The mycoparasitic fungus Clonostachys rosea responds with both common and specific gene expression during interspecific interactions with fungal prey

Abstract Clonostachys rosea is a necrotrophic mycoparasitic fungus, used for biological control of plant pathogenic fungi. A better understanding of the underlying mechanisms resulting in successful biocontrol is important for knowledge‐based improvements of the application and use of biocontrol in agricultural production systems. Transcriptomic analyses revealed that C. rosea responded with both common and specific gene expression during interactions with the fungal prey species Botrytis cinerea and Fusarium graminearum. Genes predicted to encode proteins involved in membrane transport, biosynthesis of secondary metabolites and carbohydrate‐active enzymes were induced during the mycoparasitic attack. Predicted major facilitator superfamily (MFS) transporters constituted 54% of the induced genes, and detailed phylogenetic and evolutionary analyses showed that a majority of these genes belonged to MFS gene families evolving under selection for increased paralog numbers, with predicted functions in drug resistance and transport of carbohydrates and small organic compounds. Sequence analysis of MFS transporters from family 2.A.1.3.65 identified rapidly evolving loop regions forming the entry to the transport tunnel, indicating changes in substrate specificity as a target for selection. Deletion of the MFS transporter gene mfs464 resulted in mutants with increased growth inhibitory activity against F. graminearum, providing evidence for a function in interspecific fungal interactions. In summary, we show that the mycoparasite C. rosea can distinguish between fungal prey species and modulate its transcriptomic responses accordingly. Gene expression data emphasize the importance of secondary metabolites in mycoparasitic interactions.


| INTRODUC TI ON
Chemical pesticides play an important role for maintaining high yields in agricultural productions systems in the world. However, the extensive use of agricultural chemicals has led to concerns about unwanted, negative effects on the environment and human health. One approach to minimize the amount of used pesticides is integrated pest management, which promotes the integrative use of preventive cultural practices, disease-resistant plant cultivars and mechanical and biological control of pathogen populations. Biological control can also be used in organic agriculture, where chemical pesticides are banned. The use of biological control of plant pathogens by microbial antagonists can therefore be expected to increase, and to become standard procedure, in future agricultural and horticultural production.
We recently sequenced the genome of C. rosea strain IK726 and performed a comparative genome analysis with Trichoderma and Fusarium species . Conspicuous features of the C. rosea gene content include a high number of genes encoding ATP-binding cassette (ABC) and major facilitator superfamily (MFS) membrane transporters, polyketide synthase (PKS) and nonribosomal peptide synthetase (NRPS) secondary metabolite biosynthesis proteins, cytochrome P450 (CYP) and polysaccharide lyase family 1 (PL1) pectin lyases . The selection for high number of genes in these gene families suggests that their gene products may be involved in ecological niche adaptation (Wapinski, Pfeffer, Friedman, & Regev, 2007), such as mycoparasitism. Although C. rosea and Trichoderma spp. are necrotrophic mycoparasites with broad host range, it is not known if, or to what extent, they modulate their responses towards different fungal prey species. A better understanding of the mechanisms that determine the outcome of biocontrol interactions is necessary for science-based improvements of biocontrol applications in agriculture.
The aim with the current work is to identify C. rosea genes that are specifically induced during interactions with the fungal pathogens B. cinerea or F. graminearum, and genes induced in response to both pathogens, using an RNA-seq approach. We hypothesize that the transcriptomic response in C. rosea towards the two pathogens differs due to their intrinsic differences in cell wall composition, secondary metabolite spectra, etc. We further hypothesize that there will be an overlap between genes induced during fungal-fungal interactions and genes evolving under diversifying selection (in analogy with effectors in other parasite-host interactions). The results show that C. rosea responds with both common and specific transcriptional changes during interactions with B. cinerea and F. graminearum. Induction of genes putatively encoding drug resistance membrane transporters and proteins involved in biosynthesis of secondary metabolites emphasize the importance of secondary metabolites in mycoparasitic interactions.

| Cultivation conditions
Clonostachys rosea strain IK726 (WT) and mutants derived from it, F. graminearum strain PH-1 and B. cinerea strain B05.10 were maintained on potato dextrose agar (PDA) medium (Oxoid, Cambridge, UK) at 25°C in darkness unless otherwise specified. Vogel's minimal (VM) medium (Vogel, 1956) with 0.3% glucose was used to grow fungi for transcriptome sequencing, while Czapek dox (CZ) medium (Sigma-Aldrich, St. Louis, MO, USA) was used to grow fungi for reverse transcription quantitative PCR (RT-qPCR) gene expression analyses, and phenotypic analyses of mutant strains.
For transcriptome sequencing during interactions, an agar plug of C. rosea mycelium and of the prey fungi F. graminearum or B. cinerea were inoculated at opposite sides of a 9-cm-diameter VM agar Petri plate (covered with a nylon membrane for easy harvest) and incubated at 25°C in darkness. Due to different mycelial growth rates, C. rosea was inoculated 5 days prior to the inoculation of F. graminearum or B. cinerea. The growing front (6 mm) of C. rosea mycelium was harvested together with the mycelial front (2 mm) of the prey fungi, 24 hr after hyphal contact between the fungi. Mycelium harvested from C. rosea confronted with itself at the same stage was used as the control treatment. Two biological replicates (different Petri plates) for each treatment were used.
In the control treatments, ZEA and fungicides were replaced with an equal volume of methanol or sterile distilled water, respectively. Fungal mycelia were harvested 2 hr after the addition, washed in distilled water to remove traces of ZEA or fungicides, frozen in liquid nitrogen and stored at −80°C.

| Nucleic acid isolation and manipulation
Genomic DNA was extracted from C. rosea as described previously (Nygren et al., 2008). RNA extraction was performed using the Qiagen RNeasy kit following the manufacturer's protocol (Qiagen, for 3 min on a GeneAMP PCR system 2700 (Applied Biosystems, Carlsbad, CA, USA). Running the PCR for 18 cycles was found to be optimal based on agarose gel electrophoresis (producing maximum amount of PCR products, but before entering the plateau phase of the reaction), and the PCR product was purified using a PCR purification kit (Fermentas, Waltham, MA, USA), analysed using a 2100 Bioanalyzer Instrument (Agilent Technologies, Santa Clara, CA, USA), quantified using a Qubit fluorometer (Life Technologies) and sent for paired-end transcriptome sequencing at the SNP&SEQ Technology Platform, Uppsala, Sweden, using Illumina HiSeq2000 equipment. Read lengths were 107 and 144 bp, respectively, and the average insert sizes for the different samples were 305-340 bp.

| Transcriptome data analysis
Quality filtration and adapter removal were performed using Cutadapt ver. 1.2.1 (Martin, 2011), minimum read length was set to 50 bp, and only reads still belonging to a complete pair were kept for further analysis. TopHat ver. 2.0.9 and Cufflinks ver. 2.1.1 packages (Trapnell et al., 2012) were used to map reads to the C. rosea, B. cinerea and F. graminearum reference genomes (Amselem et al., 2011;Cuomo et al., 2007;Karlsson et al., 2015), quantitate and test for differential gene expression. Parameter settings were kept at default, except for intron size adjustment.
Fungi generally have short introns; therefore, minimal intron length was set to 5 bp and maximum intron length to 5,000 bp.
Reads were analysed as unstranded, and read pair distance was adjusted for each sample. One mismatch was allowed in the analyses. The data were corrected for multiple testing by adjusting p-values to a false discovery rate (FDR) of .05 using Cuffcompare in Cufflinks. Two of three conditions contained tissue from two species, C. rosea and either of B. cinerea and F. graminearum.
Therefore, all reads were aligned to the three reference genomes and only the read pairs with a better alignment to the C. rosea genome than to either of the two other genomes were used in further analyses. To avoid biases in comparisons between conditions, all samples were filtered the same way. Expression levels were determined as reads per kilobase of transcript per million mapped reads.

| Gene expression analysis with RT-qPCR
RNA extraction was performed using the Qiagen RNeasy kit following the manufacturer's protocol (Qiagen). After DNaseI (Fermentas, St. Leon-Rot, Germany) treatment, one microgram of total RNA was reverse transcribed in a total volume of 20 μl using Maxima first stand cDNA synthesis kit (Fermentas, St. Leon-Rot, Germany).
Transcript levels were quantified by RT-qPCR using the SYBR Green PCR Master Mix (Fermentas, St. Leon-Rot, Germany) and primer pairs listed in Appendix S1, in an iQ5 qPCR System (Bio-Rad, Hercules, CA, USA) as described previously (Tzelepis, Melin, Jensen, Stenlid, & Karlsson, 2012). Melt curve analysis was performed after the qPCR reactions, to confirm that the signal was the result from a single product amplification. Relative expression levels for target genes in relation to actin (act), shown previously to be constitutively expressed (Kamou et al., 2016;Tzelepis, Dubey, Jensen, & Karlsson, 2015), were calculated from the Ct values by the 2 −ΔΔC t method (Livak & Schmittgen, 2001). Gene expression analysis was carried out in at least three biological replicates, each based on two technical replicates. Gene expression data were analysed by analysis of variance (ANOVA) using a general linear model approach implemented in Minitab ver. 18 (Minitab Inc., State College, PA, USA).
Pairwise comparisons were made using the Fisher method at the 95% significance level.

| MFS transporter gene family evolution
Seven hypocrealean fungi and Neurospora crassa were included in studying the evolutionary history of the MFS gene family. Wholegenome nucleotide and protein sequences of the mycoparasitic fungi C. rosea , T. atroviride, T. reesei and T. virens Martinez et al., 2008), and the plant pathogenic fungi F. graminearum, F. solani and F. verticillioides (Coleman et al., 2009;Cuomo et al., 2007;Ma et al., 2010), and the saprotrophic fungus N. crassa (Galagan et al., 2003) were retrieved from the National Center for Biotechnology Information (NCBI). ABC and MFS transporter genes were identified by BlastP analysis in an iterative process described previously , although the predicted C. rosea MFS transporters BN869_T00000646, BN869_T00007234 and BN869_T00002052 were additionally in- Phylogenetic relationships were taken from Karlsson et al. (2015), while branch lengths were calculated based on a four-gene alignment including actin, glyceraldehyde 3-phosphate dehydrogenase, DNA-directed RNA polymerase II subunit B and translation elongation factor 1 alpha. Coding gene sequences were retrieved from the respective genome sequences. Each gene was aligned individually using Clustal W (Larkin et al., 2007) in MEGA ver. 6 (Tamura, Stecher, Peterson, Filipski, & Kumar, 2013), concatenated and used to calculate branch lengths in MEGA ver. 6. The resulting species phylogeny was calibrated by setting the split between T. reesei and T. virens to 16 million years (de Man et al., 2016).
Major facilitator superfamily gene family evolution analysis was carried out on subgroups that contained ≥2 genes in at least one species and were present in ≥2 species. The program CAFE (Computational Analysis for Gene Family Evolution) ver. 3 (Han, Thomas, Lugo-Martinez, & Hahn, 2013) was used to test whether gene family sizes were compatible with a stochastic birth-and-death model, to estimate gene family size in extinct species and to identify lineages with accelerated rates of gene gain or loss. Mutation rate (λ) was 0.0024 .

| Phylogenetic analyses
Full-length protein sequences were aligned by MUSCLE (Edgar, 2004), and phylogenetic trees were constructed using neighbourjoining implemented in MEGA ver. 6. The JTT amino acid substitution model (Jones, Taylor, & Thornton, 1992) was used with uniform rates among sites and pairwise deletion of gaps. Statistical support for branches was assessed by 500 iterations of bootstrap resampling. Conserved protein modules and features were identified using the conserved domain database (CDD) (Marchler-Bauer et al., 2017) at NCBI.

| Analysis of molecular evolution
Regions of low amino acid conservation in MFS transporter alignments were identified by reverse conservation analysis (RCA), as described by Lee (2008). In short, Rate4Site ver. 2.01 was used to calculate the degree of conservation (S score, high scores correspond to low degree of conservation) for each amino acid position using the empirical Bayesian method (Mayrose, Graur, Ben-Tal, & Pupko, 2004;Pupko, Bell, Mayrose, Glaser, & Ben-Tal, 2002). A sliding-window average (n = 7) of normalized S scores (mean was 0 and standard deviation was 1) was plotted in Excel (Microsoft; W mean score), and significant peaks were defined by values ≥1.

| Construction of gene deletion vectors, transformation and mutant validation
Three fragment multisite gateway cloning system (Invitrogen, Carlsbad, CA, USA) was used to construct gene deletion vectors.
The ~ 1 kb 5′-flank and 3′-flank regions of mfs464, mfs602, fdo1 and cyp1 were amplified from genomic DNA of C. rosea using gene specific primer pairs ups F/ups R and ds F/ds R, respectively, as indicated in Appendix S1. Gateway entry clones of the purified 5′-flank and 3′-flank PCR fragments were generated as described by the manufacturer (Invitrogen). The gateway entry clones for the hygromycin (hph) gene, constructed during our previous studies Dubey, Ubhayasekera, Sandgren, Jensen, & Karlsson, 2012;Dubey, Broberg, Sooriyaarachchi et al., 2013), were used. The gateway LR recombination reaction was performed using entry plasmid of respective fragments and destination vector pPm43GW (Karimi, De Meyer, & Hilson, 2005) to generate the deletion vectors. Agrobacterium tumefaciens-mediated transformation (ATMT) was performed based on a previous protocol for C. rosea (Utermark & Karlovsky, 2008). Putative transformants were selected on a plate containing hygromycin (200 μg/ml). Validation of homologous integration of the deletion cassettes in putative transformants was performed using a PCR screening approach with primer combinations targeting the hph gene (Hyg F/Hyg R) and sequences flanking the deletion cassettes (Appendices S1 and S2) as described previously Dubey et al., 2014). RT-PCR analysis was conducted on WT and gene deletion strains using RevertAid premium reverse transcriptase (Fermentas, St. Leon-Rot, Germany) and primer pairs specific for the respective genes (Appendix S1). The PCR-positive transformants were tested for mitotic stability, and were purified by two rounds of single spore isolation (Dubey et al., 2014).

| Phenotypic analyses of gene deletion mutants
For growth rate analysis, a 3-mm-diameter agar plug from the grow- (20 mM), zinc chloride (20 mM) or lithium chloride (10 mM). In addition, growth rates were analysed under nutrient-limited conditions at pH 4.0, pH 9.0 or pH 10.0. Limitation for carbon, nitrogen, magnesium, iron or potassium was induced by reducing the concentration of sucrose, sodium nitrate, magnesium sulphate, ferrous sulphate or potassium chloride by 1/10. Colony diameter was measured after 5 days of growth at 25°C in darkness. The concentration of chemicals used for phenotypic analysis in this study was based on our previously published results (Dubey et al., 2014 with the exception of ions. The appropriate concentration of ions was selected based on screening of C. rosea WT spores and mycelium on CZ plates amended with a series of different concentrations of the respective compounds. For determination of conidial germination, germ tube development and colony formation, WT and deletion strain conidia were inoculated (10 3 conidia) on solid agar plates using the same concentration of chemical compounds described above with the exception of H 2 O 2 (0.8 mM) and Teldor (fenhexamid 1000 μg/ml). Data for conidial germination and germ tube development were recorded daily using a Leica 165FC microscope (Wetzlar, Germany), while colony diameter was measured 5 days postinoculation. All phenotypic analyses were performed with at least three independent mutants (to avoid phenotypes resulting from ectopic insertions of the deletion cassette), and each based on at least three biological replicates unless otherwise specified.
Antagonistic behaviour of C. rosea WT and deletion strains against F. graminearum or B. cinerea was tested in five biological replicates using an in vitro plate confrontation assay and a culture filtrate assay, as described previously (Dubey et al., 2014. In addition, a liquid culture confrontation test was performed in five biological replicates, where conidia (10 5 conidia) from WT and deletion strains were inoculated in 50 ml VM with 0.3% glucose in Erlenmeyer flasks and were allowed to grow for 24 hr at 25°C in darkness. Subsequently, an equal amount of F. graminearum spores were added to the flasks and further incubated for 3 days to allow the antagonistic interaction. All mycelia were harvested, washed with distilled water and used for DNA extraction. The mycelial biomass was estimated by measuring C. rosea and F. graminearum DNA with qPCR using primers specific to β-tubulin (Dubey et al., 2014;Reischer, Lemmens, Farnleitner, Adler, & Mach, 2004). An in vivo bioassay using a sand seedling test for fusarium foot rot disease on barley was performed in five biological replicates, where each replicate included 15 plants, following the procedure described previously (Dubey et al., 2014.

| Transcriptional response of C. rosea during interactions with B. cinerea and F. graminearum
Dual confrontation assays (C. rosea vs. C. rosea, C. rosea vs. B. cinerea and C. rosea vs. F. graminearum) were monitored using a stereo microscope on a daily basis, and mycelium from the interaction zone was harvested 24 hr after the first hyphal contacts were observed.
This occurred 10 days after inoculation of C. rosea for all interactions.
Illumina HiSeq paired-end sequencing of cDNA generated from RNA from the interaction zones resulted in a total of 117 million read pairs, equivalent to 143.5 Gbp of sequence. After removal of lowquality reads and reads originating from B. cinerea or F. graminearum, 36.8 million read pairs remained that were used for analysis and identification of differentially expressed genes. In total, 41 differentially expressed genes were identified, and functional classification showed that 61% were predicted to encode membrane transporters, 12% were predicted to encode proteins involved in biosynthesis of secondary metabolites, and 7% encoded carbohydrate-active enzymes (Table 1). Six genes were induced in C. rosea during interactions with both B. cinerea and F. graminearum, while most genes were induced specifically towards either of the two different fungal prey species (Table 1).
Among the 26 differentially expressed membrane transporter genes, 22 were identified as putative MFS transporters while only two were ABC transporters. The two ABC transporter genes were described previously as abcC8 and abcG18  and predicted to encode members of the multidrug resistance-associated protein (MRP, group C) and the pleiotropic drug resistance protein (PDR, group G) ABC transporter groups, respectively. Expression of abcC8 was induced during interaction with B. cinerea, while abcG18 was induced specifically against F. graminearum (Table 1) (Table 1).
The PKS gene pks9, putatively encoding a reduced polyketidetype PKS, and the NRPS gene nps13  were induced during confrontation with B. cinerea in comparison with F. graminearum although expression of nps13 was not significantly different from the C. rosea self-interaction control treatment. A predicted CYP (BN869_T00000852 [cyp1]) with similarity to an isotrichodermin C-15 hydroxylase was induced specifically against F. graminearum. The killer toxin-like chitinase gene chiC1  was induced 50-fold during interactions with F. graminearum, in comparison with the B. cinerea and C. rosea self-interaction treatments. A predicted FAD-dependent oxidoreductase (BN869_ T00000606 [fdo1]) with similarity to a tetracycline resistance protein was induced 86-fold against F. graminearum. A putative PL1 pectin lyase gene (BN869_T00006915) was induced during interaction with B. cinerea, but repressed during interaction with F. graminearum (Table 1).

| Validation of differential expression by RT-qPCR
Validation of RNA-seq gene expression data with RT-qPCR was carried out for nine selected genes, including three genes specifically induced during interaction with B. cinerea and six genes specifically induced against F. graminearum. There was a high level of congruence between gene expression patterns measured with the two methods (Table 1, Figure 1). Four minor differences were detected; expression of abcC8 was higher in the C. rosea vs. C. rosea treatment compared with the F. graminearum confrontation using RT-qPCR but not with RNA-seq, while expression of ptr1, abcG18 and cyp1 was higher in the F. graminearum confrontation compared with the C. rosea vs. C. rosea treatment using RNA-seq but not with RT-qPCR ( Figure 1).

| Evolution of the MFS transporter gene family in C. rosea and other hypocrealean fungi
The massive induction of membrane transporter genes in C. rosea during confrontations with fungal prey prompted us to investigate these gene families in more detail. The ABC transporter gene family in C. rosea was previously described in detail , but an iterative Blast approach identified an additional four genes resulting in a total number of 90 ABC transporter genes in the C. rosea genome. Based on a phylogenetic analysis (data not shown), these four additional ABC transporter genes were classified as members of subgroup E (BN869_T00001090) and subgroup F (BN869_T00005647, BN869_T00006720 and BN869_T00007586) according to the Kovalchuk and Driessen (2010) classification system, and named abcE1, abcF3, abcF4 and abcF5, respectively.
Using an iterative Blast approach, we further identified a total of 634 MFS transporter genes in the genome of C. rosea (Appendix S3).

| Structural divergence of MFS 2.A.1.3.65 paralogs
Alignments and RCA analyses were applied to reveal conserved and variable regions between three MFS 2.A.1.3.65 phylogenetic groups with ≥60% bootstrap support (here referred to as clades A, B and C, Figure 2). Ten regions with high amino acid diversity (W ≥ 1) were identified in the MFS 2.A.1.3.65 alignment (Figure 4). Five of these regions (1-5) were predicted to be associated with loops that connect transmembrane helices on the inside of the cell and the beginning of the transmembrane helices themselves (Figure 4).
Region 6 was predicted to be part of the C-terminal, internal loop.
Hence, all six regions were predicted to form parts of the entry of the transport tunnel, likely influencing substrate specificity.    A and B in Figure 3). Only two of these regions displayed signs of functional divergence (W ≥ 1 in group A and W < 0.5 in group B) and were labelled 1 and 2 ( Figure 5). Region 1 was located in the end of a large, highly variable internal loop region, while region 2 was located in a C-terminal internal loop ( Figure 5).

| Mycotoxin-and fungicide-induced gene expression in C. rosea
As many differentially expressed genes belonged to expanded gene families predicted to be involved in drug resistance/secondary  [Utermark & Karlovsky, 2007]) and to four fungicides with different mode of action. The ABC transporter gene abcG18 was induced 11,152-fold after exposure to zearalenone for 2 hr, while the predicted FAD-dependent oxidoreductase gene fdo1 was induced 2.7fold by zearalenone (Table 3). The ABC transporter gene abcC8, the MFS transporter gene mfs602 and fdo1 were induced during exposure to fungicides boscalid and mefenoxam, while mfs602 and fdo1 were additionally induced by iprodione (Table 3).

| Phenotypic analyses of gene deletion strains in C. rosea
Clonostachys rosea gene deletion mutants were generated by replacing mfs602, mfs464, fdo1 and cyp1 with the hygromycin selection cassette (hygB) by ATMT. Successful gene replacements in mitotically stable mutants were confirmed by PCR using primers located within the hygB cassette in combination with primers located upstream and downstream of the constructs (Appendix S2) as described previously Dubey et al., 2012Dubey et al., , 2014Dubey et al., , 2016. The expected size of PCR fragments were amplified in selected mfs602, mfs464, fdo1 and cyp1 mutant strains, while no amplification was observed in WT (Appendix S2). Furthermore, RT-PCR experiments using primers specific to mfs602, mfs464, fdo1 and cyp1 demonstrated the complete loss of the respective transcripts in each mutant (Appendix S2).
When C. rosea WT or the ∆mfs464 mutants were co-inoculated with F. graminearum in liquid VM with 0.3% glucose, there was a significant (P = .001) 2.1-3.7-fold reduction in F. graminearum biomass (measured as F. graminearum DNA/C. rosea DNA ratio) after 3 days of interaction ( Figure 6). No reductions in F. graminearum biomass were observed for the ∆mfs602, ∆fdo1 or ∆cyp1 mutants. No additional phenotypic effects in any of the four C. rosea gene deletion mutants were observed for any of the studied phenotypes, including tolerance to mycotoxins, fungicides, chemical agents for cell wall, osmotic or oxidative stress, or ion transport/tolerance, nutrient deficiencies, or in vitro dual culture interaction, or biocontrol of fusarium foot rot on barley (data not shown).

| D ISCUSS I ON
Necrotrophic mycoparasites such as C. rosea and Trichoderma spp. are assumed to have broad host ranges with little specificity. Evidence to support this view comes from the fact that single species, and sometimes single strains, are efficient biocontrol agents against a range of different plant pathogenic fungi (Jensen et al., 2007), and sometimes even against oomycete and plasmodiophorid pathogens (Lahlali & Peng, 2014;Moller, Jensen, Andersen, Stryhnz, & Hockenhull, 2003  Relative amount of F. graminearum DNA also showed that all possessed mycoparasitic activity against the plant pathogenic species A. alternata, B. cinerea and S. sclerotiorum . Therefore, it is intriguing to observe that C. rosea reacts with both common and specific gene expression during interspecific interactions with B. cinerea and F. graminearum, suggesting that C. rosea can distinguish between fungal prey species and modulate its responses accordingly. A comparative transcriptomic study of T. atroviride, T. reesei and T. virens during interaction with Rhizoctonia solani (Atanasova, Le Crom et al., 2013) revealed both common and specific responses between the three species. The common response included induction of ABC and MFS transporters, proteases and heat shock proteins.
Trichoderma atroviride mycoparasitism was characterized by induced genes involved in biosynthesis of secondary metabolites and fungal cell wall degradation, but also small, secreted cysteine-rich proteins, while the predatory mycoparasitism of T. virens was associated with induction of genes involved in gliotoxin biosynthesis (Atanasova, Le Crom et al., 2013). The weak mycoparasite and wood-degrading T. reesei induced cellulase and hemicellulase genes.
A comparative genome analysis identified high numbers of genes encoding ABC and MFS transporters, PKS, NRPS, CYP and PL1 genes, and a low number of chitinases, in the C. rosea genome , suggesting that their gene products may be involved in ecological niche adaptation (Wapinski et al., 2007) including biotic interactions, nutrient acquisition and stress modulation. It is therefore notable that members from all of these gene families are also induced during interspecific fungal interactions, indicating that mycoparasitism is the main driver for shaping the genome content in C. rosea. The common response in C. rosea against both fungal preys included sugar and small organic compound transporters that may be involved in nutrient uptake. The specific responses towards the two different fungal preys were dominated by genes predicted to be involved in membrane transport, biosynthesis of secondary metabolites and carbohydrate degradation, which fits well with our current view that degradation of the cell wall of the fungal prey, production of toxic metabolites for poisoning of the fungal prey and the ability to tolerate the counterattack by the fungal prey (by means of toxic metabolites and reactive oxygen species) are key mechanisms for a successful necrotrophic mycoparasite .
The chiC1 chitinase gene was previously shown to be induced by chitin  and is predicted to encode a killer toxinlike chitinase that permeabilize the cell wall of antagonistic species to facilitate entry of toxic metabolites (Karlsson & Stenlid, 2008;Seidl, Huemer, Seiboth, & Kubicek, 2005). This function was previously confirmed for another killer toxin-like chitinase in C. rosea; deletion of the chiC2 gene resulted in lower growth-inhibiting activity of culture filtrates against B. cinerea and R. solani, but notably, not against F. graminearum . This may be significant as the 50-fold induction of chiC1 against F. graminearum, but not against B. cinerea, may suggest adaptation of killer toxin-like paralogs towards different fungal preys (driven by differences in cell wall composition, presence of chitinase inhibitors etc.). Differential regulation of killer toxin-like paralogs was also reported from T. atroviride (Gruber, Vaaje-Kolstad et al., 2011) and T. virens , and the saprotrophic Aspergillus nidulans (Tzelepis, Melin, Stenlid, Jensen, & Karlsson, 2014) and N. crassa (Tzelepis et al., 2012). The killer toxin-like function was also confirmed in a nonmycoparasitic species; deletion of the chiC2-2 gene in A. nidulans resulted in reduced growth-inhibiting activity of culture filtrates (Tzelepis et al., 2014).
Clonostachys rosea is known to produce a limited number of secondary metabolites with biotic activity; peptaibols with toxicity against S. sclerotiorum (Rodriguez et al., 2011), polyketides with antibacterial activity (Zhai et al., 2016), glisoprenin inhibitors of appressorium formation (Thines, Eilbert, Anke, & Sterner, 1998) and nematicidal epipolysulfanyldioxopiperazines (Dong, He, Shen, & Zhang, 2005). However, it is not possible at this stage to connect the secondary metabolite biosynthetic genes induced against B. cinerea or F. graminearum with any of these compounds. The pks9 gene induced against B. cinerea is predicted to encode a reduced lovastatin diketide polyketide, and other PKS genes of the same type were induced during mycoparasitism in both T. atroviride (Atanasova, Le Crom et al., 2013) and T. reesei (Atanasova, Knox, Kubicek, Druzhinina, & Baker, 2013). Deletion of the nonreducing type pks4 gene in T. reesei resulted in loss of pigmentation and mutants that were more sensitive to toxic metabolites produced by the fungal prey (Atanasova, Knox, Kubicek et al., 2013). Induction of genes with similarity to isotrichodermin C-15 hydroxylase (cyp1) and norsolorinic acid reductase in C. rosea during interaction with F. graminearum may indicate production of hitherto unknown secondary metabolites, and similar genes were reported to be induced during mycoparasitism in T. cf. harzianum (Steindorff et al., 2012(Steindorff et al., , 2014Vieira et al., 2013).
Interpretation of the induction of C. rosea ABC and MFS transporters during interaction with B. cinerea and F. graminearum is made difficult by the fact that both groups of membrane transporters exists as large gene families with a complex phylogenetic substructure that typically reflects their involvement in a wide range of biological functions, including basic metabolism and cellular homoeostasis (Kovalchuk & Driessen, 2010). The ABC transporter gene family in C. rosea has been studied in detail , where ABC transporter group B (multidrug resistance proteins) and G (pleiotropic drug resistance proteins) were shown to evolve under selection for increased gene copy numbers. The abcG18 gene, induced against F. graminearum and during exposure to the Fusarium mycotoxin zearalenone, indeed belongs to group G, suggesting a role of the ABCG18 protein in protecting C. rosea against toxic metabolites produced by fungal prey (at least by Fusarium spp.). Three other ABC transporter genes from group G (abcG5, abcG8 and abcG29) and one gene from group B (abcB26) were also reported to be induced in C. rosea during exposure to ZEA Kosawang, Karlsson, Jensen, Dilokpimol, & Collinge, 2014), and both ABCG5 and ABCG29 were later shown to be involved in biocontrol of fusarium foot rot disease (Dubey et al., 2014. The C. rosea abcB1, abcB4, abcB18, abcB20 and abcB26 group B and the abcG8 and abcG25 group G ABC transporter genes were also reported to be induced by exposure to bacterial metabolites (Kamou et al., 2016;Karlsson et al., 2015), suggesting that C. rosea needs to defend itself not only to metabolites from the fungal prey but also to antibiosis from the surrounding microflora in the rhizosphere. The abcC8 gene belongs to the multidrug resistance-associated proteins, subgroup C-V  and the induction during interaction with B. cinerea and during exposure to fungicides boscalid and mefenoxam may suggest an involvement in cellular protection against exogenous toxic metabolites and xenobiotics.
The The phylogenetic structure of both families shows that they evolve by a rapid birth-and-death process followed by sequence diversification, indicating that functional diversification between paralogs is the driving force for the observed gene copy number increase, rather than selection for increased protein amount (where sequence conservation between paralogs are expected). MFS602 belong to family 2.A.1.3.65 predicted to be involved in drug resistance, and our analysis of structural divergence showed that loops forming the entry to the transport channel are the targets of the diversification between paralogs, indicating that selection for substrate specific- metabolite trichodermin and consequently affect in vitro antagonism (Liu et al., 2012). MFS1 was also shown to transport xenobiotics such as iprodione and a range of azole-type fungicides. This dual function fits well with the gene expression profile of mfs602 that was induced during interaction with F. graminearum but also during exposure to the fungicides iprodione, boscalid and fenhexamid.
However, deletion of the mfs602 gene in C. rosea did not result in any measurable phenotype. This may not be unexpected, given the large number of similar paralogs present in C. rosea that can mask any potential effect by partially overlapping functions, which is previously The function of MFS464 is more difficult to speculate about, as no member of MFS family 2.A.1.2.33 is characterized to any detail.
Even the type transporter of the family, HOL1 from S. cerevisiae, was only shown to transport histidinol after random mutagenesis, which not necessarily provides evidence for its true biological function (Wright et al., 1996).

ACK N OWLED G EM ENTS
We acknowledge financial support from the Department of Forest Council and the Knut and Alice Wallenberg Foundation.

CO N FLI C T O F I NTE R E S T
None declared.

DATA A R C H I V I N G S TAT E M E N T
The transcriptome sequence data generated and analysed in this work are deposited in the short read archive (SRA) at NCBI, with accession numbers SRX3389264 -SRX3389269.