Conidiating cultures of E. turcicum previously characterized as race 13N and 23N were supplied by Dr. M. Craven from the Agricultural Research Council—Grain Crops (Potchefstroom, South Africa) as described previously (Craven and Fourie, 2011). Isolates 2 (race 23N) and 103 (race 13N) were collected from the Free State province of South Africa. Maize Va26 seedlings lacking any Ht genes (Leath and Pedersen, 1985) were whorl inoculated at the trifoliar leaf stage with 400 μl of a 9,000 conidia/ml conidial suspension of E. turcicum race 23N and 13N on separate plants. Maize leaves were also painted (using a paint brush) with the conidial suspension of either race to ensure that the maize–fungus interaction would be detected in distal parts of the leaves at early time points. Inoculated seedlings were placed into a dew chamber for 16 h where after the seedlings were transferred to a growth chamber. Conditions in the growth chamber were kept at 22°C day and 18°C night temperatures (±2°C) with a flux of 25–50 (362–650 μE m⁻² s⁻¹).

Samples were harvested by cutting the stem of the seedlings 1 cm above the ground, removing the flag leaf and flash freezing the rest of the plant. Five plants per biological replicate and three biological replicates per time point were collected. Plants were harvested before inoculations (representing 0 days post-inoculation, dpi) and at disease stages representing initial chlorotic flecks (2 dpi), advanced chlorotic flecks (5 dpi), lesions (7 dpi), and mature lesions (13 dpi, Figure 1). An additional sample of maize leaves inoculated with E. turcicum race 23N was collected at 13 dpi, which showed extensive damage to leaf tips and is from here on referred to as the “severe lesion” (SL). The transcriptome of a South African race 13N isolate grown in vitro was also sequenced to allow for comparisons between in planta and in vitro conditions and to identify in planta-specific transcripts, which may be more likely to be involved in pathogenicity. Single conidia of isolate 103 (race 13N) were obtained from diseased leaf material collected during the maize inoculation trial and grown on potato dextrose agar for 2 weeks before material was harvested for RNA sequencing.

RNA was extracted from five technical replicates per biological replicate for in planta and in vitro samples by flash freezing leaf or fungal material (0.1 g) in liquid nitrogen and grinding the samples using a mortar and pestle. Total RNA was extracted from frozen material using the Qiazol lysis reagent (Qiagen, Limburg, Netherlands) according to the manufacturer's instructions. Genomic DNA contamination was removed with RNase-free DNase I (Qiagen) and extracted RNA purified using the RNeasy® Mini Kit (Qiagen) according to the manufacturer's instructions. RNA was eluted into a final volume of 30 μl with nuclease-free water. RNA concentration and purity was estimated with the Nanodrop® 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, United States), and the quality of extracted RNA was visualized with formaldehyde gel electrophoresis (Bryant and Manning, 2000).

Before sequencing, three biological replicates per time point were pooled in 15 μg quantities of each replicate to obtain 45 μg of RNA per time point. The pooled RNA samples were analyzed on the 2100 Bioanalyzer (Agilent, California, United States) to ensure that samples were of adequate quality and quantity for transcriptome sequencing. In total, 12 RNA samples (13N_in vitro, 13N_0dpi, 13N_2dpi, 13N_5dpi, 13N_7dpi, 13N_13dpi, 23N_0dpi, 23N_2dpi, 23N_5dpi, 23N_7dpi, 23N_13dpi, and 23N_SL) were sent to the Beijing Genomics Institute (BGI; Shenzhen; China, RRID:SCR_011114) for library construction and strand-specific sequencing of 90 bp reads on the Illumina HiSeq 2500 platform (Illumina, California, United States, Illumina HiSeq 2500 System, RRID:SCR_016383). The raw RNAseq data have been deposited in the National Center for Biotechnology Information (NCBI) Short Read Archive (accession number PRJNA560644).

Raw Illumina reads were quality trimmed and filtered to remove low-quality reads and adaptors using Trimmomatic v. 0.33 (Bolger et al., 2014) (Trimmomatic, RRID:SCR_011848) with the following settings: “ILLUMINACLIP = Yes, TruSeq3 (paired-end for MiSeq and HiSeq), HEADCROP = 10, MINLEN = 90.” The reference race 23N E. turcicum Et28A (Ohm et al., 2012; Condon et al., 2013) and Zea mays B73 RefGen_v3 (http://www.maizegdb.org/assembly/) genomes were concatenated, and filtered reads were aligned to the concatenated genome with Tophat v. 2.0.13 (Trapnell et al., 2009) using the following parameters: “mean inner distance between mate pairs = 170 bp, max realign edit distance = 0, library type = FR first strand.”

Read counts were determined with feature Counts (Liao et al., 2014) (featureCounts, RRID:SCR_012919) using the following command: “featureCounts –primary –p -a Settu1_GeneCatalog_genes_20110305.gff -o outputfile acceptedhits.bam.” The Bioconductor edgeR v. 3.12.1 package (Robinson and Smyth, 2008; Robinson and Oshlack, 2010; Robinson et al., 2010) (edgeR, RRID:SCR_012802) was used to remove genes that had fewer than one read per million mapped in at least one dataset. This package was also applied to normalize data. A multidimensional scatter (MDS) plot was constructed in edgeR using read counts of the genes mapped to the E. turcicum genome to visualize the similarity between datasets (Supplementary Figure 1). Datasets were normalized using the trimmed mean of M values (TMM) method implemented in edgeR (Robinson et al., 2010) (edgeR, RRID:SCR_012802). Heatmaps were drawn in the R package, pheatmap (Kolde, 2015).

A modified small-scale hexadecyltrimethylammonium bromide (CTAB) method described previously (Stewart and Via, 1993) was used to extract DNA from flash-frozen plant material collected for transcriptome sequencing. Fungal material was quantified using an E. turcicum-specific cpr1 gene primer set and normalized relative to the amount of maize material estimated using the gst3 gene primer set (Langenhoven et al., under review). The method developed and validated by Langenhoven et al. (under review) was based on the fungal quantification method developed for Cercospora zeina (Korsman et al., 2012). Quantities of E. turcicum and maize were extrapolated from standard curve graphs, and fungal quantification was determined as nanogram E. turcicum DNA per microgram maize DNA. Tests to detect significant differences were conducted using a one-way ANOVA and the Tukey multiple pairwise comparison at a 95% confidence interval in R (R Core Team, 2017) (R Project for Statistical Computing, RRID:SCR_001905) of log-transformed fungal quantities. The correlation between the percentage of reads mapped and log-transformed fungal quantities were investigated using the Spearman correlation method in R (R Core Team, 2017) (R Project for Statistical Computing, RRID:SCR_001905). A scatter plot was constructed to visualize the correlation between data using the “ggpubr” v 0.2 package in R (R Core Team, 2017) (R Project for Statistical Computing, RRID:SCR_001905).

Annotations of the E. turcicum Et28A genome (race 23N) based on gene ontology (GO), InterProScan domains, Kyoto Encyclopedia of Genes and Genomes (KEGG), and EuKaryotic Orthologous Groups (KOG) were conducted previously (Ohm et al., 2012; Condon et al., 2013) and are available for download (http://genome.jgi.doe.gov/Settu1/Settu1.home.html). Genes predicted to be secreted as well as those annotated as carbohydrate-active enzymes (CAZymes) or involved in secondary metabolite biosynthesis are also available for download from the same source. The R packages GSEABase (Morgan et al., 2017) and GOStats (Falcon and Gentleman, 2007) were used to identify fungal overrepresented GO terms using the standard hypergeometric test at a significance level of 0.05. Overrepresented GO terms were subsequently grouped into high-level summaries using the online tool GOSlimViewer (McCarthy et al., 2006) (GOSlimViewer, RRID:SCR_005665). Cell wall degrading enzymes were identified from CAZymes by considering functional annotations and performing BLASTp analysis against the Plant Cell Wall-Degrading Enzyme database (Choi et al., 2013) using the following parameters: expect (e–) value <1 × 10⁻⁵ and percent similarity >40%. The Joint Genome Institute (JGI) database (https://mycocosm.jgi.doe.gov/Settu1/Settu1.home.html) was queried to identify genes involved in β-oxidation of lipids and fatty acids and the glyoxylate cycle.

Secreted proteins were previously identified from the E. turcicum genome (Ohm et al., 2012; Condon et al., 2013) and were downloaded from the Joint Genome Initiative website (http://genome.jgi.doe.gov/Settu1/Settu1.home.html) These protein sequences were investigated to identify candidate secreted effectors based on three categories: (1) protein characteristics, (2) evidence of expression, and (3) genome annotations or similarity to known proteins. Protein characteristics investigated were presence of a signal peptide, absence of a transmembrane domain, protein size, and cysteine content. The presence of a signal peptide was confirmed with SignalP (Petersen et al., 2011) (SignalP, RRID:SCR_015644) and TMHMM v2.0 (Sonnhammer et al., 1998) (TMHMM Server, RRID:SCR_014935) was used to detect transmembrane domains. A protein was labeled as a putative transmembrane protein if it contained more than two transmembrane domains in total, or at least one domain after the first 60 amino acids (Sperschneider et al., 2016). Since the exact size limit on effectors is still unclear, 350 aa was chosen as the limit of candidate effector size, and larger proteins were removed. Proteins containing less than two cysteine residues were also removed.

Proteins were considered to show evidence of expression if they contained read counts values >2 (based on transcriptome sequencing) in at least one dataset. Candidates were investigated for similarities to proteins with known roles in pathogenicity by performing BLASTp analysis on the Pathogen–Host Interactions database (PHI-base, RRID:SCR_003331) (Urban et al., 2017). BLASTp analysis was also performed against the non-redundant NCBI protein database using DIAMOND (Buchfink et al., 2015). Obtained hits from BLASTp searches against PHI-base and NCBI database with an e-value of <1 × 10⁻⁵ and similarity >40% were considered significant. In addition, similarity of E. turcicum proteins to secreted in xylem (SIX) proteins were investigated by querying the NCBI database (NCBI, RRID:SCR_006472) for proteins with the keyword “secreted in xylem SIX” (date search was performed: 18 April 2019). Gene names were manually investigated to identify search results that were not annotated as secreted in xylem. The protein sequences of all available SIX gene sequences were downloaded, and BLASTp analysis was performed against the E. turcicum Et28A proteome with DIAMOND (DIAMOND, RRID:SCR_009457) using an e-value cut-off of 1 × 10⁻⁵ and the parameter “more sensitive” (Buchfink et al., 2015). Heatmaps were constructed in pheatmap (Kolde, 2015) (pheatmap, RRID:SCR_016418), a package in R (R Core Team, 2017) (R Project for Statistical Computing, RRID:SCR_001905).

Races were compared by scoring read count values as 1 (read count value >2) or 0 (read count value <2) across time points 2–13 dpi. Two in planta groups were subsequently created, which consisted of presence/absence per protein for 13N_2dpi, 13N_5dpi, 13N_7dpi, and 13N_13dpi as well as 23N_2dpi, 23N_5dpi, 23N_7dpi, 23N_13dpi, and 23N_SL—from here on referred to as 13N_in planta and 23N_in planta, respectively. Presence/absence was also scored for the 13N_in vitro data. Venn diagrams were constructed to compare expression between the 13N_in vitro data and the 13N_in planta group, as well as between the in planta groups. Proteins with expression in only one group or dataset or shared between groups and the in vitro dataset were submitted for BLASTp analysis against the non-redundant protein database with DIAMOND (Buchfink et al., 2015). Parameters were set to return the 20 most significant hits per query, and an e-value cut-off of 1 × 10⁻⁵ was used. In addition, putative functions of proteins were assigned based on InterProScan domains and KEGG and KOG annotations of the E. turcicum Et28A genome conducted previously (Ohm et al., 2012; Condon et al., 2013).

Complementary DNA (cDNA) was synthesized from RNA using the High Capacity RNA-to-cDNA™ Kit (Thermo Fisher Scientific, Waltham, United States) according to the manufacturer's instructions. Synthesis of cDNA from biological replicates was performed separately and not in a pool as for transcriptome sequencing. Primer sets were designed for reverse transcriptase PCR (RT-PCR) from the E. turcicum Et28A v. 1.0 genome based on open reading frames (Supplementary Table 1). The following conditions were used to amplify fragments of each candidate effector gene from cDNA as templates for standard curves: 12.5 μl Amplicon Taq 1.1 Master Mix, 1.6 μM of each primer, and 30 ng DNA in a final volume of 25 μl. The initial denaturation step was at 95°C for 3 min, followed by 30 cycles of 95°C for 15 s, 61°C for 15 s, and 72°C for 15 s, with a final extension step of 72°C for 40 min. Amplicons were purified using Sephadex G50® columns.

Reference and target gene amplicons were cloned into the pJET1.2/blunt vector to use as DNA templates for standard curves for RT quantitative PCR (RT-qPCR). Ligation reactions were performed using the CloneJET PCR Cloning Kit (Thermo Fischer Scientific, Waltham, United States) according to the manufacturer's instructions. Purified amplicons were cloned into Escherichia coli DH5α using the heat shock method (Froger and Hall, 2007). Competent cells were prepared using the calcium chloride protocol (Holsters et al., 1978). Putative transformants were investigated for the presence of the insert in a colony PCR using 6.25 μl Amplicon Taq 1.1 Master Mix and 1.6 μM of each M13 primer in a total volume of 12.5 μl. Cycling conditions were the same as described above. Transformants containing the insert were cultured and plasmid DNA extracted using the Zyppy™ Plasmid Miniprep Kit (Zymo Research, Irvine, United States) as per the manufacturer's instructions.

Reverse transcriptase quantitative PCR reactions were performed for E. turcicum candidate effector genes and three reference genes, namely, 40S ribosomal protein (40S, ProtID 168532), elongation factor 1-α (EF1-α, ProtID 36922), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH, ProtID 183842) using the Bio-Rad CFX96 Touch™ Real-Time PCR Detection System. Reactions were set up in a total volume of 11 μl consisting of 5 μl RealQ Plus 2x Master Mix Green without ROX™ (Amplicon, Brighton, United Kingdom), 0.5 μM of each primer, 1 μl cDNA template, and sterile, distilled water. The cycling conditions were as follows: 95°C for 10 min, 45 cycles of 95°C for 10 s, 60°C for 15 s, and 72°C for 10 s. A standard curve was constructed for each gene (from starting concentrations ranging from 250.2 to 712 ng/μl) using serial dilutions of plasmids at 1 × 10⁻²-1 × 10⁻⁶.

The melt curve analysis of each gene was performed in the software package Bio-Rad CFX Manager™ to ensure that only a single cDNA product was produced and that primer dimers were absent. A product from each RT-qPCR experiment was purified (using the Sephadex® G50 protocol) and sequenced using 1 μl BigDye, 4 μl purified product, and 1.6 μM primer in a volume of 10 μl. Products were purified using the Sephadex® G50 spin columns and submitted for Sanger sequencing to confirm that the correct fragment was produced. RT-qPCR results were analyzed in qBASE^PLUS (Biogazelle, Zwijnaarde, Belgium).

Significant differences in gene expression between datasets were detected using a one-way ANOVA (ANOVA, RRID:SCR_002427), and the Tukey multiple pairwise comparison was conducted to identify significant differences between datasets at a 95% confidence interval. Analyses were conducted in R (R Core Team, 2017) (R Project for Statistical Computing, RRID:SCR_001905) using log-transformed calibrated normalized relative quantity (CNRQ) values. A constant value of 1 was added to all values to obtain positive log values. The null hypothesis tested was that no significant differences existed between datasets. At least two biological replicates were available for analyses, except for the 0 dpi time point, as no fungal transcripts were detected and this point was removed from the ANOVA (ANOVA, RRID:SCR_002427).

Twenty E. turcicum isolates from a previous population genetic study of E. turcicum were selected for sequencing based on genetic differences (Nieuwoudt et al., 2018). Five isolates were selected from each host (maize and sorghum) and location (Delmas and Greytown, South Africa) to represent a diverse set of isolates. DNA was extracted from pure cultures after 4–7 days growth using the Zymo Research Fungal/Bacterial DNA extraction kit (Zymo Research, Irvine, United States) as per the manufacturer's instructions, with extension of the vortex time (45 min rather than the 5 min as suggested). Sequences of each candidate effector from the sequenced E. turcicum Et28A v. 1.0 (race 23N, http://genome.jgi.doe.gov/Settu1/Settu1.home.html) and NY001 v. 2.0 (race 1, https://mycocosm.jgi.doe.gov/Settur3/Settur3.info.html) genomes were included to compare sequence variation among isolates from different continents. Candidate effector sequences from an E. turcicum isolate (Et73), which has been shown to be specific to sorghum, were also included (Langenhoven et al., under review).

Primers were designed to amplify the full sequences as well as flanking regions of two candidate effectors (Supplementary Table 1) based on the genome sequence of E. turcicum Et28A v.1.0 (Ohm et al., 2012; Condon et al., 2013). Conditions to amplify candidate effector gene sequences were as follows: 12.5 μl Amplicon Taq 1.1 Master Mix, 0.48 μM of each primer, and 30 ng DNA in a final volume of 25 μl using the same cycling conditions as described above. Amplicons were purified using the Sephadex® G50 spin columns and sequenced. Sequences were aligned using MUSCLE (Edgar, 2004) (MUSCLE, RRID:SCR_011812) in MEGA v.7.0.26 (Kumar et al., 2016) (MEGA Software, RRID:SCR_000667).

Phylogenetic trees depicting evolutionary relationships among the amino acid sequences of the candidate effectors sequenced during this study were inferred by maximum likelihood using the package PHANGORN in R v. 3.4.0 (Schliep, 2011). The optimal model of amino acid substitution was inferred using the “model test” function in the same package. The phylogenetic trees were drawn using 1,000 bootstraps, the optimal model of amino acid substitution and nearest-neighbor interchange to improve the likelihood of the tree. The haplotype network was drawn from nucleotide sequences using the median-joining method and an epsilon value of 0 in POPART v. 1.7 (Leigh and Bryant, 2015). Tajima's D test was performed in pegas, a package in R (Paradis, 2010), to determine if candidate effectors are undergoing positive selection.

We applied dual RNA sequencing of pooled biological replicates to conduct genome-wide expression profiling of E. turcicum during infection of maize at five time points. Light, chlorotic flecks were observed at 2 dpi, which became more numerous at 5 dpi (Figures 1A,B). Tan-colored lesions were observed at 7 dpi, which enlarged and became gray in color at 13 dpi. At 13 dpi, maize seedlings displayed lesions without damage to leaf tips as well lesions with significant damage to leaf tips (referred to as severe lesions). Similar symptoms were observed for both races, as expected, since susceptible maize line Va26 lacks Ht genes (Figures 1A,B). For each time point, 15–17.5 million paired-end reads were produced (Table 1). In addition, 15.4 million paired-end RNAseq reads were generated from an in vitro grown race 13N E. turcicum culture (Table 1). Expression profiles of the 11,702 genes predicted from the E. turcicum Et28A genome (Ohm et al., 2012; Condon et al., 2013) revealed evidence of expression for 70% of genes in vitro (normalized expression values >2) and 86.8% during Z. mays infection (Supplementary Table 2). A low percentage of reads that mapped to the E. turcicum genome were detected before inoculation in the 13N_0dpi (0.0015%) and 23N_0dpi (0.0019%) datasets. The possibility that a small number of maize genes may have mapped to the fungal genome, or that reads mapped to endophytes present in maize, cannot be excluded. However, we do not believe that these results have a significant impact on our main findings as the RNAseq data was used for gene discovery and not differential expression. In this study, a dataset was considered to be either the read count values for the in vitro grown isolate or values for isolates of race 13N or 23N at a particular in planta time point.

The percentage of transcripts mapped at each time point showed the same trend (low at 0–7 dpi, increasing at 13 dpi) for both races. E. turcicum genomic DNA (gDNA) quantity measured by qPCR (as a proxy for fungal biomass) was significantly higher at 13 dpi as compared to 0–7dpi with no differences detected between 13N_13dpi and 23N_13dpi (Figure 1C). There was a significant positive correlation between the percentage of reads mapped and log-transformed fungal gDNA content (Figure 1D). These results indicate extensive colonization of maize leaves by E. turcicum between days 7 and 13.

Several lines of evidence were followed to identify E. turcicum secreted proteins that could be candidate effectors with a role in pathogenicity, namely (i) protein characteristics, (ii) sequence similarity to known effectors, (iii) virulence function shown for a similar protein in another fungal phytopathogen, and (iv) expression in planta. A total of 1,388 secreted proteins previously identified from the race 23N E. turcicum Et28A genome were downloaded and queried to identify candidate effectors (Ohm et al., 2012; Condon et al., 2013). Of these, 1,186 had more than 1 read per million mapped in at least one dataset in our study. A total of 351 proteins met the criteria to be classified as candidate effectors, of which 346 showed evidence of expression (read count value >2) in at least one dataset (Supplementary Table 7).

Similarity searches of candidate effectors against PHI-BLAST revealed significant hits to previously characterized effectors or effector candidates (8 hits) or to proteins with an increased (3), reduced (15), or mixed effect (3) on virulence when knocked out (Table 2). Known effectors included Ecp6 (ProtID 136414), which was previously characterized in E. turcicum (Xue et al., 2013). One protein (ProtID 30080) had a significant hit to the secreted in xylem 5 (SIX5) effector from Fusarium oxysporum. Another protein (ProtID 20746) exhibited similarity to the XEG1 protein from Phytophthora sojae. Three proteins (ProtIDs 154392, 164382, and 28054) had significant hits to the MoCDIP4 effector from M. oryzae. A hit to BEC1019 (ProtID 91360), a candidate effector from Blumeria graminis, was identified. A protein (ProtID 30084) similar to the secreted lipase effector FGL1 from Fusarium graminearum (Blümke et al., 2014) was detected.

Candidate effectors identified with similarity to proteins that resulted in reduced virulence when knocked out included cell wall degrading enzymes and proteins involved in appressorial penetration. Interestingly, a putative necrosis-and-ethylene inducing precursor protein was identified (ProtID 41216). A protein with similarity to Sfp-type 4′-phosphopantetheinyl (ProtID 177420) transferase was identified. Three significant hits were detected to proteins that resulted in increased virulence when overexpressed and included a cutinase from Monolinia fructicola (ProtID 85317), a heat shock protein from Saccharomyces cerevisiae (ProtID 88127) and SP1 from Parastaganospora nodorum, which shows high sequence homology to cerato-platanin (ProtID 164814), a phytotoxic protein of Ceratocystis fimbriata f. sp. plantani.

Homology searches against the NCBI database revealed eight additional candidates with putative annotations related to pathogenicity (Table 3). Four significant hits to known effectors were identified: ProtID 34559 showed similarity to the secreted in xylem 13 (SIX13) protein from F. oxysporum, ProtID 29144 was similar to the celpoo28 effector-like protein, and two proteins (ProtIDs 174473 and 184152) were identified with significant similarity to the biotrophy-associated protein 2 (BAS2). One cell death inducing protein (ProtID 25241) and the hypersensitive response inducing protein 1 from Alternaria alternata (ProtID 164162) was identified. One significant hit was identified as a chitin-binding protein of the Dothideomycete tomato pathogen, S. lycopersici (ProtID 135655). A significant hit to the PR1-like protein was obtained against A. alternata (ProtID 177800).

An additional seven candidates contained annotations indicative of pathogenicity (Supplementary Table 7). Two proteins (ProtIDs 165307 and 165528) were annotated as containing “common in several fungal extracellular membrane proteins” (CFEM) domains. A protein (ProtID 34628) was identified, which contains a peptidoglycan-binding LysM domain, and another protein (ProtID 166607) had a chitin-binding domain. Three proteins with peptidase activity were identified, of which two are annotated as metalloproteases (ProtID 23005 and 30401) and one as a serine peptidase with trypsin activity (ProtID 93425).

A total of 558 proteins representing SIX1–SIX14 from different fungi were identified in the NCBI database by a keyword search (Supplementary Table 8). BLASTp analysis of these protein sequences against E. turcicum proteome yielded a total of 40 significant hits (Supplementary Table 8). Four E. turcicum proteins (ProtIDs 34559, 30080, 18972, and 24515, Supplementary Table 8) matched multiple proteins previously characterized as SIX13 or SIX5 from F. oxysporum, Colletotrichum fructicola, and Neofusicoccum parvum. The E. turcicum proteins most similar to SIX13 and SIX5 were ProtID 34559 and ProtID 30080, respectively (Supplementary Table 8). Although ProtID 34559 was the top hit to F. oxysporum SIX13, it had a low percentage identity and no functional annotation. Therefore, to gain further evidence that it might be a SIX13 ortholog, reciprocal BLASTp analysis was performed against the NCBI non-redundant protein database using the F. oxysporum f. sp. cubense SIX13 amino acid sequence (GenBank accession number ALQ80840.1) as input. Among the top 30 results were proteins previously identified as SIX 13 proteins from F. oxysporum as well as ProtID 34559 from E. turcicum (query coverage = 84%, e-value = 2 × 10⁻¹⁶, identity = 27%). Known SIX13 effectors identified from the NCBI database were aligned with E. turcicum Et28A ProtID 34559 (Supplementary Figure 3). Owing to the low percentage similarities of these proteins to the F. oxysporum sequences, these proteins were renamed as SIX13-like and SIX5-like.

A gene expression heatmap was generated of candidate E. turcicum effectors from this study that were annotated based on matches to E. turcicum Et28A secreted proteins, effectors in PHI-base, the NCBI database, or literature (Figure 6). The trend observed for all datasets was highest read count values for transcripts at 13 dpi, which is possibly due to the higher fungal reads mapped at this time point. The candidates with the highest expression values at 13 dpi were MoCDIP4 (ProtID 164382 and 154392), cerato-platanin (ProtID 164814), and the candidate annotated with trypsin activity (ProtID 93425). The candidate effectors showing the highest read counts at 2, 5, and 7 dpi included Ecp6, cerato-platanin, and ProtID 162716. The latter protein did not have a functional annotation, but it was also expressed earlier (at 2 dpi) in race 13N than race 23N (Figure 6). Ecp6 (ProtID 136414) and cerato-platanin were also expressed in vitro. Interestingly, AVRHt1 showed no evidence of expression in the 13N_in vitro or in the 13N_in planta datasets but was detected in the 23N_in planta datasets. SIX5-like (ProtID 30080) was expressed at the latest time points in planta, whereas SIX13-like (ProtID 34559) showed expression at all time points in both races (Figure 6).

Comparison of genes expressed in race 13N (all in planta time points vs. in vitro), as well as race 13N and 23N (all in planta time points) was carried out to identify race- or in planta-specific gene expression (Figure 7). Comparisons of the 13N in planta and in vitro conditions revealed that the majority of genes were shared between the datasets (8,173, Figure 7A, Supplementary Table 9). Of these, 200 were putatively identified as effectors. A total of 1,911 showed evidence of expression in the 13N_in planta group but not in vitro and included 136 putative effectors, which included five known effectors (SIX13-like, SIX5-like, MoCDIP4, XEG1, and CELP0028).

A total of 10,080 genes showed evidence of expression in both the 13N_in planta and 23N_in planta groups (Figure 7B, Supplementary Table 9). Of these, 336 were classified as putative effectors, of which 11 were similar to known effectors (including Ecp6). Seven genes were uniquely expressed in the 13N_in planta group, none of which were putatively identified as effectors or involved in pathogenicity. A total of 194 genes showed expression in the 23N_in vitro group only and included genes possibly involved in pathogenesis, cell wall hydrolysis, and secondary metabolite biosynthesis, as well as genes encoding transporter proteins, a hard surface induced and a defense-related protein (Supplementary Table 9).

Ten candidate effectors were uniquely expressed in the race 23N_in planta group and included AVRHt1. Mideros et al. (2018) reported a non-synonymous mutation in the AVRHt1 (ProtID 179218), which distinguished race 1 and 23N isolates. Race 1 isolates encoded a “T” nucleotide and race 23N isolates a “C” allele. Transcriptome sequences generated during this study for each dataset revealed that the race 23N E. turcicum isolate encoded a “C” allele at the reported genomic location (scaffold 2: 3,549,698).

The conserved effector Ecp6 and candidate effector SIX13-like were chosen for expression validation. The RNAseq data had shown that Ecp6 was expressed in vitro and in planta. Candidate effector SIX13-like was not expressed in vitro in race 13N and was among the most highly expressed transcripts, with the highest expression at 2 dpi in race 23N and 5 and 7 dpi in both races.

Significant differences in expression were detected between time points for Ecp6 and SIX13-like (Figure 8). In vitro expression in race 13N was low for both candidate effectors. The expression of both Ecp6 and SIX13-like peaked at 5 and 7 dpi as compared to 2 and 13 dpi. Ecp6 expression was significantly greater at these mid-time points compared to the outer time points for race 23N, whereas SIX13-like expression was significantly greater at 5 dpi in 13N and 7 dpi in 23N compared to the outer time points in the corresponding race. The pattern of expression of each gene did not differ significantly between the two races. The identity of RT-qPCR products produced were confirmed with sequencing (Supplementary Figure 4).

Two effector genes, SIX13-like and SIX5-like, were sequenced to identify host-specific differences among E. turcicum from maize and sorghum. A total of 22 polymorphisms were detected in the sequences of SIX13-like among the 20 E. turcicum isolates from maize and sorghum sequenced during this study, a sorghum-specific isolate (Et73, Langenhoven et al., under review), as well as genome sequences of Et28A and NY001 (Figure 9A). All SNPs detected in exon regions of SIX13-like resulted in non-synonymous amino acid changes, and no premature stop codons were detected. A SNP was detected in SIX5-like that resulted in a non-synonymous amino acid change (Figure 9B).

Alignment of the SIX13-like amino acid sequences showed that maize isolates were distinct from sorghum isolates with particular amino acids present in all or most isolates from a host (Supplementary Figure 5). This host specificity was borne out by maximum likelihood analysis that produced a phylogram with distinct maize and sorghum clades (Figure 10A). The optimal model of amino acid substitution used for this analysis was FLU, as determined by the lowest Akaike and Bayesian Information Criteria and highest log likelihood. The E. turcicum maize isolates from the United States (ET28A and NY001) formed a subclade of the South African maize isolates. The collection site in South Africa (Delmas or Greytown) did not influence the grouping of isolates (Figure 10A). The distinction between hosts was also visible from the haplotype network (Figure 10B) and showed a greater haplotype diversity among isolates from sorghum than maize. Tajima's D test was performed to determine if candidate effector SIX13-like is undergoing positive selection. Results indicated that the observed mutation rate was not significantly different from the null hypothesis of neutral selection.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.