terminal nucleotide composition showed a reduced proportion of reads (27%) with 59 end uracil (59-U) in the Ddcl2 strain, compared to a 32 to 37% proportion of reads with 59-U from the WT and Ddcl1 strains (Fig. 5B). The origin of sRNAs was not significantly affected by the deletion of dcl genes, with most reads mapping to coding sequences (CDSs; 49%), followed by intergenic regions (25%), promoters (12.3%), 39 untranslated region s (UTRs) (8%), introns (4%), and 59 UTRs (1.5%). A higher proportion (83.5%) of sRNAs was mapped with the sense orientation, rather than the antisense one, similar to what was reported in previous studies in F. graminearum and T. atroviride (20,69), and this might be due to by-products of mRNA degradation. However, the relative proportion of sRNAs mapping to the antisense direction was reduced from an average of 17.5% during WT-B. cinerea interaction to 14.3% during Ddcl2 strain-B. cinerea interaction (see Table S6A).
(i) milRNA prediction in Clonostachys rosea. Mirdeep2 analysis predicted 61 milRNAs in C. rosea with lengths between 18 and 25 nt, and they were named cromir's. These milRNAs originated from a variety of positions in the genome including promoters, introns, CDSs, and UTRs, but mainly (28 of 61) from intergenic regions (see Table S6B). The expression of 15 cro-mir's was common against both mycohosts, whereas 29 and 17 cro-mir's were expressed specifically during interaction with B. cinerea or F. graminearum, respectively (see Table S6B). Interestingly, no cro-mir was found to be differentially expressed in the Ddcl1 mutant during the interspecific interactions, while 11 cro-mir's were significantly downregulated in the Ddcl2 mutant during interaction with both mycohosts (Table 3). This downregulation was confirmed through stem-loop RT-qPCR (Table 3). A single milRNA (cro-mir-23) was identified as upregulated in the Ddcl2 mutant in the RNA-seq analysis but downregulated according to stem-loop RT-qPCR.
(ii) Identification of cro-milRNAs endogenous gene targets. Twenty-one putative endogenous gene targets were identified for the 11 cro-mir's downregulated in Ddcl2 (Table 4). Eight gene targets were commonly upregulated in Ddcl2 during the interaction with B. cinerea and F. graminearum, while seven and six gene targets were uniquely upregulated during the interactions with B. cinerea and F. graminearum, respectively (Table 4). Among the predicted gene targets, several had putative regulatory roles: CRV2G00015277, CRV2G00002266, and CRV2G00002043 were predicted to encode putative transcription factors, CRV2G00001868 encodes an ATP-dependent helicase, while CRV2G00004332 and CRV2G00008014 encode a GTP binding protein and a GTPase with a putative role in signal transduction. Moreover, CRV2G00014914 was located in a secondary metabolite gene cluster and might have a role in regulating secondary metabolism (Table 4).  (iii) Cross-species gene target identification. Using the criteria described for the endogenous gene target prediction, we identified 513 putative cross-species gene targets in B. cinerea (see Table S6C). Among these, the seven genes bcpls1, bcpka1, bcnoxA, bcste11, bccap9, bccrh1, and bcchsIV were previously characterized for their role in growth and development, proteolysis, and consequently virulence (Table 5). Moreover, a gene encoding a B. cinerea homolog of SSAMS2 (BCIN_08g03180) was also among the putative targets, and this gene encodes a GATA transcription factor required for appressoria formation and chromosome segregation in Sclerotinia sclerotiorum (70). In addition, bcnog1 and bchts1 encoding proteins putatively involved in ribosome biogenesis, and bcphy2 and bchhk1 encoding signal transduction proteins were also identified as putative targets. Finally, three genes coding for a protein with a putative role in chitin recognition (bcgo1), chromatin remodeling (bcyta7), and intracellular trafficking and secretion (bcvac8) were also identified ( Table 5).
Thirty-five cross-species gene targets were predicted in F. graminearum as well. We identified three previously characterized virulence factors (FGSG_07067, FGSG_02083, and FGSG_00376) as putative targets of cro-mir-3, cro-mir-4, and cro-mir-5, respectively (Table 5). In addition, three membrane transporter genes (FGSG_13747, FGSG_13747, and FGSG_13747) and two genes coding for proteins with a putative role in intracellular trafficking and secretion (FGSG_09686 and FGSG_09686) were identified as putative targets (Table 5). In summary, several mycohost genes with a role in virulence, intracellular trafficking, secretion, and regulation were identified as putative targets of C. rosea dcl2-dependent milRNAs.
Botrytis cinerea and Fusarium graminearum responded differently toward Clonostachys rosea WT and dcl deletion strains. Transcriptome analysis of B. cinerea and F. graminearum was performed to investigate whether the deletion of dcl genes affects their response mechanism to C. rosea. Read pairs unique to B. cinerea from the C. rosea-B. cinerea interaction and unique to F. graminearum from the C. rosea-F. graminearum interaction were used in the analysis. From the total number of read pairs that originated from the C. rosea-B. cinerea or C. rosea-F. graminearum interactions, 25 and 23% reads were uniquely assigned to B. cinerea and F. graminearum, respectively (see Table S3).
In comparison to the WT-B. cinerea interaction, 24 genes (21 upregulated and 3 downregulated) were differentially expressed in B. cinerea during the Ddcl1 mutant-B. cinerea interaction. However, 721 genes were found to be differentially regulated (655 upregulated and 66 downregulated) in the interaction with the Ddcl2 mutant ( Fig. 6A; see also Table S6C). The 21 B. cinerea genes that were upregulated against the Ddcl1 strain were also upregulated against the Ddcl2 strain (Fig. 6A). We specifically investigated genes coding for hydrolytic enzymes, transcription factors, membrane transporters, known virulence factors, RNA silencing component proteins, and genes that are part of secondary metabolite BGCs. During Ddcl1 mutant-B. cinerea interaction, one gene (BCIN_14g03930) coding for a known virulence factor and two genes coding for MFS transporters were upregulated, while two genes that were part of secondary metabolite BGCs were downregulated in B. cinerea. Deletion of dcl2 induces increased expression of 12 genes previously characterized for their role in growth and development, virulence, and pathogenesis in B. cinerea. Among the other genes, we detected the upregulation of GTPases, kinases, chitinases, squalene monooxygenases, and genes involved in chitin synthesis and chitin recognition ( Table 6).
The other differentially expressed genes did not have a characterized functional role, but a function was predicted for some of them. In particular, among the genes upregulated during the Ddcl2 mutant-B. cinerea interaction, we detected 49 putatively coding for hydrolytic enzymes, 24 located in putative secondary metabolite BGCs, 22 transcription factors, 17 genes involved in RNA silencing, 15 protein kinases, and 13 MFS transporters (see Table S6C). GO enrichment analysis of upregulated genes during the Ddcl2 mutant-B. cinerea interactions identified terms for metabolic processes, including gene expression (GO:0010467), cellular component organization or biogenesis (GO:0071840), and RNA processing (GO:0006396) (Fig. 6B). However, GO terms oxidoreductase    activity (GO:0016491), oxidation-reduction processes (GO:0055114), and polysaccharide and glucan catabolic processes (GO:0000272 and GO:0009251) were enriched for the downregulated genes (Fig. 6B). During Ddcl2 mutant-F. graminearum interaction, 397 (169 upregulated and 128 downregulated) F. graminearum genes were differentially expressed, while only 97 (78 upregulated and 19 downregulated) were differentially expressed during the Ddcl1-F. graminearum interaction (Fig. 6C; see also Table S6D). Totals of 47 and 16 genes were upregulated and downregulated, respectively, against both mutant strains, whereas the rest were differentially expressed only during contact with one of the mutants (Fig. 6C). Furthermore, we found 26 (9 upregulated and 17 downregulated) previously characterized F. graminearum genes that were differentially regulated during the interaction with dcl deletion strains compared to the WT ( Table 6). The downregulated genes included several involved in deoxynivalenone, neurosporaxanthin, torulene, and fusarielin biosynthesis. Moreover, eight of the nine upregulated genes were previously characterized for having a role in F. graminearum virulence, and six of them encoded transcription factors (FgMYT3, GzHMG005, FgERB1, GzC2H066, GzZC230, and GzC2H059) ( Table 6). Additionally, during the interaction with the Ddcl2 mutant, 14 F. graminearum CAZyme genes showed upregulation with respect to the WT, all of them predicted to encode glycoside hydrolases, whereas only 3 genes were downregulated. MFS transporters were among the DEGs as well, with five of them being upregulated while seven were downregulated (see Table S6D).

DISCUSSION
While the Ddcl1 mutant had a phenotype largely similar to the WT, the Ddcl2 mutant displayed evident differences, including a higher number of differentially expressed genes during the interaction with the plant-pathogenic mycohosts. This number of DEGs was significantly higher than the number of genes predicted to be directly targeted from DCL2-regulated milRNAs, but it has already been observed in F. graminearum and T. atroviride how RNAi can be involved in regulating the activity of transcription factors and other regulatory elements and therefore indirectly influencing the expression of a vast array of genes and pathways (20,69). In our data set, we could observe four C. rosea transcription factors downregulated in the WT during interaction with the mycohosts and putatively targeted by milRNAs downregulated in the Ddcl2 mutants. Among these, CRV2G00015277 and CRV2G00002266 were involved in the interaction with both the mycohosts, while CRV2G00002043 was involved only in response to B. cinerea. CRV2G00002266 exhibited significant sequence similarities with the PRZ1 transcription factor, known for regulating the expression of the vacuolar ATPase Ca 21 pump PMC1 (71). This pump shown to regulates the level of cytoplasmic Ca 21 by activating Ca 21 -dependent enzymes involved in protein secretion in the nuclear envelope, endoplasmic reticulum, Golgi complex, and trans-Golgi/endosomal network in S. cerevisiae (71).
Furthermore, several other putative milRNA targets could have regulatory roles, including the predicted helicases CRV2G00001868 and CRV2G00004339 and the putative Rho-type GTPase activating protein CRV2G00008014. In particular, the transcript of gene CRV2G00004339, putatively targeted by milRNAs during interaction with F. graminearum, encodes a helicase of superfamily SNF2, involved in chromatin remodeling by deposition of H2A (72).
Beyond the direct action of milRNAs on targets, the deletion of dcl1 and especially dcl2 induced the differential expression of several secondary metabolite BGCs in C. rosea. The BGC containing the PKS gene pks22, involved in the synthesis of the antifungal compound clonorosein (50) was downregulated in the Ddcl2 mutant during the interaction with both mycohosts. In contrast, no difference in clonorosein A production was detected between the WT and the dcl mutants in the metabolome analysis. However, since the metabolome analysis was performed under in vitro conditions, it is possible that the dcl2dependent regulation of clonorosein production is more pronounced during contact with the mycohosts. In fact, pks22 was previously shown to be induced during interactions with B. cinerea and F. graminearum (50). The sorbicillin BGC, responsible for the yellow coloration of WT C. rosea colonies (50), is downregulated in the Ddcl2 mutant, and both sorbicillin and sorbicillinol were underproduced in the Ddcl2 mutant and had their biosynthesis restored in the complementation mutant in the in vitro trials, explaining the difference in pigmentation of the Ddcl2 mutant. This gene cluster was also induced during the interaction of C. rosea strain ACM941 with F. graminearum in the study of Demissie et al. (48). However, it is interesting that the positive regulator of the cluster, YPR1 (CRV2G00015416), is not differentially expressed in our study, whereas the transcription factor YPR2 (CRV2G00015419) is downregulated and hence coregulated with the other genes in the gene cluster in the Ddcl2 mutant. YPR2 is a Gal4-like transcription factor predicted to positively regulate a negative regulator of sorbicillin biosynthesis (62), and its coregulation with the biosynthetic genes suggests that the deletion of DCL2 affects the control of sorbicillin production at a currently unknown level.
Furthermore, two putatively important BGCs were specifically downregulated in the Ddcl2 mutant during contact with F. graminearum: these were the pks29 BGC involved in antagonism and biocontrol (50) and the BGC with the NRPS-like CRV2G00015275 as the core enzyme. This last cluster was studied as "cluster 3" in the work of Demissie et al. (47), where it was found to be induced in C. rosea after exposure to the F. graminearum secretome, and it presents strong homology with the fumisoquin cluster of Aspergillus fumigatus (73). Deletion of the core NRPS-like enzyme of the cluster leads to reduced growth and sporulation in A. fumigatus (74), but fumisoquins were not produced in detectable amounts by either the WT or the Ddcl2 mutant in our in vitro analysis. Biosynthesis of the corresponding compound in C. rosea may be specifically triggered during contact with F. graminearum. The transcription factor CRV2G00015277, putatively targeted by DCL2-dependent novel milRNAs cro-mir-5, cro-mir-10, and cromir-11, is located next to the cluster and is upregulated in the Ddcl2 mutant. It is possible that CRV2G00015277 is a negative regulator of the cluster, targeted by milRNAs to induce the production of fumisoquins, but this hypothesis should be examined in a future study. None of these gene clusters (sorbicillin, clonoroseins, pks29, and fumisoquins) were downregulated in the Ddcl1 mutant. The reduced production of bisorbicillinol in the Ddcl2 mutant also suggests that the deletion might hamper this fungus' antibacterial properties, since several bisorbicillinoids synthesized by C. rosea have significant antibacterial activity (75).
A further reason for the diminished capacity of the Ddcl2 mutant to control the plantpathogenic mycohosts can be found in the downregulation of genes encoding enzymes involved in the degradation of the fungal cell wall. In the Ddcl2 mutant, between 55 and 64 glycoside hydrolase genes were downregulated compared to the WT. Among these were three GH18 chitinases (ech37, ech42, and chiA5) and one GH20 N-acetylhexosaminidase (CRV2G00012950), which were downregulated during interaction with both mycohosts. Furthermore, four genes putatively involved in cell wall degradation of F. graminearum (48) were found to be downregulated in the Ddcl2 mutant: these were two glycoside hydrolases of classes GH2 (CRV2G00016896) and GH114 (CRV2G00003509), as well as two metallopeptidases (CRV2G00010851 and CRV2G00011092). Interestingly, the gene chiC1, predicted to encode a killer toxin-like chitinase that permeabilizes the cell wall of antagonistic species to facilitate entry of toxic metabolites (76,77), is upregulated in the Ddcl2 mutant. This may be explained by the fact that chiC1 is induced by chitin (44) and that the Ddcl2 mutant is compromised in its ability to antagonize the mycohosts, resulting in larger amounts of chitin exposed to the Ddcl2 mutant.
Moreover, 17 genes upregulated during C. rosea response to mycohosts in the study of Nygren et al. (49) were downregulated in the Ddcl2 mutants in comparison with the WT upon contact with the same mycohost. Among them is a putative isotrichodermin C-15 hydroxylase (cyp1), a type of protein also induced during mycoparasitism in T. cf. harzianum (78), but the majority of these genes is constituted by transporters, especially MFS transporters. This group includes gene mfs464, suggested in the study of Nygren et al. (49) to perform an important function in the mycoparasitic attack against F. graminearum, due to its extreme induction (fold change . 693). mfs166 and mfs464, downregulated in the Ddcl2 mutant, were found to be upregulated during the C. rosea response to F. graminearum in the studies of both Nygren et al. (49) and Demissie et al. (48), making their involvement in response to the mycohost very likely. The other detected differentially expressed MFS transporters are commonly involved in efflux-mediated protection against exogenous or endogenous secondary metabolites and sugar uptake, suggesting a DCL-dependent influence on this aspect of C. rosea mycoparasitic action. This group also includes nine genes belonging to the drug-H 1 antiporter-2 family, which underwent a significant gene expansion during C. rosea evolution and has therefore a putative important role in the fungus lifestyle (79). DCL-based control of these transporters is most likely indirect because most MFS genes detected in this way are downregulated in the mutants, whereas direct targets of RNA silencing are expected to be upregulated after dcl deletion. Reinforcing this hypothesis, none of the MFS transporters predicted in C. rosea is a putative target of differentially expressed milRNAs detected in this study. Identification of several upregulated genes coding for MFS transporters used by mycohosts to tolerate harmful secondary metabolites of their own production strengthens the hypothesis that these proteins enable C. rosea to withstand mycohost-produced toxins during fungus-fungus interaction.
The differential expression of this vast number of genes is likely due to the 128 putative transcription factors differentially expressed in the Ddcl2 mutant. Among these, CRV2G00006707 is a homolog of the CCAAT-binding subunit HAP3, regulating growth and secondary metabolism in other filamentous fungi such as F. verticillioides (68,80). This gene is downregulated in the Ddcl2 mutant during interaction with both mycohosts (log2 fold change [log 2 FC] of 21.6 in Cr-Bc [C. rosea 1 B. cinerea] and 21.3 in Cr-Fg [C. rosea 1 F. graminearum]). Another transcription factor downregulated in the Ddcl2 mutant was CRV2G00004759, a homolog of the filamentous growth regulator 27 (fgr27) of Trichoderma lentiforme, which is involved in adherence regulation and could have a role in reduced growth rate of the mutant (67). Moreover, two putative homologs of the sucrose utilization protein 1 (SUC1) are upregulated in the Ddcl2 mutant, and its upregulation is associated with a delay in mitotic and meiotic nuclear divisions in Schizosaccharomyces pombe (66).
It is possible that part of the reduced ability of the Ddcl2 mutant to overgrow B. cinerea in vitro and control F. graminearum in vivo comes from a cross-regulating action of C. rosea milRNAs targeting mycohost genes involved in the development or reduction of virulence. Specifically, three F. graminearum virulence factors were both downregulated during interaction with the WT C. rosea and putatively targeted by milRNAs downregulated in the Ddcl2 mutants. These genes included FGSG_07067, the GzZC232 transcription factor whose deletion impaired virulence in the work of Son et al. (81); FGSG_00376, the NOS1 NADH ubiquinone oxidoreductase proven to be a factor of virulence by Seong et al. (82); and FGSG_02083, the transcription factor ART1, whose deletion causes reduced starch hydrolysis and virulence, as well as the incapability of trichothecenes biosynthesis (83). Regarding B. cinerea, among the putative milRNA-targeted downregulated genes, there were those encoding BCIN_09g06130, the BcPls1 tetraspanin necessary for appressoriummediated penetration into host plant leaves (84), and BCIN_16g01130, the bcpka1 catalytic subunit of the cAMP-dependent protein kinase, whose deletion affects the lesion development and leaves rot caused by the fungus (85). Two other putative targets were BcnoxA (BCIN_05g00350), a component of the B. cinerea NADPH oxidase complex necessary for the colonization of host tissues (86), and the MAP triple kinase BcSte11 (BCIN_03g02630), whose deletion is known to cause defects in germination, delayed vegetative growth, reduced size of conidia, lack of sclerotium formation, and loss of pathogenicity in B. cinerea (87). Moreover, a B. cinerea homolog of Ssams2 (BCIN_08g03180) was also among the putative targets, and this gene encodes a GATA transcription factor required for appressoria formation and chromosome segregation in S. sclerotiorum (70).
Several other genes encoding virulence factors were found to be upregulated in the pathogenic mycohosts during the interaction with the Ddcl2 mutant, even if they were not among the putative targets of milRNAs. Among the F. graminearum genes upregulated during contact with the Ddcl2 mutant were the transcription factors MYT3, ERB1, GzHMG005, GzC2H066, GzZC230, and GzC2H059, whose disruption reduces the virulence of the pathogen (81,(88)(89)(90)(91), as well as the phospholipase PLC1, known for its involvement in hyphal growth, conidiation, deoxynivalenol production, and virulence (92). Regarding B. cinerea, among the genes upregulated during contact with the Ddcl2 mutant, we found nop53 and noxR, crucial for fungal development and virulence through the regulation of reactive oxygen species (93,94); frq1, involved in circadian regulation of fungal virulence (95); and vel1, whose deletion affects virulence and light-dependent differentiation (96). Moreover, among the upregulated genes there was also a homolog (BCIN_14g03930) of the S. sclerotiorum transcription factor SsNsd1, necessary for pathogenicity and appressorium formation (97). Furthermore, upon contact with the Ddcl2 mutant, B. cinerea upregulated several genes encoding proteins involved in chitin and cell wall synthesis, such as Bccrh1, BcchsIV, BcchsV, BcchsVI, and BcchsVII (98)(99)(100)(101). The upregulation of BcCHSVI and BcCHSVII is of particular interest because these proteins have a role in plant infection (101).
Genes encoding two virulence factors of F. graminearum (TRI5 and TRI6) and one of B. cinerea (MFSG) were downregulated during interaction with the Ddcl2 mutant. The gene mfsG is involved in B. cinerea virulence by providing tolerance to glucosinolatebreakdown products (102), but the C. rosea Ddcl2 mutant shows downregulation in several putative secondary metabolite clusters compared to the WT. Therefore, it is possible that the expression of mfsG is reduced during contact with the mutant because the lack of production of harmful compounds makes it unnecessary for the mycohost to express resistance genes. TRI5 and TRI6 are involved in the synthesis of trichothecenes (103,104), and other genes involved in the biosynthesis of these mycotoxins are similarly downregulated during contact with the Ddcl2 mutant, including the genes TRI1, TRI3, TRI4, TRI9, and TRI14 (105). This is surprising because F. graminearum overexpresses the transcription factor gene ART1 during contact with the Ddcl2 mutant, and this transcription factor is known to be a positive regulator of trichothecene biosynthesis (83). The reduced ability of the Ddcl2 mutant to control F. graminearum may make it unnecessary for the mycohost to produce DON in high quantities, despite ART1 overexpression. Interestingly, among the most relevant genes proven to be DON-responsive in C. rosea in a previous study (106), only 1 of 16 was found to be less expressed in the Ddcl2 mutant than in the WT during interaction with F. graminearum: a homolog of glucose repressible protein GRG1 (CRV2G00000966). Given the reduced expression of DON-biosynthesis genes by F. graminearum, the downregulation of a higher number of DON-responsive genes was expected.
Another important mycotoxin produced by F. graminearum is zearalenone, and the zearalenone hydrolase gene zhd101 (CRV2G00011056) was found to be downregulated by the Ddcl2 mutant. The deletion of this gene undermines C. rosea mycoparasitic action against F. graminearum (107), and its downregulation is therefore a possible reason for the impaired biocontrol action of the Ddcl2 mutant. Another zearalenone-responsive gene, one encoding a putative bacteriorhodopsin-like protein (106), is also downregulated in the Ddcl2 mutant, but its role in the C. rosea-F. graminearum interaction is still unknown.
Interestingly, F. graminearum showed altered production of red pigment at the point of contact with the Ddcl2 mutant, which could plausibly be due to downregulation of genes belonging to the gene clusters of carotenoid and fusarielin (108,109). However, the gene cluster of aurofusarins, known for their red colorations, was not differentially expressed during the interaction with the Ddcl2 mutant.
Conclusions. DCL-dependent RNA silencing plays a determinant role in the regulation of many biological processes. In the present study, the role of DCL-like enzymes was investigated for the first time in the antagonistic action of the fungus C. rosea. Our result show that DCL2-mediated RNAi plays a central role in regulating endogenous cellular processes involved in growth, secondary metabolite production, and antagonism toward the mycohosts, whereas the function of DCL1 is redundant except for conidium production. The observed phenotypic effect in Ddcl2 strains is due to the diminished production of antifungal metabolites in the mutant, as well as to downregulation of genes known to be involved in mycohost response and resistance to secondary metabolites. Identification of 11 milRNAs, which were downregulated in the Ddcl2 strain, and their putative endogenous gene targets, including transcription factors and chromatin remodeling proteins, indicates DCL-dependent regulation of C. rosea antagonistic interactions. Furthermore, we predicted putative cross-species gene targets in the mycohosts B. cinerea and F. graminearum previously characterized for their role in fungal virulence, posing the bases for future studies focusing on the role of DCL-dependent RNA silencing in interspecific fungal interactions.

MATERIALS AND METHODS
Fungal strains and culture conditions. C. rosea strain IK726 WT and mutants derived from it, B. cinerea strain B05.10, and F. graminearum strain PH1 were used in this study. The fungal cultures were maintained on PDA (Oxoid, Cambridge, UK) medium at 25°C.
Gene identification and phylogenetic analysis. C. rosea strain IK726 genome version 1 (41) and version 2 (55) were screened for the presence of genes encoding DCL, AGO, and RDR by BLASTP analysis. The presence of conserved domains was analyzed using the Simple Modular Architecture Research Tool (SMART) (110), InterProScan (111), and conserved domain search (112).
Amino acid sequences of DCLs (DCL1 and DCL2), AGOs (AGO1 and AGO2), and RDRPs of several fungal species (see Table S1A) were retrieved from the UniProt and GenBank databases (113,114). The sequences of Dicer1, Argonaute1, and RDR of the model plant Arabidopsis thaliana were retrieved from the UniProt database (113) and used as outgroups. Sequences were aligned with mafft v.7 (115) with options suggested for ,200 sequences (L-INS-i), and the phylogenetic trees were generated using iqtree v.1.6.12 (116) with 1,000 bootstrap replicates and option "MFP" to find the best substitution model. Figtree v. 1.4.4 (117) was used to visualize the trees.
Construction of deletion vector, transformation, and mutant validation. The ;1-kb 59-flank and 39-flank regions of dcl1 and dcl2 were amplified from genomic DNA of C. rosea using gene-specific primer pairs (see Table S1B), as indicated in Fig. S1 (53). Gateway cloning system (Invitrogen, Carlsbad, CA) was used to generate entry clones of the purified 59-flank and 39-flank PCR fragments as described by the manufacturer (Invitrogen, Carlsbad, CA). The hygromycin resistance cassette (hygB) generated during our previous studies (43,118) from pCT74 vector, as well as a Geneticin resistance cassette generated as a PCR product from the pUG6 vector (119), were used. A three-fragment multisite gateway LR recombination reaction was performed using the entry plasmids of respective fragments and destination vector pPm43GW (120) to generate the deletion vectors. Complementation cassettes for dcl1 and dcl2 were constructed by PCR amplification of the full-length sequence of dcl1 and dcl2, including ;800bp upstream and ;500-bp downstream regions from genomic DNA of C. rosea WT using gene-specific primers (see Table S1B). The amplified DNA fragments were purified and integrated into destination vector pPm43GW using two-fragment gateway cloning technology to generate complementation vectors.
Agrobacterium tumefaciens-mediated transformation was performed based on a previous protocol for C. rosea (43,121). Transformed strains were selected on plates containing either hygromycin for gene deletion or Geneticin for complementation. Validation of homologous integration of the deletion cassettes in putative transformants were performed using a PCR screening approach with primer combinations targeting the hygB cassette and sequences flanking the deletion cassettes (see Fig. S1), as described previously (45,122). PCR-positive transformants were tested for mitotic stability and then purified by two rounds of single-spore isolation (118). To determine the transcription of dcl1 and dcl2 in the WT, deletion, and complementation strains, total RNA from the respective strains were isolated (Qiagen, Hilden, Germany). After DNase I treatment, according to the manufacturer's instructions (Merck, Kenilworth, NJ) reverse transcription-PCR (RT-PCR) was performed using RevertAid premium reverse transcriptase (Fermentas, St. Leon-Rot, Germany) and gene-specific primer pairs (see Table S1B).
Phenotypic analyses. Phenotypic analyses experiments were performed with C. rosea WT, deletion strains dcl1 (Ddcl1) and dcl2 (Ddcl2), and their respective Ddcl11 and Ddcl21 complemented strains. Each experiment was repeated twice with similar results.
The growth rate, colony morphology, and conidium production were analyzed in four biological replicates as described previously (43). To analyze mycelial biomass, agar plugs of C. rosea strains were inoculated in 50-ml conical flasks with 20 ml of PDB (Oxoid, Cambridge, UK), followed by incubation at 25°C under constant shaking (100 rpm). Biomass production was determined by measuring the mycelial dry weight 5 days postinoculation. The antagonistic behavior against B. cinerea and F. graminearum was tested using an in vitro plate confrontation assay on PDA medium, as described previously (51). The growth of B. cinerea and F. graminearum was measured daily until their mycelial fronts touched the C. rosea mycelial front. The experiments were performed in four biological replicates. The biocontrol ability of C. rosea strains against F. graminearum was evaluated in a fusarium foot rot assay, as described previously (123). In brief, surface sterilized wheat seeds were treated with C. rosea conidia (1 Â 10 7 conidia/ ml) in sterile water, sown in moistened sand, and kept in a growth chamber after pathogen inoculation (51). Plants were harvested 3 weeks postinoculation, and disease symptoms were scored on scale of 0 to 4, as described previously (51,123). The experiment was performed in five biological replicates with 15 plants in each replicate.
Statistical analysis. ANOVA was performed on phenotype data using the general linear model approach implemented in Statistica version 16 (TIBCO Software, Inc., Palo Alto, CA). Pairwise comparisons were made using the Tukey-Kramer method at a 95% significance level.
Metabolite analysis. An agar plug of C. rosea strains was inoculated on PDA (Oxoid) and allowed to grow for 10 days at 25°C. Agar plugs, together with mycelia, were harvested from the centers of plates using 50-ml Falcon tubes (53). The mycelial plug was sonicated for 15 min in 20 ml of methanol, and then 1 ml of extract was transferred to a 1.5-ml centrifuge tube for centrifugation at 10,000 Â g for 5 min. Supernatants were collected and then analyzed by UHPLC-MS on a reversed-phase column (2.1 Â 50 mm, 1.5 mm; Accucore Vanquish; Thermo Scientific, Waltham, MA) using a gradient of acetonitrile (MeCN) in water, both with 0.2% formic acid (10 to 95% MeCN in 3 min and 95% MeCN for 1.2 min, at 0.9 ml min 21 ). The MS was operated in positive mode with scanning of m/z 50 to 1,500, and the mass spectra were calibrated against sodium formate clusters using the Compass DataAnalysis 4.3 software (Bruker Daltonics, Bremen, Germany) that was also used for general data analysis. UHPLC-MS/MS was run with the same instrument, column, and UHPLC conditions, using the auto-MS/MS function (11 precursor ions, m/z 50 to 1,500, with ramped fragmentation energies of 20/30/35 eV for m/z 200/500/1,000). The UHPLC-MS data were converted to mzXML format using DataAnalysis 4.3, and ion chromatogram peak picking in the range 5 to 200 s was performed using the program XCMS in software environment R using the centWave method (peak width, 3 to 20 s; m/z tolerance, 5 ppm; noise, 1,000) (124,125). XCMS was used for subsequent peak grouping and missing peak filling. For each sample, the resulting molecular feature peak areas were normalized against the sum of peak areas, and the resulting relative peak areas were 10 logarithmized. The data were used for PCA, and ANOVA was used to evaluate significant differences in concentrations between strains. Tentative compound identification was done by comparing high-resolution mass spectrometry data on fungal compounds from the databases Antibase and combined chemical dictionary. The identity of the tentatively identified compounds was further corroborated by analysis of MS/MS data. The experiment was performed in five biological replicates.
Dual culture interaction experiment for sRNA and transcriptome sequencing. An agar plug of C. rosea strains was inoculated at edge of a 9-cm-diameter PDA (Merck, Kenilworth, NJ) petri plate covered with a Durapore membrane filter (Merck) for easy harvest of mycelia. The mycohost fungi B. cinerea or F. graminearum were inoculated at opposite side of the plate (43). Due to different mycelial growth rates, C. rosea was inoculated 7 days prior to the inoculation of F. graminearum or B. cinerea. The mycelial front (5 mm) of C. rosea was harvested together with the mycelial front (5 mm) of B. cinerea (Cr-Bc) or F. graminearum (Cr-Fg) at the hyphal contact stage of interactions (see Fig. S2A) and snap-frozen in liquid nitrogen. The experiment was performed in three biological replicates.
RNA extraction, library preparation, and sequencing. Total RNA was extracted using the mirVana miRNA isolation kit according to the manufacturer's protocol (Invitrogen, Waltham, MA). The RNA quality was analyzed using a 2100 Bioanalyzer Instrument (Agilent Technologies, Santa Clara, CA) and concentration was measured using a Qubit fluorometer (Life Technologies, Carlsbad, CA). For sRNA and mRNA sequencing, the total RNA was sent for library preparation and paired-end sequencing at the National Genomics Infrastructure (NGI), Stockholm, Sweden. The sRNA library was generated using TruSeq small RNA kit (Illumina, San Diego, CA), while the mRNA library was generated using a TruSeq Stranded mRNA Poly(A) selection kit (Illumina, San Diego, CA). The sRNA and mRNA libraries were sequenced on a NovaSeq SP flow cell with a 2 Â 50-bp reads and NovaSeqXp workflow in S4 mode flow cell with 2 Â 151 setup, respectively, using Illumina NovaSeq6000 equipment at NGI Stockholm. The Bcl to FastQ conversion was performed using bcl2fastq_v2.19.1.403 from the CASAVA software suite (126). The quality scale used was Sanger/phred33/Illumina 1.81.
(i) Functional annotation of genomes. The predicted proteomes of C. rosea strain IK726, B. cinerea strain B05.10 (ASM14353v4), and F. graminearum strain PH-1 (ASM24013v3) were annotated through BLAST2GO v.5.2.5 (127) and InterProScan v.5.46-81.0 (111) with default parameters to identify transcription factors. Secondary metabolite clusters were predicted through antiSMASH v.5.0 (128), while predicted enzymes involved in the metabolism of carbohydrates (CAZymes) were identified using the dbCAN2 meta-server (129). The amino acid sequences of B. cinerea and F. graminearum were compared to the PHI-base database using BLAST (130) with a minimum of 80% in both identity and query coverage. All identified matches described in the PHI-base annotation by the keywords "reduced virulence" or "loss of pathogenicity" were considered to be potential virulence factors.
(ii) Differential expression and GO enrichment analyses. Reads were trimmed with bbduk v.38.86 (131) with the following options: bbduk.sh in1=read1.fastq in2=read2.fastq out1=read1_clean.fastq out2=read2_clean.fastq ref=reference.fa ktrim=r k =23 mink=11 hdist=1 tpe tbo qtrim=r trimq=10. Successful cleaning and adapter removal was verified with fastqc v. 0.11.9 (https://www.bioinformatics .babraham.ac.uk/projects/fastqc/). Since all the samples represented the interaction of two organisms, the genome of C. rosea was concatenated with the one of either B. cinerea or F. graminearum, creating two "combined genomes" (Cr-Fg and Cr-Bc), and the same was done with the annotations in .gff format. Reads from the C. rosea-B. cinerea interaction were aligned to the Cr-Bc genome, whereas reads from the C. rosea-F. graminearum interaction were aligned to the Cr-Fg. The chosen aligner was STAR v.2.7.5c (132), with default options, and the count tables were then generated through featureCounts v.2.0.1 (133). Finally, the differential expression analysis was done with DESeq2 v.1.28.1 (134), where an FDR of ,0.05 in combination with a log 2 FC of .1.5 or ,21.5 was considered to define differentially expressed genes (DEGs). Enrichment in gene ontology (GO) terms of DEGs was determined through Fisher tests integrated in the BLAST2GO suite, with an FDR threshold of 0.05.
(iii) Mapping of sRNAs. sRNA reads were trimmed with bbduk v.38.86 (131) with the same options used for mRNA read trimming, and successful cleaning and adapter removal was verified with fastqc v.0.11.9. The program SortMeRna v.4.2.0 (135) was used to remove structural sRNA (rRNA, tRNA, snRNA, and snoRNA) from the reads, and sequences within the length range of 18 to 32 bp were isolated with the command reformat.sh of the BBTools suite (131). The database of structural RNAs used for SortMeRna consisted in the rRNA sequences of the SILVA database (136), while snRNA, tRNA, and snoRNA sequences were downloaded from the NRDR database (137). After filtering, the sRNA reads were mapped to the Cr-Bc and Cr-Fg genomes with STAR, with the following options recommended for sRNA mapping: STAR -genomeDir index/-readFilesIn read1.fq read2.fq -outFileNamePrefix prefix -outFilterMismatchNoverLmax 0.05 -outFilterMatchNmin 16 -outFilterScoreMinOverLread 0 -outFilterMatchNminOverLread 0 -alignIntronMax 1 -alignEndsType EndToEnd. For the STAR default option, reads with good mapping results on more than 20 different loci were considered "not mapped." Untranslated regions (UTRs) and introns were added to the .gff files of the genomes through "add_utrs_ to_gff" (https://github.com/dpryan79/Answers/tree/master/bioinfoSE_3181) and GenomeTools with the "-addintrons" option (138), respectively. Promoters were also added through an ad hoc Python script (https://github.com/EdoardoPiombo/promoter_extractor), considering promoters to be composed of the first 1,000 bases upstream of a gene or of all the bases until the end of the precedent gene. Introns, promoters, and UTRs were all considered when featureCounts was used to generate the count tables.
(iv) Prediction of miRNA-like RNAs and subsequent analyses. Putative milRNAs were predicted with mirdeep2 v.2.0.1.2 (139). The miRbase database (140), as well as all the fungal milRNA sequences from RNAcentral (141), were used to provide reference sequences from other species. To ensure the novelty of newly detected milRNAs, BLAST was used to compare them to the fungal milRNAs identified in several other studies, plus all the fungal milRNAs available in RNAcentral, requiring 95% minimum identity and query coverage (25,33,(141)(142)(143)(144)(145). Nonstructural sRNA reads, previously mapped to the genomes with STAR, were counted with featureCounts, and the differentially expressed milRNAs were identified with DESeq2, with the same thresholds used for DEG analysis.
The sRNA_toolbox (146) was used to predict putative targets for the identified milRNAs. The prediction was carried out with the animal-based tools PITA, Miranda, TargetSpy (147)(148)(149), and simple seed analysis and with the plant-based tools psRobot, TAPIR FASTA, and TAPIR RNAhybrid (150,151). Target-milRNA couples identified by at least three animal-based tools or two plant-based ones were retained for the following analyses. Predicted targets were retained only when they were significantly expressed (FDR , 0.05) with a log 2 FC .1.0 opposite to the milRNA. Putative targets of downregulated milRNAs were therefore considered only when they were overexpressed. The predicted targets present in double copy in their genome were then removed from the analysis. Repetitive elements in the genome of C. rosea were predicted according to the guidelines for basic repeat library construction (http://weatherby .genetics.utah.edu/MAKER/wiki/index.php/Repeat_Library_Construction-Basic), using all fungal transposons in RepetDB as known transposons (152), and putative milRNA targets within 700 bp from any C. rosea transposon were removed from the analysis.
(v) Validation of milRNA-expression through stem-loop RT-qPCR. milRNA specific stem-loop RT-qPCR primers (see Table S1B) were designed as described previously (153). Stem-loop RT primers (1 mM) were denatured at 65°C for 5 min and immediately transferred to ice. For each milRNA RT reaction, a "no RNA" master mix was prepared with 0.5 ml of 10 mM dNTP (Thermo Scientific, Waltham, MA), 5Â SSIV buffer, 2 ml of 0.1 M dithiothreitol, 0.1 ml of RiboLock RNase inhibitor (40 U/ml), 0.25 ml of SSIII reverse transcriptase (Invitrogen, Waltham, MA), 1 ml of denatured stem-loop RT primer, and 1 ml of 5 mM reverse primer of C. rosea actin (act) reference gene. Next, 10 ng of RNA template used for next-generation sequencing analysis was added into respective reactions. The tubes were then incubated in a thermal cycler at 16°C for 30 min, followed by 60 cycles of pulsed RT at 30°C for 30 s, 42°C for 30 s, and 50°C for 1 s and then enzyme inactivation at 85°C for 5 min. Quantitative PCR was performed using DyNAmo Flash SYBR green kit (Thermo Scientific, Waltham, MA) according to the manufacturer's instructions. The C T values of milRNA were normalized to that of act to be used for quantification using the DDC T method (154).
Data availability. The raw sequencing data were submitted to ENA in under BioProject accession number PRJEB43636. This project contains both the transcriptome and the sRNA sequencing data for each of the samples.