Whole Genome Sequencing of Wisconsin Fusarium Isolates

Overview

This project documents a complete field-to-genome pipeline for characterizing the genomic diversity of Fusarium species causing ear rot of corn in Wisconsin — from statewide disease scouting through genome assembly, annotation, and effectorome prediction. To date, the pipeline has been applied to over 20 fungal isolates, generating high-quality reference genomes that are being used to understand pathogenicity, secondary metabolite biosynthesis, and adaptation to temperature stress.

All scripts used from basecalling through annotation and effectorome search are publicly available on GitHub.

---

Phase 1 — Field disease survey and sampling

Sampling was conducted across all major corn-producing cropping districts of Wisconsin — a statewide systematic disease survey designed to capture the geographic and ecological breadth of Fusarium populations present in the state. Survey sites spanned the Central Sands, southwest, southeast, and northern production regions.

At each site, ears showing symptoms of Gibberella ear rot (pink-red mold associated with Fusarium graminearum and related species) were collected at R5 maturity just before harvest.

---

Phase 2 — Pathogen isolation and culture

Symptomatic tissue was surface-sterilized and plated on selective media under standard aseptic conditions. Putative Fusarium isolates were hyphal-tipped or single-spore purified and confirmed by morphological assessment and molecular identification using ITS and species-specific PCR primers (RPB2 and TEF1 $\alpha$). Isolates (spores and mycelia plugs) were archived in 15% glycerol at −80°C for long-term storage.

Over 20 isolates representing multiple Fusarium species within the Fusarium graminearum species complex (including F. graminearum, F. cerealis), Fusarium fujikuroi species complex, and related taxa — were selected for whole genome sequencing based on geographic representation, toxin chemotype diversity, and phenotypic variation in pathogenicity assays.

---

Phase 3 — DNA extraction and library preparation

Metagenomic-quality high-molecular-weight (HMW) DNA was extracted from fresh mycelial tissue using commercial DNA extraction kits with modifications optimized for fungal material. DNA purity and integrity were assessed by NanoDrop spectrophotometry, Qubit fluorometry, and agarose gel electrophoresis prior to library preparation.

Native barcoding library preparation was performed using Oxford Nanopore Technologies (ONT) native barcoding kits, enabling multiplexing of multiple isolates per sequencing run. Libraries were loaded onto R10.4.1 flow cells and sequenced on the GridION and PromethION platforms at the United States Department of Agriculture (USDA-ARS) Dairy Forage Research Center in Madison, WI.

---

Phase 4 — Basecalling (USDA-SCINet Atlas GPU cluster)

Raw signal data (pod5 format) generated by the sequencer was processed using ONT’s proprietary software MinKNOW or on the Atlas GPU cluster (USDA-SCINet), which provides the GPU compute necessary for super accurate Dorado-based basecalling at scale. All pod5 files were basecalled at Q>15 accuracy using Dorado with native barcoding demultiplexing — a run that took approximately two days per batch. Output was separated into per-barcode BAM files, then converted to gzipped FASTQ using Samtools.

---

Phase 5 — Read quality control and pre-processing (USDA-SCINet Ceres)

Pre-processing was executed on the Ceres HPC cluster. The per-barcode FASTQ files were processed through a sequential QC pipeline:

Adapter removal — Porechop was used to identify and trim Oxford Nanopore sequencing adapters, converting gzipped FASTQ to clean FASTQ files.

Duplicate removal — Seqkit (via Miniconda) was applied to remove duplicate reads that can arise from library preparation artifacts.

Length filtering — NanoFilt was used to remove reads shorter than 500 bp (Batch 1) or 1 kb at Q≥20 (Batch 2), retaining only high-quality long reads for assembly.

Quality visualization — NanoPlot was used to generate per-sample summary statistics and read-length distribution plots, providing a quality checkpoint before assembly.

---

Phase 6 — Read correction and genome assembly

Read correction (Atlas GPU) — Filtered reads were passed through Dorado correct to perform deep-learning-based read error correction, converting FASTQ to FASTA format. This step removes base-quality scores and produces consensus-corrected sequences optimized for assembly.

Genome assembly (Ceres) — Corrected reads were assembled into draft genomes using Flye, a long-read assembler designed for Oxford Nanopore data. Flye takes corrected FASTA as input and outputs a draft assembly FASTA.

Alignment (Atlas) — Original (uncorrected) reads were aligned back to the draft assembly using Dorado aligner, producing per-barcode BAM alignment files that capture the full read-level variation against the assembled contigs.

Assembly polishing (Atlas) — The draft assembly was polished using Dorado polish, which uses the aligned raw reads to correct residual errors in the draft — a step that runs approximately nine hours per barcode on GPU compute.

---

Phase 7 — Assembly quality assessment (Ceres)

Polished assemblies were evaluated through a multi-tool quality pipeline:

• BUSCO / Compleasm — benchmarked universal single-copy orthologs to assess genome completeness against the Hypocreales lineage database
• GenomeTools (gt seqstat) — contig statistics including N50, total assembly size, and GC content
• Merqury — k-mer-based quality value estimation
• QUAST — reference-independent assembly statistics
• CRAQ — chimeric read-aware quality assessment

Results were visualized using Genious Prime, Galaxy, or R packages like gggenomes, ggpubr, ggsci, and ggplot2.

---

Phase 8 — Genome annotation (Ceres and local machine)

Gene models were predicted and annotated using Funannotate, a fungal-specific genome annotation pipeline that integrates ab initio gene prediction, RNA-seq evidence (where available), and functional database searches. Funannotate was run on Ceres for compute-intensive steps and on a local Ubuntu machine for interactive annotation review.

Telomere identification — TelomereSearch.py was used on the local machine with a 10 kb search window and 50-repeat threshold to identify telomeric repeat sequences in assembled contigs, confirming chromosome-level assembly quality.

---

Phase 9 — Secondary metabolite and effectorome analysis

Biosynthetic gene cluster (BGC) prediction — Assembled and annotated genomes were submitted to antiSMASH (Antibiotics and Secondary Metabolite Analysis Shell) to identify and characterize biosynthetic gene clusters — including trichothecene (DON), zearalenone, and other clusters relevant to mycotoxin production.

Effectorome prediction — The predicted proteome from each genome was screened through a two-step effector identification pipeline:

1. SignalP-6.0 — identifies proteins with secretion signal peptides, the first requirement for candidate effectors. Large proteome FASTAs were split into batches of ≤5,000 proteins for the web server, or run as a full proteome on local compute.

2. EffectorP 3.0 — filters signal peptide-containing proteins to predict which are likely fungal effectors and their subcellular localization targets.

The resulting predicted effectorome — the full complement of candidate secreted effector proteins — was compared across isolates to identify conserved and isolate-specific effectors potentially relevant to host specificity and virulence.

---

Compute infrastructure

Stage	Platform
Basecalling, alignment, polishing	USDA-SCINet Atlas (GPU cluster)
QC, assembly, annotation	USDA-SCINet Ceres (CPU cluster)
Telomere search, local annotation review	Personal machine (Ubuntu)
antiSMASH, SignalP-6 (web)	Web interfaces

---

Repository

All scripts — from basecalling and preprocessing through genome assembly, annotation, and effectorome search — are documented and shared publicly on GitHub. The repository follows a standardized project structure with separate directories for raw data, basecalled reads, scripts, results, and environment files.