Two algicolous Arthrinium species, A. saccharicola KUC21221 and A. koreanum KUC21332, were obtained from the Korea University Culture (KUC) collection. Arthrinium species found in the egg mass of A. japonicus are known to have originated from an adjacent brown alga, as A. japonicus spawns on S. fulvellum, one of the major endophytic hosts of Arthrinium spp. (Heo et al., 2018). For genome sequencing, they were cultured in 1-L Erlenmeyer flasks containing 500 mL potato dextrose broth for 7 days at 25°C in dark. Genomic DNA was extracted using the DNeasy Plant Mini Kit (Qiagen, Valencia, CA, United States) according to the manufacturer’s instructions with slight modifications (Kohler et al., 2011). To improve the accuracy of genome annotation, total RNA was extracted from the same cultures using the RNeasy Plant Mini Kit (Valencia, CA, United States) according to the manufacturer’s instructions. The DNA and RNA qualities were determined using the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, United States) with a DNA 1000 chip.

Arthrinium koreanum KUC21332 was inoculated on malt extract agar (MEA, Bacto, Sparks, MD, United States). The fungus was cultured for 7 days, and three agar plugs with mycelium were used as the inocula. To assess the difference in gentisyl alcohol production according to the nitrogen source [peptone, sodium glutamate, KNO₃, and (NH₄)₂SO₄], the fungus was cultured in 250-mL Erlenmeyer flasks containing 50 mL medium (40 g glucose, 10 g nitrogen source, 0.5 g MgSO₄, 0.5 g KH₂PO₄, and 0.5 g K₂HPO₄ in a liter of distilled water) for 10 days in dark. To determine the carbon source (glucose, sucrose, maltose, soluble starch, and mannitol) and to perform RNA-Seq, it was cultured in the medium (40 g carbon source, 10 g sodium glutamate, 0.5 g MgSO₄, 0.5 g KH₂PO₄, and 0.5 g K₂HPO₄ in a liter of distilled water) for 3 days in dark. Total RNA was extracted in the same way as described above.

A DNA library with approximately 20-kb fragment sizes was constructed and the WGS was acquired using the PacBio Sequel platform at Macrogen Co., Ltd., (Seoul, South Korea). Meanwhile, a single molecule real-time (SMRT) library was constructed and sequenced with a single SMRT cell. The reads were trimmed, corrected, and filtered. Data on high-quality and error-corrected Illumina reads were used as input for the program Proovread v2.14.0 to correct the potential sequencing errors in the PacBio long reads (Hackl et al., 2014). Assembler Falcon was used for de novo assembly of the corrected PacBio reads (Chin et al., 2016). The assemblies were finalized and manually corrected after polishing using the paired-end Illumina reads and Pilon v1.21 (Walker et al., 2014).

To improve the accuracy of genome annotation, mRNA in the samples was sequenced. The RNA-Seq library was constructed and Illumina HiSeq 4000 sequencing was performed at Macrogen Co., Ltd., (Seoul, South Korea). Trimmed and corrected RNA-Seq reads were aligned to the reference genome using HISAT2 v2.1.0 (Kim et al., 2019). The genes were predicted using Augustus v3.3.3, BRAKER v2.1.5, GenMark-ES v4.61, and GlimmerM v2.5.1 (Majoros et al., 2003; Stanke et al., 2006, 2008; Ter-Hovhannisyan et al., 2008; Hoff et al., 2016, 2019). The final consensus gene model was constructed using all predictions using EVidenceModeler v1.1.1 (Haas et al., 2008). Repeats were masked using RepeatMasker v4.1.0 and RepeatModeler v2.0¹. Annotation completeness was evaluated using BUSCO v4.1.2 on fungi_odb10, ascomycota_odb10, and sordariomyceta_odb10 gene sets (Seppey et al., 2019). Orthologous protein families with the reference genome datasets were identified using Orthofinder v2.4.0 with default setting except for an inflation parameter (I = 2.5) (Emms and Kelly, 2019).

The proteins were annotated by predicting functional domains from Pfam using InterProScan (Hunter et al., 2009; El-Gebali et al., 2019). To further facilitate functional interpretation, proteins were aligned to the non-redundant database of NCBI². Gene ontology (GO) terms were mapped using InterProScan v5.29-68.0 (Gene Ontology Consortium, 2012; Jones et al., 2014). Carbohydrate-active enzymes (CAZymes) were analyzed using dbCAN2 (hmmer) (Potter et al., 2018; Zhang et al., 2018). Gene clusters related to secondary metabolism were analyzed using antiSMASH Fungi v5.1.2, and secondary metabolite regions were identified using strictness “relaxed” (Blin et al., 2019).

To obtain fungal extracts, the fungal cultures were filtered with 0.45 μm syringe filters (Chromafil PE- 20/15 MS, Macherey-Nagel, Düren, Germany) and extracted with 50 mL of ethyl acetate for 24 h. The ethyl acetate layer was collected and dried at 35°C using a rotary evaporator, and the obtained extracts were stored at −18°C until use.

The 2,2-diphenyl-1-picrylhydrazyl (DPPH, Sigma-Aldrich Inc., St. Louis, MO, United States) was dissolved in 80% methanol at 150 μM. The 198 μL of DPPH solution and 22 μL of the fungal extracts (10 mg/mL DMSO) was mixed in each well in a 96-well plate. The plate was allowed to reach a steady state for 30 min at room temperature in dark. The absorbance was measured at 540 nm using a microplate reader (Sunrise^TM, Tecan Group Ltd., Port Melbourne, VIC, Australia).

To analyze differentially-expressed genes (DEGs), RNA-Seq libraries were constructed using the Illumina TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, San Diego, CA, United States) and were sequenced on the Illumina NovaSeq 6000 platform at Macrogen Co., Ltd., (Seoul, South Korea). Quality check of sequenced reads was performed using the FastQC v0.11.7³.

The RNA-Seq data were analyzed using the “new Tuxedo” pipeline according to a published protocol (Pertea et al., 2016). Briefly, the raw reads were quality filtered using the program Trimmomatic v0.39 and aligned to each genomic DNA reference obtained in this study using HISAT2 v2.1.0 (Bolger et al., 2014; Kim et al., 2019). Data on the transcripts and their expression levels were assembled and estimated using the StringTie v1.3.4 (Kovaka et al., 2019). The DEG analysis was conducted using the DESeq2 (Love et al., 2014). To reduce systematic bias that could affect biological meaning in comparison between samples, the size factor was estimated using count data and normalized using median of ratios method. Between the experimental groups, there was no significant difference (p = 0.564) in the expression level of pyruvate kinase (GO:0004743), which catalyzes the final step of glycolysis, indicating that the difference in secondary metabolism was independent of the amount of available energy (Abdel Fattah et al., 2010; Zhang et al., 2016).

Statistical analyses were performed with R version 3.5.3 (R Development Core Team, 2013). Non-metric multidimensional scaling (NMDS) and permutational multivariate analysis of variance (adonis2) were performed using the package “vegan” (Oksanen et al., 2013).

In the present study, the genomes of two marine Arthrinium spp., A. koreanum KUC21332 and A. saccharicola KUC21221, isolated from the inner tissue of a brown alga Sargassum fulvellum and egg mass of A. japonicus, respectively, were analyzed. A. koreanum demonstrates a genome size of 48.75 Mbp (GC content: 52.09%), including 14,381 gene models, and A. saccharicola exhibits a 55.08-Mbp genome (GC content: 50.07%) with 14,773 models (Table 1). Both genomes are larger than those of A. phaeospermum AP-Z13 (48.45 Mbp) and A. arundinis NRRL 25634 (47.67 Mbp), with A. saccharicola genome being the largest among the four species. The number of gene models of both species is similar to A. phaeospermum (14,055 models), while that of A. arundinis NRRL 25634 is higher (16,992 models). The genome assembly completeness of both species ranges from 96.3–98.8%.

The four Arthrinium spp. shared 10,001 genes (Figure 1A). A. koreanum shared more genes (615) with A. arundinis compared to the genes shared with A. saccharicola (212) and A. phaeospermum (146). A. saccharicola shared more orthogroups (672) with A. phaeospermum compared to those shared with A. arundinis (208) and A. koreanum (212). Compositions of shared and specific genes were analyzed by gene ontology (GO) categories and have been illustrated in Figure 1B. The genes that only existed in A. koreanum (circle XII) were unidentified. Approximately half (mean 51.2%) of the genes were involved in biological process (BP), and 37.0 and 11.5% were involved in molecular function (MP) and cellular component (CC), respectively. Metabolic process (GO:0008152), cellular process (GO:0009987), and localization (GO:0051179) accounted for 48.6, 29.0, and 13.1% of the total genes in BP, respectively. Additionally, there were seven genes involved in interspecies interactions between organisms (GO:0044419), all of which were common in at least two Arthrinium spp. Four of them were shared by all four Arthrinium spp., of which two were involved in the defense response to gram-negative bacteria (GO:0050829), one was responsible for anti-bacterial defense response (GO:0042742), and the other were involved in pathogenesis (GO:0009405). One gene shared by A. arundinis, A. koreanum, and A. phaeospermum was involved in the DNA restriction-modification system (GO:0009307). One gene shared by A. arundinis, A. saccharicola, and A. phaeospermum, and the other shared by A. koreanum and A. saccharicola, were involved in pathogenesis (GO:0009405). In the case of CC, cellular anatomical entity (GO:0110165), protein-containing complex (GO:0032991), and intercellular (GO:0005622) accounted for 84.6, 7.8, and 7.2% of the total, respectively. In the case of MF, catalytic activity (GO:0003824) and binding (GO:0005488) accounted for 46.9 and 40.1% of the total genes, respectively, followed by transporter activity (GO:0005215) and molecular function regulator (GO:0098772), accounting for 7.5 and 3.6%, respectively.

Additionally, the genes encoding CAZymes and plant cell wall degrading enzymes (PCWDEs) and compare to those of other ascomycetes exhibiting various lifestyles. Five, eight, and twelve species were selected for dark septate endophytes (DSEs), plant pathogens, and saprobes, respectively, by referring to the Mycocosm group of the Joint Genome Institute and previous studies (Supplementary Table 1; Galagan et al., 2003; Dean et al., 2005; Güldener et al., 2006; Hane et al., 2007; Van Den Berg et al., 2008; Ellwood et al., 2010; Andersen et al., 2011; Rouxel et al., 2011; Gianoulis et al., 2012; Ohm et al., 2012; O’Connell et al., 2012; Grigoriev et al., 2013; Koike et al., 2013; Traeger et al., 2013; Nordberg et al., 2014; Xu et al., 2015; David et al., 2016; Jourdier et al., 2017; Knapp et al., 2018). DSE is a paraphyletic group (Sordariomycetes, Pezizomycetes, Dothideomycetes, Leotiomycetes, Eurotiomycetes, etc.) of endophytic fungi belonging to Ascomycota, and was selected because it represented endophytic ascomycetes. The sequence data of A. arundinis NRRL 25634, Entoleuca mammata CFL468, Pestalotiopsis sp. NC0098, and Truncatella angustata MPI-SDFR-AT-0073 were produced by the United States Department of Energy Joint Genome Institute⁴ in collaboration with the user community.

The Arthrinium spp. had a total of 601-626 CAZymes, of which 307–323, 96–104, 9–11, 62–68, 11–14, and 113–117 were associated with glycoside hydrolase (GH), glycosyltransferase (GT), polysaccharide lyase (PL), carbohydrate esterase (CE), carbohydrate-binding module (CBM), and auxiliary activity (AA), respectively. The average number of CAZyme by lifestyle was 713, 553, and 397 for DSE, plant pathogen, and saprobe, respectively. Arthrinium spp. had 127–134 of PCWDEs, while DSE, plant pathogen, and saprobe had an average of 168, 130, and 65 PCWDEs, respectively. In the NMDS plot generated using the CAZyme profile of these species, Arthrinium-DSE-plant pathogen was clustered together, and the saprobe was widely distributed at a distance from this cluster. Based on the permutational multivariate analysis of variance, significant dissimilarities were observed between clusters according to lifestyles, and A. koreanum and A. saccharicola, whose lifestyles have not been identified, showed significant dissimilarity only with the saprobe (Figure 2). The NMDS plot generated using the PCWDE profile showed similar results, except that the dissimilarity between the DSE and the plant pathogen cluster was not significant. Further, among PCWDEs, GH53, GH127, PL4, and CE12 were significant predictors in the dimension 2 direction, and the representative enzymes of these CAZyme families are arabinogalactan endo-β-1,4-galactanase (EC 3.2.1.89), non-reducing end β-L-arabinofuranosidase (EC 3.2.1.185), rhamnogalacturonan endolyase (EC 4.2.2.23), and acetylxylan esterase (EC 3.1.1.72), respectively.

The analysis of biosynthetic gene clusters (BGCs) of the four Arthrinium spp. showed that the highest number of BGCs was found in A. arundinis (84), followed by A. koreanum (76), and A. phaeospermum (69) and A. saccharicola (69) (Supplementary Figure 1). The secondary metabolite backbone genes consisted of 23–28 type I polyketide synthase (T1PKS) genes, 16–21 non-ribosomal peptide synthetase (NRPS) genes, 13–18 terpene genes, 0–3 indole genes, and 1–2 type III polyketide synthase (T3PKS) genes. Six BGCs were annotated with 100% similarity in the four Arthrinium spp. NRPS and two T1PKS BGCs, known to produce dimethylcoprogen, alternapyrone and ACR toxin I, respectively, were observed in all four species, and another two T1PKS BGCs, known to produce (R)-mellein and alternariol, were commonly found in A. koreanum and A. arundinis (Fujii et al., 2005; Izumi et al., 2012; Chen et al., 2013; Chooi et al., 2015a, b). Additionally, a T1PKS BGC, known to produce pyranonigrin E, was observed in A. phaeospermum (Andersen et al., 2011).

Arthrinium koreanum KUC21332 and A. saccharicola KUC21221, the gentisyl alcohol-producing species, commonly harbor 6-MSA BGCs (42.3 and 42.5 kb, respectively), which show 40% similarity with patulin BGC, while A. phaeospermum AP-Z13 does not harbor one (Figure 3). The nine genes that they commonly harbor in their BGCs are GsaA-GsaI. GsaH is orthologous to PatK, the backbone gene encoding 6-methylsalisylic acid (6MSA) synthase that catalyzes the first step of patulin biosynthesis (conversion of acetyl-CoA and malonyl-CoA to 6MSA); GsaE is orthologous to PatG encoding 6MSA decarboxylase that catalyzes the second step (conversion of 6MSA to m-cresol); GsaD is orthologous to PatH encoding m-cresol methyl hydroxylase that catalyzes the third step (conversion of m-cresol to m-hydroxybenzyl alcohol); GsaF is orthologous to PatI encoding m-hydroxybenzyl alcohol hydroxylase that catalyzes the fourth step (conversion of m-hydroxybenzyl alcohol to gentisyl alcohol); GsaA is orthologous to PatL encoding C6 transcription factor that activates gene expression; GsaB and GsaG are orthologous to PatO and PatJ encoding isoamyl alcohol oxidase and a putative dioxygenase, respectively, that catalyze the sixth step (conversion of gentisaldehyde to isoepoxydon); and GsaC and GsaI have been predicted to encode major facilitator superfamily (MFS) transporter (77.98% identical to EKG18982.1) and aldehyde reductase (73.39% identical to KAF6798856.1), respectively (Li et al., 2019). A. arundinis NRRL 25634 also harbors a 6-MSA BGC (44.9 kb) similar to patulin BGC (46% similarity), with two additional genes named GsaJ and GsaK, orthologous to PatM and PatC encoding ATP-binding cassette (ABC) transporter and MFS transporter, respectively; PatM and PatC were proposed to be involved in the extracellular patulin secretion (Li et al., 2019). In the other two Arthrinium spp., GsaC may replace the function of PatC, and the function of GsaI in the BGC should be further studied.

By measuring the DPPH radical-scavenging activity of A. koreanum KUC21332 extract using different organic (peptone and sodium glutamate) and inorganic nitrogen compounds (KNO₃ and (NH₄)₂SO₄) as a nitrogen source, the highest activity (90.6–98.3%) was observed with sodium glutamate (Figure 4A). The DPPH radical-scavenging activity was also determined using a monosaccharide (glucose), disaccharides (sucrose and maltose), a polysaccharide (soluble starch), and a sugar alcohol (mannitol) as the carbon source and sodium glutamate as the nitrogen source. As the activities rapidly increased within 2 days in the nitrogen source test, the carbon source test was conducted using 3-day cultures. The highest activity (79.2%) was observed with glucose, which was approximately 25 times higher than the lowest activity (3.2%) observed with mannitol (Figure 4B). To verify whether the Gsa BGC was responsible for gentisyl alcohol production, DEG analysis was performed with the transcriptomes from the mannitol- and glucose-supplemented cultures. The expression of GsaA, GsaC, GsaG, and GsaH was found to be significantly lower in the mannitol-supplemented group compared to that in the glucose-supplemented group. The fold-change values of the four genes in the glucose-/mannitol-supplemented group were 2.03, 27.28, 8.53, and 5.10 (q < 0.001 for all; the Wald test with Benjamini–Hochberg correction), respectively.