Similarities among isolates

All Fv isolates clustered within a single clade when either the BUSCO or the Effector genes were used to produce phylogenetic trees (Figs 1, S1 and S2). The phylogenetic tree derived from effectors had a significantly lower resolution among Fv isolates when compared to the BUSCO tree, which was due to the high degree of homology and conservation between effector profiles and effector sequences as well as the number of genes from which the trees were derived (786 vs. 4494). The effector-based and BUSCO-based phylogenetic trees are compared directly in the tanglegram presented in Fig 1. The tanglegram confirmed that the relationships determined from putative effector profiles mirror those of the BUSCO profiles in their ability to differentiate isolates at a species level, but they do not provide sufficient resolution for clear differentiation within species-clades. There were no disagreements between what isolates belonged in which clades and no differences in the classification of out-species clades. The lack of resolution between Fv isolates in the effector tree is evident given the bootstrap confidence estimates. Many isolates contained identical effector profiles, so arrangement of those isolates was done in alphabetical order. The same was true for BUSCO alignments, however fewer isolates were identical, and this provided greater resolution in the classification of isolates. We additionally compared the average nucleotide identity (ANI) across all isolates in an all-by-all manner. This provided us with similarity percentages across the isolates; for example, the F. circinatum isolate was, on average, 88.4% similar to the Fv clade (S3 Table). Two heatmaps were drawn from the ANI data, the first included all the isolates while the second included the 43 isolates included in the Fv clade so that a better differentiation could be made between highly similar isolates (S3 and S4 Figs).

Relationships between toxin content and disease severity

Fumonisin concentrations were plotted and used to estimate correlations with disease severity. Disease severity scores (infected kernel counts) are displayed in S5A and S5B Fig, and though slightly skewed to the left, the distributions were nearly normal (Shapiro-Wilk normality test p-value = 0.3625). Correlations were first estimated between each individual ear’s fumonisin PPM and that ear’s infected kernel number (Fig 2A). The positive correlation of r = 0.38 indicated that 14.52% of the variation in fumonisin concentration was explained by the number of infected kernels. There were two clusters of fumonisin concentrations in the scatterplot, with the first identifying infections that had zero fumonisin contamination (X = 0) and a range in the number of kernels infected. The second cluster identifies ears with fumonisin concentrations around 0.6 ppm^1/5 (0.07 ppm), which appeared to be a common contamination level for the inoculation and incubation time. This correlation was similar to the Spearman’s correlation coefficient of 0.41 for the same data, done to adjust for non-normality. The scatterplot in Fig 2B used the average fumonisin PPM and the average infected kernel number by isolate. Two correlations were calculated for this dataset, where the first was a linear correlation (r = 0.415) and the second was from a logarithmic regression (r = 0.515). These were again similar to the Spearman’s correlation coefficient of 0.46.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Correlations between fumonisin PPM and infected kernel counts.

A. Scatterplot displaying estimates for all ears measured for fumonisin concentration, with fumonisin PPM^1/5 presented on the X-axis and the same ear’s infected kernel count number is on the Y-axis. B. Scatterplot displaying average fumonisin PPM and average infected kernel number for each Fusarium isolate. The line of best fit equations and R-squared values are displayed on each plot, with Fig 2B providing two regressions and equations for the linear (right) and logarithmic (left) relationships.

https://doi.org/10.1371/journal.pone.0306144.g002

In addition to fumonisin concentration measurements, sequences of the 16 genes involved in the FUM biosynthetic gene cluster were identified in the isolates using BLAST to determine the presence, absence, or similarity of each gene for all isolates. In S6A Fig, the heatmap shows the similarity of each gene, where the darker brown colors indicate a greater similarity between the isolate’s gene sequence and the gene from the Fv reference. Isolates Fpro1B7, FV32964, and FV32966 (thought to be F. proliferatum), showed, on average, an 81% similarity in genes that were present and the absence of 3 genes (FUM8B, FUM11, and FUM18). Four isolates lacked all genes in the FUM cluster, which was expected because no previous studies had detected fumonisin production in those species (F. subglutinans (FV62720 included), F. circinatum, and F. temperatum). Two isolates (FV32968 and FV-FL-02) contained only two FUM genes (at 96% identity) and lacked the other 14 genes. In S6B Fig, a scatterplot displays the correlation between the disease severity and the average similarity percentage of the FUM cluster. While the correlation was r = 0.621, isolates containing all the genes in FUM showed a large range in the disease severity estimates derived from kernel counts.

Comparisons of genome size and composition

As previous studies have shown significant genome size differences among Fv isolates and have proposed the presence of a twelfth chromosome [49,50], further investigation was undertaken using two different Fv reference genomes differing in that regard. All the Fv isolates were scaffolded using the two reference genomes independently, genes were predicted, and RagTag alignment statistics were summarized (S7 Table). The first reference genome (RefSeq sample Fv7600) has a genome size of 41.8 Mb, 11 chromosomes, and 36 scaffolds. The second reference genome assembly (Fv10027_ITA) was 44.6 Mb consisting of 20 contigs. In Fig 3A, a Circos plot is shown with Fv10027 as the outermost ring with Fv7600 aligned to it. Several regions are highlighted to show the absence of these contigs in Fv7600. When the order is reversed (S7 Fig) no such gaps are seen in the alignment, suggesting that these regions are present in Fv10027 but not in Fv7600. BUSCO genes were again predicted for all isolates after scaffolding with both reference genomes independently to see if a significant difference in the absent regions existed. The average number of BUSCO genes increased by one (range 0–2) when aligned to Fv10027 compared to Fv7600 and decreased by one (range 0–3) when BUSCO genes were predicted without scaffolding (S1 Table). Comparing scaffolding statistics produced by RagTag, isolates scaffolded using reference Fv10027 placed an additional 100–1400 Kb when compared to Fv7600 while scaffolding using Fv7600 reduced the number of gap base pairs (bps) from 1200–23000 when compared to Fv10027. Because the increase in placed bps was significant when using Fv10027 as the reference, we predicted genes and repeating elements from the extra regions to determine if they harbored novel functions. Fig 3B provides a heatmap of all isolates’ normalized alignment scores for each of the regions that were absent in Fv7600 with contig names and lengths included using the Fv10027 assembly. Contig WAJQ01000013.1 was the only extra contig present in all isolates, with normalized alignment scores ranging from 24% to 85%. Augustus predicted 299 genes from the Fv10027 regions missing in Fv7600, but when these regions were BLASTed against the Fv7600 Augustus gene prediction, every single gene was accounted for at a 93% or higher identity. This confirmed that while the alignment of Fv7600 had a significant number of missing regions, the genes were either copies of genes in other chromosomes or they were simply transposed. Transposable element (TE) prediction of the extra contigs was completed with the expectation that a greater number of TEs would be present if those regions were lineage specific or supernumerary chromosomes. The regions were shown to be nearly identical, if not a slightly lower TE percentage in the core chromosomes. TE percentages ranged from 0.9% to 1.29% in the extra contigs and ranged from 0.95% to 1.44% in the core genome, changing proportionally. Because no functional differences could be identified, any software requiring a reference genome and any reference to chromosome names were subsequently completed using Fv7600 as the reference to maintain consistency with the literature.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Investigation of large (>10kb) non-chromosomal contigs.

A. Circos plot between reference genomes Fv7600 and Fv10027_ITA. Reference Fv10027_ITA is on the outside, Fv7600 is the interior ring. Blue circles name and indicate contigs present in Fv10027 but are absent in Fv7600. Orange blocks refer to differences in alignment and green blocks are identical alignments. B. Heatmap displaying all Fv isolates’ normalized alignment scores of the extra contigs in the Fv10027_ITA reference. Scores range from 0–100 and are displayed in each respective cell. Reference Fv7600 is also included (“RefSeq7600”).

https://doi.org/10.1371/journal.pone.0306144.g003

Content and distribution of effectors, variants and CAZymes

Effectors were predicted for all isolates to determine their coordinates and to identify relationships between the total number of effectors and virulence estimates. The putative effectors of all isolates within the Fv clade were plotted in S8A Fig and show the distribution and density of effectors across the genome. High effector density was evident on chromosomes 2, 8, and 10 (NC_031676.1, NC_031682.1, NC_031684.1) and disproportionately high effector totals were found on chromosomes 8 and 10. Though both chromosomes are smaller than 3Mb (2.8 and 2.2Mb respectively), they contained the greatest number of effectors (2235 and 2440 respectively). S8B Fig shows a heatmap arranged from the greatest to the smallest number of effectors per isolate. F. subglutinans (Fsub4L1) had the greatest number of effectors (543) and F. temperatum (Ft25622) had the fewest (492), while the total number of unique effectors identified across all isolates was 786. To incorporate disease severity estimates into our comparative genomic framework, the average number of infected kernels per isolate were correlated with the number of effectors in that isolate. The correlation coefficient was nearly zero (r = 0.001), supporting the idea that specific effectors rather than total effector number regulate the severity of FER disease symptoms.

As the total number of effectors per isolate was unrelated to disease severity, we sought methods to narrow down the pool of effectors to identify those that may contribute to disease severity. We used OrthoFinder (illustrated in Fig 4A using OrthoVenn3) to cluster the effectors of the three most severe (FV34715, FV54917, FV-NE-01) and two least severe (Fc25332, Ft25622) isolates. This yielded two new sets of effectors with 55 unique to the least severe isolates and 61 unique to the most severe isolates (Fig 4A). Each new effector set was then correlated with disease severity scores and plotted (Fig 4B & 4C) with significant correlations (r = 0.633) for the effectors present in more severe isolates and for the effectors that were absent in more severe isolates but present in the less severe isolates (r = 0.654) (Fig 4B and 4C). Using the subset of effectors that were present or absent in the most severe isolates, there was significant clustering in the scatterplots with isolates in the Fv clade grouping together regardless of disease severity estimates. Given the lack of resolution within Fv isolates, we selected two isolates for more detailed comparisons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Specific effector orthogroup selection.

A. OrthoVenn3 display of effector orthogroups between five isolates. Effectors from two isolates with low disease severity and three isolates with high disease severity were compared to determine unique effectors. The red circle indicates the number of unique effectors between the highly severe isolates and the purple circle indicates the number of effectors unique to the least severe isolates. B. Scatterplot between effectors from severe isolates and disease severity. The effectors unique to the three most severe isolates are on the Y-axis, and the average number of kernels infected by each isolate in on the X-axis. C. Scatterplot between effectors from the least severe isolates and disease severity. The effectors unique to the two least severe isolates are on the Y-axis, and the average number of kernels infected by each isolate in on the X-axis.

https://doi.org/10.1371/journal.pone.0306144.g004

The two most genetically similar Fv isolates with contrasting disease severity estimates were FV54917 and FV-FL-03. Phylogenetic trees (Fig 1) indicated that these isolates were closely related, with FV54917 providing the second highest disease severity estimate and FV-FL-03 providing one of the lowest (S5A Fig). Of FV54917’s 527 effectors and FV-FL-03’s 533 effectors, only 36 were either present or absent when the opposite was true in the other isolate. The density and distribution of FV54917’s effectors displayed in Fig 5A are slightly different than total isolate effector distribution discussed above (S8A Fig). The effector amino acid sequences searched for in the UniProt protein database identified 29 of the 36 effectors that had either conserved domains or putative functions annotated (Table 1). Of the 491 effectors that the two isolates shared, 79 showed a change in position greater than 100kb, and 12 were located on different chromosomes. The sequences of these 79 effectors were BLASTed to determine if there was any known function, and those annotations are displayed in S5 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Effector distribution and movement between FV54917 and FV-FL-03.

A. Density and distribution plot of effectors from high disease severity isolate FV54917. B. Synteny plot of effectors which had a positional difference greater than 100kb between FV-FL-03 and FV54917. Lines represent individual effectors and their positional changes between the two isolates.

https://doi.org/10.1371/journal.pone.0306144.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Effectors of isolates FV54917 and FV-FL-03 when one is differentially present or absent compared to each other.

Under each isolate, 0 indicates absence while 1 indicates presence of the effector. Gene type or conserved domain of the predicted effector from BLAST results and the reference to that gene are also listed.

https://doi.org/10.1371/journal.pone.0306144.t001

Variants that were differentially present or absent between the two isolates and were predicted to have a “high impact” on gene function were pulled out for each isolate, resulting in 62,958 for FV54917 and 58,439 for FV-FL-03. The genes and the proposed impact of the variant are presented in S6 Table along with the chromosome and position where the variant occurred with a total of 40 protein variants identified in the isolates. The location and density of the unique variants for each isolate are shown in S9A and S9B Fig.

Carbohydrate-active enzymes (CAZymes) were also identified and correlated with disease severity estimations. S10A–S10F Fig show the correlation of disease severity with each CAZyme class, the highest correlation being with the classes “Auxiliary Activities” (r = 0.420) and Carbohydrate esterase (r = 0.393).

Isolate origin and virulence

Finally, the substrate and geographic origins of each isolate were used to investigate the differences in the disease severity scores. Eight isolates lacked information about the original host and seven had no information regarding geographic origin. No patterns were seen between the geographic origin and the virulence of the isolates. Substrate data of each isolate was divided into two groups: maize and not maize. S11B Fig displays a boxplot comparing disease severity between the two original substrate groups and two tests of significance. The T-Test significance value = 0.050 suggested that isolates coming from maize produced significantly larger lesions and or infected more kernels.

Disease severity and toxin production

The species of Fusarium that cause Fusarium Ear Rot (FER) in maize are highly variable in their ability to infect, cause damage, produce toxins, and reduce maize yield. While the primary causal agent of FER is Fusarium verticillioides (Fv), several other Fusarium species are found in infected maize ears. The availability of a diverse set of Fusarium isolates with sequenced genomes provided a means to conduct both intraspecific and interspecific comparative genomic studies for a pathosystem that has been associated with quantitative disease resistance. Because of the extensive genomic homology between Fusarium species, the quantitative nature of the FER pathosystem, the large number of pathogenicity-associated genes, and the availability of disease severity data from controlled inoculations, we present this comparative genomic analysis of 50 Fusarium isolates as a means to identify genetic mechanisms controlling disease severity.

Initially, only four isolates were considered alternative species and were included to ensure differences in genome composition and disease symptom expression for comparisons with Fv isolates. Our preliminary genome assembly and phylogenomic comparisons determined that several isolates differed from their listed species names. Isolate FV62720 is likely to be F. subglutinans, while FV32964 and FV32966 are likely F. proliferatum isolates. Two isolates, FV32968 and FV-FL-02, were placed outside of the confirmed Fv clade and are potentially members of the Fusarium fujikuroi species complex. The pine (Pinus sp.) pathogen F. circinatum (Fc25332) provided a reliable and distinct phylogenetic root as the most distantly related of the out-species that has been shown to infect maize and produce little to no disease severity [41]. Using infected kernel counts, Fc25332 consistently showed the lowest disease severity of all the isolates, confirming previous findings. Although the putative effectors did not provide sufficient information to differentiate Fv isolates to the same degree as the BUSCO-based tree, the effector tree (S2 Fig) did mirror the clades and species-level differentiation was possible with Fc25332 shown to be the most distantly related isolate in both comparisons. This can be seen in the tanglegram (Fig 1), with the alternative species clearly differentiated and the order of the Fv isolates lacking alignment across trees due to insufficient information provided by the sequences from the putative effectors. The identical species level designations show the potential of using the effectorome (all putative effector genes) for the simultaneous identification of species using pathogenicity-related genes that may also allow for the differentiation of virulent and avirulent isolates. The same clade-level differentiation can be seen in the average nucleotide identity (ANI) scores across the isolates and the trees on the ANI heatmaps (S3 and S4 Figs). The ANI scores additionally confirmed the species level differentiation seen in the BUSCO tree and provided a numerical comparison between individual isolates (S3 Table). Isolates FV32968 and FV-FL-02 were 97.5% similar to all true Fv isolates, compared to the average 99.07% of the Fv isolates compared to that clade. Out-species ranged from 88.3% (F. circinatum)– 93.8% (F. temperatum) identity compared to all other isolates.

Phenotyping inoculated ears by counting visually infected kernels provided large quantitative differences in disease severity and differentiation among isolates. Due to the variability in phenotyping methods available for the FER pathosystem and the practical importance of fumonisin contamination in the food supply, we validated relationships between our kernel-based phenotyping method and measurements of fumonisin levels. The correlations between kernel counts and fumonisin measurements for all phenotyped ears yielded a low but positive correlation (r = 0.381). The correlation was slightly higher when averages of infected kernel counts and of the fumonisin measurements were taken (r = 0.415), while a logarithmic regression increased the correlation between disease symptoms and fumonisin content to r = 0.515. While the correlations in the literature range greatly, on average, our correlations are slightly lower than those found in the literature [51–53]. We believe the correlation is insufficient to justify using visual severity scores as a proxy for fumonisin measurements and subsequent breeding efforts if any significant genotypic diversity is present in either population. The impact of the genotypic diversity of both Fusarium isolates and maize varieties should be considered when making these comparisons. Our use of 170 unique maize genotypes specifically chosen to maximize genotypic diversity along with 50 Fusarium isolates, six of which should not have been able to produce any fumonisin at all (S6A Fig), makes direct comparisons infeasible. While these isolates did produce fumonisin levels that were lower than the other 44 Fv isolates on average, some still produced small amounts of FUM where genes in the FUM pathway were absent. This suggests some experimental or external sources of error were present in the ELISA estimated fumonisin concentration data.

Introducing additional sources of variance in the form of diverse maize varieties and Fusarium isolates is expected to increase the observed phenotypic variation. For example, phenotyping ears from a single maize variety with a single Fusarium isolate should produce a more reproducible and less variable estimate of fumonisin content. Conversely, the inclusion of a more disease resistant variety may lead to reduced disease severity while fumonisin production is similar. Likewise, a single Fv isolate may produce a high concentration of fumonisin in a particular maize variety without causing high disease severity. These data, taken with the improved logarithmic correlation, suggest that using infected kernel counts is a problematic and non-linear method of quantifying fumonisin levels. Although unable to be concluded from these data, it is likely that visual estimates of disease severity should not be used to measure fumonisin content when genotypic diversity is present. Reduced scales measuring severity (i.e. 1–10) are also likely to inflate the correlation compared to using a raw scale such as kernel counting which ranged from 1–110.

As expected, all out-species were in the lower half of the virulence scores (S5A Fig), although FV32966 and FV32964 are thought to be the same species as Fpro1B7 (F. proliferatum) and they differed in disease severity considerably, ranking 28^th, 37^th, and 46^th, respectively. This is perhaps explained by the host range of F. proliferatum, which includes a wide range of plant species with isolates developing adaptations that vary in disease severity across different hosts. Evidence contrary to this hypothesis is that Fpro1B7 was isolated from a maize plant while FV32964 was isolated from human blood. Results from the T-test between isolates from maize or non-maize hosts indicated a slightly significant association between isolate host and disease severity (T-test p = 0.05 (S11B Fig)), providing evidence that Fusarium can adapt to their hosts to cause higher disease severity and Fv isolates cause more severe maize infections compared to other Fusarium species that are also found in maize.

In the literature on Fv and other Fusarium species, evidence suggests that a twelfth chromosome is present in some isolates [47,50]. As we had access to 43 Fv genomes (FV-FL-02 and FV32968 were included), we investigated this theory by comparing two different reference genomes, Fv7600 and Fv10027_ITA. Fv10027 was larger by 1.6 Mb and contained 7 extra contigs, 3 of which were over 200 Kb compared to Fv7600. After identifying these extra contigs by scaffolding each reference with the other reference and providing comparisons as Circos plots (Figs 3A and S7), there was sufficient evidence of extra-chromosomal regions in Fv10027 and several of our isolates (Fig 3B). These extra-chromosomal regions have many names (lineage specific, conditionally dispensable, supernumerary, accessory, mini, or B-chromosomes) and are best studied in the forma speciales of F. oxysporum.

In 2010, Ma et. al. investigated these regions, referring to them as lineage-specific (LS) genomic regions [47]. They found that in F. oxysporum f. sp. lycopersici, LS regions occupied four full chromosomes, had a high density of transposons, had genes related to pathogenicity, and could be transferred horizontally to convert non-pathogenic strains into pathogenic strains. Following this methodology, we identified genes and repeat sequences in the extra chromosomal regions and compared them to the core chromosomes to determine if there was evidence to consider these extra contigs as LS genomic regions. Not only did the extra contigs contain a lower average percentage of repetitive sequences than the core chromosomes (1.11% vs 1.23%, respectively), but all the 299 predicted genes in the extra contigs were identified in the Fv7600 reference at an identity of 93% or higher. Also, there was no correlation between the presence of extra contigs and disease severity for each isolate (r = 0.06). Given these data, we do not consider these regions to be similar in function to the LS regions identified in F. oxysporum f. sp. lycopersici. Although they are significant in size, they do not contain any functional genes that are absent in the core chromosomes of Fv7600. Because of these factors and the RefSeq status of Fv7600, we chose to scaffold and use Fv7600 as the reference whenever necessary.

Comparisons of Fv genome composition

One group of phytopathogen genes known to regulate disease severity are effectors. Effectors can act as virulence or avirulence factors depending on the current state of the molecular arms race between host and pathogen. The “zig zag model” describes this phenomenon well, where pathogen effectors that are unrecognized by the host can contribute to disease, but effectors recognized by the host (then termed avirulence genes) can trigger the defense response characteristic of effector triggered immunity (ETI) [54]. Hemibiotrophs, such as Fusarium spp., evade or repress the plant immune system before deploying effectors when they switch to a pathogenic lifestyle [55,56]. Because effectors can act as pathogenicity factors, avirulence factors, or neither, we investigated the correlation between groups of effectors and disease severity. Putative effectors’ density and distribution across the Fv genome (Fig 5A for FV54917 and S8A Fig) showed chromosomes 8 and 10 contained the most predicted effectors while being two of the shortest. No correlation was identified between effector number and disease severity (r = .001). This was the expected result, showing that effectors do not act additively to cause disease severity. Identifying effectors that may actively contribute to disease severity by clustering effector groups from isolates with contrasting virulence levels revealed that 301 of the predicted effectors were shared across the chosen isolates, 61 to be present in only the most severe isolates, and 55 in the least severe isolates. This search for effectors provided further evidence that FER is highly quantitative with many alleles contributing variable degrees to disease severity. Of the effectors that were similar to a reference protein, several were of particular interest due to their known roles in Fusarium pathogenicity: pectate lyase and acetylexylan esterase in the high-severity effector group; cutinase and glucuronoxylanase in the low-severity effector group [57–60].

Though these methods led to a more specific list of potentially significant effectors, using out-species isolates rather than low-severity Fv isolates had drawbacks. First, the effectors unique to the out-species isolates do not necessarily have direct implications about the effectorome of Fv. Fortunately, using two different out-species limits the number of non-maize-specific effectors. For example, effectors unique to F. circinatum that may contribute to disease on Pinus spp. should not be present in F. temperatum (the other out-species and does not infect Pinus) and the identified effectors are likely conserved across additional Fusarium spp. A second drawback is that because no Fv isolates representing low disease severity were included, narrowing the list of their effectors was limited. This was done because when multiple low-severity Fv isolates were used there were either no unique effectors, or they were inconsistent. To address this, we narrowed the comparison to two closely related isolates with contrasting levels of virulence.

We chose the contrasting isolates FV54917 and FV-FL-03, ranking 2^nd and 43^rd in disease severity, respectively, to identify effector and variant differences that may be associated with FER severity. We chose these isolates based on their close phylogenomic relationship (as seen in the BUSCO-based tree in S1 Fig) and distinct disease severity scores. Table 1 displays the effectors that are unique to each isolate with a reference gene or conserved domain. FV54917 had 8 unique effectors with a database reference and FV-FL-03 had 21. Several of the proteins encoded from the effectors have previously been identified as pathogenicity related proteins such as cutinase, chitin-binding proteins, glucanase, xylanase, and other conserved domains [59,61]. Notable too were the effectors identified using both this approach and OrthoVenn3/OrthoFinder. A cell wall integrity and stress response component (WSC) domain-containing protein was present the high-severity group (Table 1) and absent in FV-FL-03 (S1 Table), while a pectate lyase A protein was present in the low-severity group and absent from FV54917. Limited literature exists on WSC domains in phytopathology; however, it has been shown to act as a substrate anchor for the enzyme in Pyricularia oryzae [62]. Pectate lyases, conversely, are well known pathogenicity factors for multiple fungal pathogens including Fusarium [57,63]. We also compared the density and distribution of these two isolate’s variants (SNPs or indels) called against the Fv reference genome (S9A and S9B Fig) to determine if there were significant differences. The most obvious difference was the number of variants in chromosome 5, with FV54917 having 9990 variants and FV-FL-03 having only 3928. We selected a subset of the variants if they were unique between the two isolates and their effects on protein composition were rated as “high” by SNPeff. S6 Table contains a record of the position, variant effect(s), protein type targeted by the variant, and that protein’s reference for all high effect variants for both isolates. Of note were multiple splice sites identified in a pectate lyase copy in FV54917. This mutation of pectate lyase, taken in conjunction with the identification of pectate lyase in low-severity isolates and its absence in FV54917 provides supports its role as an avirulence gene.

Various pectate lyase genes were again identified when determining the 79 effectors that were present in both FV54917 and FV-FL-03 (with at least 95% identity) but with a positional difference of more than 100 Kb (Fig 5B and S5 Table). Twelve of the effectors, including a pectate lyase, were positioned in entirely different chromosomes in the two isolates. Control over transcription is more likely to vary when genes are in such different positions or have additional copies in other locations. This strategy is well known in Fusarium pathogenomics, where the high concentration of transposable elements allows for rapid gene diversification without the need for sexual recombination [64,65]. Theoretically, this is particularly impactful in quantitative diseases such as FER, where accumulating multiple minor changes in the expression and the sequences of non-host-specific pathogenicity genes make the pathogen more resilient and could provide a competitive advantage over other microbes.

Obtaining and sequencing Fusarium isolates

The 50 isolates consisted of 46 isolates believed to be Fv, and one isolate each of F. circinatum, F. temperatum, F. proliferatum, and F. subglutinans. Each of these “out-species” has been isolated from infected maize in previous studies. The isolates were primarily sourced from the USDA-ARS Culture Collection (NRRL) and others were donated from across the United States by external researchers or were isolated from maize plants locally. Isolate names were retained for all NRRL isolates. Isolates were cultured on potato dextrose agar (PDA) or isolated directly from plant material when collected from the field. To ensure a single genotype was represented, a single spore was extracted from each isolate and grown on a PDA plate before storing at -80°C. Before whole genome sequencing, a PCR assay was performed on all isolates that were not sourced from the NRRL using primers FV-F2/FV-R from Faria et al (2011) [66] to confirm their identity as Fv. Isolates were then grown on PDA plates for 14-days before the mycelial mat was scraped off, placed in a 2ml tube, and lyophilized. Lyophilized tissue was then placed into a 96 well plate and sent to Rapid Genomics LLC, Gainesville FL, for extractions and Illumina pair-end whole genome sequencing. Isolates were obtained under the APHIS permit number P526P-21-07097.

De novo assembly, alignment, annotation, and gene/protein prediction

Unpaired FASTQ files were initially trimmed and assessed for quality using Trimmomatic and FastQC respectively [67,68]. Trimmed reads were assembled de novo using Spades V3.9.0 into FASTA files [69]. Using RagTag, assemblies were scaffolded and patched using two reference genomes independently, Fusarium verticillioides 7600 (GCF_000149555.1) and Fv10027_ITA (GCA_020882315.1) for all Fv isolates [70]. Other species were scaffolded to previously assembled genomes of their respective species (S1 Table). Next, BUSCO v4.0.6 was used for BUSCO gene identification in order to assess genome completeness using the “hypocreales_odb10” dataset [71]. Based on low quality genome completeness for some isolates, phylogenetic trees created using the Basic Local Alignment Sequencing Tool (BLAST) to align BUSCO genes were used to identify suspect isolates. Five isolates initially identified as Fv were subsequently identified as other species and were re-assembled, re-aligned, and genes were re-identified using the reference genome that was closest to their species (S1 Table). Augustus V3.4.0 was then used for gene prediction of all genomes with “species = Fusarium” [72].

To obtain additional assembly statistics, Bowtie2 was used to align reads into contigs and chromosomes in SAM and BAM formats and samtools was used to view those statistics (S1 Table) [73]. Freebayes v6.1 was used to create Variant Call Files (VCF) using the Fv reference for all isolates so that variants could be compared across all isolates [74]. Variants were filtered using VCFtools, the Ensembl Variant Effect Predictor tool, and SnpEff/SnpSift, where the severity of impact estimations were estimated and then viewed on the Variant Effect Predictor (VEP) website [75–77]. Alignment differences between the references were extracted and investigated using Augustus to predict all the genes in the genomes and BLASTp to determine if any were absent or significantly different. Repeat sequence and transposable element prediction were predicted using RepeatModeler and RepeatMasker [78,79]. Normalized alignment scores were calculated using -1 for misalignments and +1 for correct alignments and then adjusted for sequence length over reference length. Genes predicted by Augustus were then filtered using SignalP v5.0b to identify those with a signal peptide [80]. EffectorP was then run through Python to predict any putative fungal effectors [81]. Protein sequences of genes previously identified in the fumonisin biosynthetic gene cluster, FUM, were downloaded from NCBI (FVEG_00316, 317, 319–329, 14634, and 14635 in the Fv reference genome) and BLASTp was used locally to determine presence, location, and sequence identity in the Fusarium isolates. The putative effectors predicted from EffectorP were run through OrthoFinder v.2.5.2 for clustering and comparisons across all isolates [82]. The heatmap representing effector presence across isolates was created by using the table output of OrthoFinder and converting outputs into a binary presence/absence table. To predict carbohydrate-active enzymes (CAZymes), dbCAN v.4.0.0 was used with the output of SignalP, and a custom Python script was used to extract only the CAZymes that were predicted by all three tools used by dbCAN [83]. Another Python script was used to extract the table from the OrthoFinder output and convert it to a format usable for making a heatmap in R. Quast [84] was used to evaluate the quality of genomes and to compare alignment of the two different reference genomes using Circos [85] Average nucleotide identity scores were generated using pyani [86].

Figures and descriptive analyses

The BUSCO phylogenetic tree was built using IqTree v2.1.3 and a custom python script [87]. The phylogenetic tree of putative effectors was an output of OrthoFinder. Both trees used the least similar Fusarium isolate (F. circinatum or Fc25332) as the root. All heatmaps were drawn in R using the “pheatmap” package [88]. The tanglegram of the two phylogenetic trees was created in R using the “cophylo” function from the “phytools” package [89]. The density and distribution plot of effector locations was also created in R using the package “CMplot”, obtaining the coordinates from the output from EffectorP. Scatterplots and corresponding statistical analyses were created and calculated in either Microsoft Excel or Python’s Matplotlib [90]. OrthoVenn3 diagrams were created using OrthoVenn3 run locally through Docker as described in the instructions. The boxplots in S11B Fig were also created locally on Python using the “matplotlib”, “seaborn”, “pandas”, and “scipy” packages.

Methods of inoculation and phenotyping disease severity

Spore solutions were prepared by growing isolates in 1/3^rd strength potato dextrose broth (PDB) for 5 days while being shaken continually. Spore counts were determined using a hemocytometer and concentrations were adjusted to 5x10⁵ spores/ml. Maize plants used to produce ears for inoculations were grown in 3 seasons: Spring 2022, Fall 2022, and Spring 2023, with sixty seeds sown from a total of 250 maize varieties in Citra, FL at the University of Florida Plant Science Research and Education Unit. Each season, varieties that produced 30 or more undamaged and uninfected ears were harvested 21–24 days after the 50% silking date of that variety, reducing the number of varieties that were assessed for disease symptoms to 160. Healthy ears were brought back to a lab where they were surface sterilized, inoculated, and incubated for one week before phenotyping disease symptoms. Inoculations were performed to simulate wounding followed by the introduction of infectious spores. This was done by piercing each ear with a sterile 18G needle, then using a new needle attached to a PrimaTech Adjustible Dose Vaccinator that was loaded with 0.75ml of conidial suspension and injected through the husk into the ear. Thirty ears were used to evaluate each maize variety, receiving inoculations from ten different Fusarium isolates in triplicate. Inoculations were completed using a circular design that allocated each Fusarium isolate to ten maize varieties. Inoculated ears were placed in labelled bags and incubated for one week in a growth chamber set to have 12/12 hours of light/dark and the temperature at 30°C. Ears were then removed from the bag, husked manually, and infected kernels were counted and recorded. Images were also taken using a custom-built imaging device for subsequent image analysis and disease phenotyping. The total number of ears that were phenotyped was 3716. Additional details and information regarding methods of inoculation and phenotyping can be found in Hudson et al. (2023) [91].

Fumonisin concentration measurements

Fumonisin concentrations were measured from 417 maize ears that were inoculated as described above in the Spring 2023 season. Ears inoculated with isolates known to have the FUM biosynthetic gene cluster were prioritized for sampling, and samples representing species known to not have the FUM cluster were minimized in sampling. Otherwise, sampling was randomly generated using the index number pre-assigned to each ear. Ears were shelled and the seeds were ground to a fine powder using a coffee grinder before fumonisin was extracted. We used the protocol as described in the Helica Fumonisin Hydro ELISA kit by Hygiena. Briefly, the ground maize was shaken in distilled water for three minutes and allowed to precipitate before the top layer was transferred to a new tube and centrifuged for five minutes. The supernatant was transferred to a new tube and then diluted. The ELISA test was performed following the manufacturer’s instructions and the optical density (OD) was read on an Epoch optical density plate reader by Biotek® following the manufacturer’s instructions. A dose-response curve was calculated using the standards provided by the Hygiena and the fumonisin parts per million (PPM) was calculated for each sample using the standard curve. As the distribution of fumonisin concentrations was positively skewed, the fumonisin concentrations were adjusted by transforming the scores to PPM^1/5 to normalize scores prior to estimating correlations between fumonisin content and infected kernel counts. The data remained non-normal according to a Shapiro-Wilk normality test, but no better transformation could be determined. Because of this, two correlations were estimated: the first assumed normality and was between infected kernel counts and fumonisin PPM^1/5 then again using averages per isolate. The second did not assume normality and used Spearman’s Rank Correlation Coefficient on both raw values and isolate averages. All correlations and scatterplots were presented using Excel or Matlab with a custom Python script.

Sequence Data Availability: All genome assemblies have been uploaded to NCBI under Bioproject ID number: PRJNA1047553.