Comparative genomics reveals lineage‐associated structural variation and diversification in a barley fungal pathogen Authors: Xiaohui Yu, Mumta Chhetri, Mafruha Hasan, Thomas H. Roberts, Peng Zhang, Robert F. Park, Yi Ding Published: 2026 ABSTRACT Summary Leaf rust, caused by Puccinia hordei, is a major barley disease worldwide. Despite repeated shifts in virulence, contrasting reproductive histories, and emerging fungicide insensitivity, the genomic basis of its diversification and adaptation remains poorly understood.In this study, we generated haplotype‐resolved, chromosome‐level genome assemblies for two isolates with contrasting virulence and analyzed 41 Australian isolates collected over 54 yr (1966–2020), integrating comparative and population genomics, mating‐type gene phylogenies, chromosome‐specific k‐mer profiling, genome‐wide copy‐number variation (CNV) analysis, and gene‐expression analysis.We identified a structurally dynamic chromosome characterized by repeat‐associated rearrangements, structural variation, and lineage‐associated CNV, representing the first evidence in a rust fungus of chromosome‐scale structural diversification of this extent. Population analyses distinguished clonally expanded lineages from recombination‐associated lineages, with mating‐type gene phylogenies providing further support for lineage differentiation. More recently collected isolates showed increased duplication‐associated variation, and CNV boundaries were associated with structural‐variant breakpoints. We also identified lineage‐associated amplification of Cyp51, with increased copy number associated with higher transcript abundance, supporting a potential role in fungicide adaptation.Overall, our findings highlight structural variation, contrasting reproductive histories, and lineage‐associated CNV as important contributors to diversification in P. hordei, providing insights for future rust pathogen surveillance and management strategies. Introduction Barley leaf rust, caused by the basidiomycete fungus Puccinia hordei, is one of the most important diseases threatening global barley production (Park, 2003). Breeding barley cultivars with genetic resistance is an effective and sustainable strategy for managing this disease; however, virulence matching deployed resistances has often emerged following the deployment of resistance (Park et al., 2015). A deeper understanding of biology and of virulence adaptation in P. hordei requires detailed studies of genomic content. Like other cereal infecting rust species, urediniospores and aeciospores of P. hordei have two separate haploid nuclei. It exhibits a complex macrocyclic, heteroecious life cycle that includes alternation on species of several genera including Ornithogalum (Duplessis et al., 2014; Park et al., 2015). Understanding of the biology of some cereal rust species has been greatly improved by haplotype‐phased genome assemblies and comparative genomics, which have revealed substantial genomic plasticity driven by transposable elements (TEs), gene duplications, and large‐scale structural variations (Duplessis et al., 2014; Schwessinger et al., 2020). The ability to generate high‐quality genome assemblies has further improved the resolution of complex genomic regions, enabling the characterization of mating‐type loci, effector repertoires, and mechanisms underlying pathogenicity (Wu et al., 2021; Henningsen et al., 2024; Luo et al., 2024; Guan et al., 2025; Tam et al., 2025). Population dynamics and adaptive responses to selective pressures such as fungicide application play a critical role in shaping the evolutionary trajectories of many fungal plant pathogens. In agroecosystems, extensive use of demethylation inhibitor (DMI) fungicides exerts strong selection pressure that often drives the rapid emergence of fungicide insensitive pathogen lineages (Cools & Fraaije, 2013; Lucas et al., 2015). In Australia, P. hordei undergoes sexual recombination on O. umbellatum, an introduced environmental weed species that has become naturalized in parts of south‐eastern Australia (Wallwork et al., 1992). Long‐term race analysis has documented recurrent shifts in pathotype composition and virulence acquisition, suggesting both clonal proliferation and recombination contribute to adaptation (Park, 2003). However, the genetic mechanisms underlying adaptation to fungicide pressure remain unknown. Recent work in other plant and human fungal pathogens including Rhynchosporium commune, Zymoseptoria tritici, and Candida albicans has shown that variation in genes like Cyp51 can mediate fungicide insensitivity through gene duplications and dosage effects (Ford et al., 2015; Mair et al., 2016; Stalder et al., 2023). Surveys of P. hordei across Australia from 2021 onwards indicated that isolates within a clonal lineage typified by pathotype 5453 P‐, first identified in Western Australia in 2001, exhibit insensitivity to the DMI fungicide tebuconazole. Fungicide insensitivity was later detected in other pathotypes, raising the possibility that genomic mechanisms may underlie azole resistance (Park et al., 2022). Elucidating population structure and genomic structural variations in P. hordei is thus essential for understanding its evolutionary potential and for developing sustainable disease control strategies. Compared with many other basidiomycetes, rust fungal genomes are characterized by extensive TE content and large predicted gene repertoires, with substantial variation among isolates within species (Cuomo et al., 2017; Corre et al., 2025). This complexity made the development of accurate genome assemblies difficult, with early attempts generating haplotype collapsed assemblies with dramatically compromised contiguity. Chen et al. (2019) reported a preliminary genome assembly for P. hordei that, despite being incomplete, provided initial insights into gene content and effector complement. Sequencing and assembling cereal rust genomes (Li et al., 2023; Liang et al., 2023) have been facilitated by long‐read technologies, particularly PacBio high fidelity (HiFi) sequencing and assemblers with sophisticated algorithms such as hifiasm and hicanu (Wenger et al., 2019; Nurk et al., 2020; Cheng et al., 2021). Haplotype phasing is crucial in analyzing heterokaryotic rust genomes. Advances in phasing and genome‐continuity improving technologies such as chromosome conformation capture (Hi–C) have enhanced the assembly of haplotype‐phased contigs to scaffold or chromosomal scale without the need for genetic and physical maps (Kronenberg et al., 2021). To gain deeper insight and understanding of the genome of P. hordei, we generated haplotype‐phased and chromosome‐level genome assemblies for P. hordei using two Australian isolates of contrasting virulence. The two haplotype‐phased and chromosome‐level assemblies were used to investigate genome architecture in P. hordei and to investigate variability in the pathogen from 1966 to 2020 in Australia. They provide a valuable genomic resource for an important plant pathogen and a crucial foundation for identifying key genes responsible for pathogen adaptation, elucidating virulence evolution, and advancing genomic studies of P. hordei. Materials and Methods Plant and rust materials and inoculation Two P. hordei G.H. Otth isolates Ph518 and Ph560 were maintained at the Plant Breeding Institute Cobbitty, The University of Sydney. Ph518 was isolated in eastern Australia in the 1970s and is avirulent to all Rph genes except Rph8 and Rph10. Ph560, collected in Western Australia in 2001, is virulent to Rph1, Rph2, Rph4, Rph6, Rph8, Rph9, Rph10, and Rph12, and likely represents an exotic incursion from which single‐step derivative mutant pathotypes arose (Chen et al., 2019; R. F. Park, unpublished). Both isolates were purified from single pustules and increased 2–3 times on the susceptible barley genotype Gus, with purity monitored by including a differential set (Park et al., 2003). An additional 31 P. hordei isolates, along with eight reported previously by Chen et al. (2019), across a time span of over 50 yr (Supporting Information Table S1), were included in population genomics. Inoculation of the additional isolates was performed and rust response examined as previously described (Park & Karakousis, 2002). DNA extraction and sequencing High molecular weight DNA was extracted from 1 g urediniospores using a modified CTAB protocol and purified as in Yu (2024). For each isolate, one PacBio Sequel II SMRT cell generated 30.9 Gb (mean read 17 252 bp, Ph518) and 37.15 Gb (18 515 bp, Ph560) HiFi reads (Novogene, Singapore). Hi–C sequencing was performed for isolate Ph560 only. Briefly, 100 mg urediniospores were crosslinked in 1% formaldehyde (15 ml, 20 min), quenched with glycine (1 g , 20 min), and centrifuged (1000 g , 1 min) to remove supernatant for shipment. Hi–C library was prepared using a Proximo Hi–C Microbe Kit (Phase Genomics, Seattle, WA, USA) and sequenced at Genewiz (South Plainfield, NJ, USA) to produce 100 million 150‐bp paired‐end reads. For population genomics, DNA from c. 50 mg urediniospores was extracted following Ding et al. (2021). Pair‐end (150‐bp) short reads library was prepared and sequenced at over 50× coverage using the NovoSeq X platform (Novogene, Singapore, Singapore). Phased genome assembly and cleanup For Ph560, the HiFi and Hi–C reads were fed into hifiasm v0.19.5 for genome assembly with the parameters ‘‐t22 ‐l3 ‐s0.55’. The resulting haplotype assemblies were blasted against a mitochondrial database retrieved from NCBI (https://ftp.ncbi.nlm.nih.gov/refseq/release/mitochondrion/) to identify and filter out mitochondrial sequences (Camacho et al., 2009). Haplotigs were then removed with purge haplotigs v.1.1.2 (Roach et al., 2018). First, a subset of HiFi reads was mapped to each haplotype using minimap2 v.2.22 (Li, 2018) with the parameters ‘‐ax map‐hifi ‐‐secondary=no’, and the resulting alignment files were sorted and indexed using samtools (Danecek et al., 2021). Last, the ‘hist’, ‘cov’, and ‘purge’ commands of purge haplotigs were executed sequentially to remove haplotig sequences. Hi–C scaffolding and telomere identification Ph560 haplotypes were scaffolded using yahs v.1.1a (Zhou et al., 2022). Raw Hi–C reads were mapped to each haplotype assembly according to Arima Genomics mapping pipeline (https://github.com/ArimaGenomics/mapping_pipeline/blob/master/arima_mapping_pipeline.sh). The resulting alignment files and haplotype assemblies were fed into YaHS to produce initial scaffolded assemblies using command ‘yahs’ with additional parameters ‘‐e GATC ‐‐telo‐motif TTAGGG’. The outputs were processed using ‘juicer pre ‐a’ command of YaHS to generate text files of Hi–C links, which were then loaded into Juicer tools to produce Hi‐C contact maps. The resulting files were manually curated with juicebox v.1.11.08 (Durand et al., 2016). The final scaffolded haplotype assemblies were generated using the ‘juicer post’ command of YaHS. Assembly quality was assessed with quast v.5.0.2 and busco v.5.7.1 (basidiomycota_odb10) (Gurevich et al., 2013; Simão et al., 2015). Telomeric motifs in each scaffolded haplotype were identified using the tidk v.0.2.64 (Brown et al., 2025). For Ph518, isolate‐specific Hi–C data were not available. Ph518 was assembled using hifiasm in HiFi‐only mode with Hi–C data from Ph560 used to assist scaffolding. The resulting Ph518 haplotype assemblies were subjected to homology‐based misassembly correction using RagTag (Alonge et al., 2022). RagTag was executed with additional parameters ‘‐f 5000 ‐d 500000 ‐b 10000 ‐‐inter’ to only detect and break large structural misassemblies. Hi–C and genome feature analysis Hi–C contact matrices were generated, normalized, and visualized using runhic v.0.8.7 (Wang, 2016) and R packages hicexperiment v.0.99.9, hicool v.1.7.1, hicontacts v.1.9.1, and interactionset v.1.36.1 (Lun et al., 2016; Serizay et al., 2024). Chromosomal content analysis and cis‐contact eigenvector calculation were conducted using the python package cooltools v.0.7.1 (Open2C et al., 2024). Summed raw HiC contact numbers were calculated for both haplotypes respectively using the R package hicexperiment v.0.99.9. To evaluate compartment‐like genome organization, intrachromosomal Hi–C contact frequencies were analyzed using the oriented first eigenvector, hereafter referred to as E1, derived from balanced Hi–C matrices at 100 kb resolution. E1 summarizes the dominant pattern of chromatin interactions across genomic windows and is commonly used to describe compartment‐like genome organization (Open2C et al., 2024). For each chromosome, bins were classified according to positive or negative E1 sign. Detrended interaction scores were calculated as observed/expected cis‐contact frequencies after correcting for genomic distance decay. Interaction frequencies between same‐sign (positive–positive or negative–negative) and opposite‐sign bin pairs were compared using one‐sided Welch t‐tests. Chromosome‐level compartment‐like interaction strength was quantified as the difference between mean same‐sign and opposite‐sign detrended interaction frequencies. The genomic span of positive and negative E1 regions and the size of the largest contiguous regions with consistent E1 sign were calculated from consecutive 100 kb bins. Copy number variation (CNV) calling The reference genome was divided into nonoverlapping 10 kb bins. For each isolate, read depth per bin was calculated using the coverage command of bedtools, excluding bins overlapping > 50% low‐complexity or repetitive regions (Quinlan & Hall, 2010). Depth values were normalized to the genome‐wide median within each isolate and smoothed with a 10‐kb window median. CNVs were identified as previously with the control‐FREEC program (Breakpoint_Threshold 0.05; Window 10 000) (Ding et al., 2021). CNVs were further assessed with the DNAcopy algorithm in R package dnacopy v.1.70.0 (Seshan & Olshen, 2025). Circular binary segmentation was applied to each sample independently to detect statistically significant candidate CNV breakpoints. We used a significance threshold α = 0.01 and required a minimum of three bins per segment. CNVs with log2 ratios above +0.3 were classified as duplications and below −0.3 as deletions. Gene‐level CNVs were defined by ≥ 50% overlap between transcripts and CNV segments using BEDTools intersect. Genes with multiple transcripts were resolved to the transcript with greatest overlap. To identify shared or isolate‐specific gene‐level CNVs, the same CNV event present in over 90% of the isolates was considered fixed, and others polymorphic. To reduce potential misclassification due to low coverage or edge effects, genes with conflicting calls (deletion and duplication across isolates) were excluded from specific enrichment tests. Distances between CNV‐affected genes and the nearest TE were calculated using GenomicRanges in R based on the RepeatMasker annotations. TE class and family were extracted, and CNV‐TE associations visualized in ggplot2 (Wickham, 2016). Additional methodological details for repeat annotation, comparative genomics, population genomics, RNA‐seq analyses, CNV analyses, k‐mer analysis, mating‐type gene identification, and yeast complementation assays are provided in Methods S1. Results Haplotype‐phased genome assemblies Ph560 and Ph518 were selected to generate haplotype‐phased, chromosome‐level assemblies. For Ph560, initial haplotype assemblies were produced using hifiasm in Hi–C integrated mode, resulting in haplotype A (Ph560A) with a total length of 158.5 Mb and an N50 of 10.6 Mb and haplotype B (Ph560B) with a total length of 154.6 Mb and an N50 of 11.8 Mb. After assembly cleanup and Hi–C scaffolding, both haplotypes contained 18 chromosome‐sized scaffolds with cumulative lengths of 142.6 Mb and 144.0 Mb for Ph560A and Ph560B, respectively. To validate phasing quality, we performed Hi–C mapping on the genome assembly comprising both haplotypes. At 100 kb resolution, the processed haplotypes A and B Hi–C matrices contained 307.1 million and 305.5 million raw contact counts, respectively. Among these, Ph560A and Ph560B each contained c. 136.3 million valid intrachromosomal cis contacts. The genome‐wide contact matrix showed a clear diagonal of high interaction frequency, indicating strong intrachromosomal contacts. Two major blocks along the diagonal correspond to Ph560A and B chromosomes, respectively, with minimal signal between them (Fig. 1a). A breakdown of contact types across Ph560A chromosomes (Figs 1b, S1) showed that cis interactions dominated (c. 75–80%) across most chromosomes, supporting successful haplotype phasing and spatial location of chromosomes in each nucleus. A total of 35 telomeres were identified in each haplotype (Fig. 2a), with all chromosomes except 3A and 13B assembled to telomere‐to‐telomere resolution. busco analysis revealed 95.0% and 94.3% gene completeness for Ph560A and Ph560B, respectively (Table 1). Over 90% of assessed BUSCOs (Ph560A: 91.5%, Ph560B: 90.4%) were both complete and single copy. busco completeness for the combined Ph560 haplotypes reached 95.5%. Haplotype‐resolved assembly of Puccinia hordei isolate Ph560 guided by Hi–C scaffolding. (a) Hi–C contact maps of the two Ph560 haplotypes, with haplotype A on the top left and haplotype B on the bottom right along the diagonal. Shown as balanced core with log10 scale at a resolution of 100 kb bins. (b) Proportion of reads mapping to haplotype A chromosomes representing cis contacts confirms accurate haplotype phasing. Stacked bar plot showing the percentage of intrachromosomal (cis, blue) and interchromosomal (trans, green) Hi–C interactions for each haplotype A chromosome. The absence of excessive trans contacts further indicates minimal cross‐haplotype mapping errors and high assembly contiguity. Pairwise whole‐genome alignments of haplotype‐resolved Puccinia hordei isolates Ph560 and Ph518. (a) Chromosome‐scale synteny plots between haplotypes A and B of Ph560 (left) and Ph518 (right). Each vertical pair of bars represents the full set of chromosomes for hapA (left) and hapB (right), with colored ribbons connecting syntenic regions. Syntenic regions are shown in grey. Ribbon orientation reflects alignment direction (co‐linear or inverted), and crossing patterns or overlap indicate structural rearrangements such as inversions, duplications, or translocations. Telomeres are highlighted as blue dots at chromosomal distal ends. (b–g) Pairwise alignments of whole haplotype assemblies using minimap2. All pairwise combinations of haplotypes from Ph560 and Ph518 were aligned at the chromosome scale. The x‐ and y‐axes correspond to the genome coordinates of the two haplotypes being compared. Diagonal lines represent co‐linear alignments; off‐diagonal or repeated blocks indicate structural rearrangements or duplications. Color shading denotes sequence similarity. Table 1Assembly metrics of the two Puccinia hordei isolates Ph518 and Ph560.ParametersPh518APh518BPh560APh560BChromosomal scaffolds18181818Chromosomal length (bp)147 501 061141 623 950142 632 809143 952 256Longest chromosome (bp)13 261 99211 537 97410 599 43411 537 491GC content (%)41.641.641.641.6Repeat content (%)72.272.771.872.9Complete BUSCOs (%)93.191.595.094.3Single‐copy BUSCOs (%)85.386.291.590.4Duplicated BUSCOs (%)7.85.33.53.9Fragmented BUSCOs (%)0.91.00.90.8Missing BUSCOs (%)6.07.54.14.9 Assembly metrics of the two Puccinia hordei isolates Ph518 and Ph560. A high coverage (130×) of PacBio HiFi reads was used to generate initial haplotype assemblies for Ph518. After deduplication, contigs from the two haplotypes contained a total of 61 telomeres at contig ends, indicating highly continuous contig assembly approaching chromosomal level. Because Hi–C data were not available for Ph518, contigs were scaffolded by using the Ph560 Hi–C reads followed by homology‐based genome correction. This cross‐isolate scaffolding strategy yielded 18 chromosomes for each haplotype (Fig. 2a). busco completeness analysis revealed slightly reduced quality compared with the Hi–C scaffolded Ph560 assemblies, with 93.1% completeness for Ph518A and 91.5% for Ph518B (Table 1). The combined Ph518 haplotypes achieved 95.1% complete BUSCOs. Whole‐genome alignment against a Puccinia triticina reference (Pt15 haplotype 1) (Li et al., 2023) revealed chromosome‐to‐chromosome collinearity for all four P. hordei haplotypes, supporting the overall chromosome‐scale structure of the assemblies (Fig. S2). Long terminal repeat (LTR) aging and gene structure annotation The four haplotype assemblies were highly repetitive, with over 70% of each comprising repetitive elements (Fig. S3A). Among these, LTR retrotransposons were most common, accounting for over 38% of each genome. We then calculated pairwise LTR‐retrotransposon divergence for RepeatMasker‐annotated LTR elements. The divergence distributions showed a shared peak across all four haplotypes (Fig. S3B). When converted to approximate insertion ages using an external fungal substitution‐rate calibration, this peak corresponded to c. 3 Ma. Thus, this estimate should be interpreted as an approximate relative timing rather than a precise absolute age. The shared divergence peak suggests a common period of LTR‐retrotransposon activity before or during haplotype divergence, rather than continuous uniform accumulation. Gene prediction based on both transcriptome and orthologous protein evidence annotated over 11 000 genes per haplotype (Table 2). Subsequent secretome prediction identified c. 1500 putative secreted proteins in each haplotype, of which 739–822 were further classified as candidate effectors by EffectorP 3.0 depending on the haplotype (Tables 2, S2A–D). Table 2Summarized results of gene structure annotation and predictions of secreted proteins and effectors for the four Puccinia hordei haplotypes.Ph518APh518BPh560APh560B Gene structure annotation Number of genes11 49311 48511 19411 065Mean gene length (bp)1729175117791733Genome covered by genes (%)13.514.214.013.3 Secretome prediction Number of secreted proteins1532154715111457Percentage of total proteins (%)13.313.513.513.2 Effector prediction Candidate effectors822815793739Percentage of secreted proteins (%)53.752.752.550.7 Summarized results of gene structure annotation and predictions of secreted proteins and effectors for the four Puccinia hordei haplotypes. Comparative genomics reveals synteny and structural diversity between haplotypes Pairwise D‐GENIES heatmaps provided an overview of haplotype sequence similarity (Fig. S4), and comparative orthogroup analyses summarized species‐specific gene content and phylogenetic relationships among Puccinia species (Fig. S5). Syntenic plots of the four haplotypes (Fig. 2a–g) revealed chromosome‐scale collinearity and structural conservation. Most homologous chromosome pairs aligned end‐to‐end, supporting accurate haplotype resolution and chromosome‐level scaffolding. Pairwise haplotype comparisons further quantified the extent of synteny and structural divergence (Table S3). The two Ph518 haplotypes were the most similar pair, with 72.4% of Ph518A and 75.6% of Ph518B classified as syntenic. By contrast, comparisons between Ph560B and Ph518 haplotypes showed the lowest syntenic proportions, particularly Ph560B vs Ph518A, in which 61.2% of Ph560B and 59.0% of Ph518A were classified as syntenic. Among nonsyntenic categories, unaligned regions accounted for the largest fraction of each comparison (13.7–24.3%), followed by duplications (4.5–9.9%) and translocations (5.4–8.3%), whereas inversions represented a smaller fraction (1.0–2.3%). These results show that the four haplotypes retain substantial chromosome‐scale synteny while differing in proportions of duplicated, translocated, inverted, and unaligned sequence. At the chromosome level, Chr9 showed large, complex rearrangements that disrupted synteny across all haplotypes (Figs 2a, S6). To quantify this pattern, we calculated the syntenic proportion of Chr9 and compared it with that of the remaining chromosomes combined. Across all six pairwise haplotype comparisons, Chr9 consistently showed a lower syntenic proportion than the remaining chromosomes (Table S4). Chr9 syntenic coverage ranged from 43.0 to 56.2% across comparisons, whereas the remaining chromosomes showed higher syntenic coverage, ranging from 60.8 to 75.7%. This consistent reduction in syntenic coverage supports that Chr9 contains a higher proportion of structurally diverse sequence than the rest of the genome. Notably, the elevated nonsyntenic fraction of Chr9 was largely explained by increased duplication and translocation coverage (Table S4). For example, in the Ph560A–Ph518A comparison, reference/query‐averaged duplication and translocation coverage reached 24.4% and 20.4% on Chr9, respectively, compared with 6.9 and 4.1% across the remaining chromosomes. This pattern was not observed to the same extent on other chromosomes, suggesting that Chr9 has undergone particularly strong structural divergence both within and between isolates. Chromosome 9 exhibits repeat‐associated structural variation Prompted by the extensive haplotype‐specific structural variation observed on Chr9, we investigated whether this chromosome displays distinct architectural and compositional features relative to the rest of the P. hordei genome. Chr9 has the lowest gene density and one of the highest TE contents across chromosomes (Fig. S7A,B). Self‐alignment of Chr9A and Chr9B generated dense secondary diagonal and off‐diagonal alignments (Fig. 3a, S7C), indicating that repeated genomic regions aligned to multiple positions. Such patterns are characteristic of tandem and inverted duplications and are consistent with large‐scale sequence duplication. A c. 3 Mb region showed strong self‐similarity (> 95%) and is gene‐sparse (Figs 3a, S7C and S8). Comparative alignment of Chr9A/B between Ph560 and Ph518 showed several structural variation breakpoints within or adjacent to repeat‐rich regions. Candidate effector genes were also observed at the flanks of these intervals and near some structural breakpoints (Figs 3a, S7C). Consistent with the repeat‐rich architecture of Chr9A, genes on Chr9A were located at marginally greater mean distances from the nearest upstream and downstream TEs than genes on the remaining chromosomes. The mean distances from gene 5′ and 3′ ends to the nearest TE on Chr9A were 0.65 and 0.64 kb, respectively, compared with 0.60 and 0.60 kb for the remaining chromosomes (Fig. 3c). Although this difference was small, it is consistent with local expansion of intergenic regions in repeat‐rich regions of Chr9A. Structural and compositional features of Puccinia hordei isolate Ph560 Chr9A. (a) Whole‐chromosome self‐alignment of Chr9A reveals tandem and inverted repeat arrays. Gene (green) and predicted effector (purple) annotations are shown along the axes. Interisolate alignment between Ph560 and Ph518 Chr9A highlights structural variation breakpoints, represented as gaps. The upper panel indicates percentage identity of alignments. (b) Hi–C contact matrix and first eigenvector (E1) profile of Chr9A in Ph560 along with Chr7A, Chr8A, and Chr10A. Grid lines indicate positive/negative E1 interfaces. Ordered chromosome regions cover from 60 to 100 Mb as shown on the y‐axis. (c) Distances from gene boundaries (5′ and 3′ ends) to the nearest transposable element (TE), comparing Chr9A with all other chromosomes. (d–f) Genome‐wide distributions of GC content, gene density, and TE coverage across eigenvector‐defined domains (Q1–Q4). Higher eigenvector values are associated with increased gene density and reduced TE content. (g–i) Feature distributions for Chr9A. Pairwise t‐tests are indicated (ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001). (J) Whole‐chromosome Hi–C maps and oriented E1 tracks for Chr9A. The left heatmap shows balanced Hi–C contact scores (Iterative Correction and Eigenvector decomposition‐normalized interaction frequencies). The right heatmap shows detrended interaction scores, calculated as observed/expected contact frequencies after correcting for genomic distance; values are centered at zero, where 0 indicates expected contact frequency, positive values indicate enrichment, and negative values indicate depletion. The bottom track shows the oriented E1 eigenvector. (k) Detrended contact scores for intrachromosomal bin pairs with the same E1 sign (A–A or B–B) vs opposite sign (a, b) within Chr9A and all other chromosomes. Scores are log2(observed/expected) contact values after normalization and correction for genomic distance‐dependent contact decay. Sample sizes are shown above the boxplots. Boxes indicate the median and interquartile range, and whiskers extend to the most extreme values within 1.5 × IQR. P‐values are from one‐sided Welch tests comparing same‐sign to opposite‐sign contacts. Hi–C contacts were observed at 100 kb resolution. We next examined chromosome conformation in Ph560 urediniospores using Hi–C contact maps and assessed compartment‐like organization using the first eigenvector (E1), the principal eigenvector derived from the normalized Hi–C contact matrix. Ph560‐derived E1 values were divided into four quartiles and compared with gene density and repeat coverage. Genome‐wide, gene density and repeat coverage showed opposite patterns across E1 quartiles (Fig. 3e,f). Gene density increased with higher E1 values, and Q4 had significantly higher gene density than the lower quartiles (P < 0.0001 for all comparisons; Fig. 3e). By contrast, repeat coverage decreased with higher E1 values, dropping from a mean percentage of c. 70% in Q1 to c. 60% in Q4, with Q4 showing significantly lower repeat coverage than Q1–Q3 (P < 0.0001 for all comparisons; Fig. 3f). While for Chr9A, an opposite relationship between E1 values and genomic features was observed (Fig. 3h,i). Gene density decreased with higher E1 values, with Q4 showing significantly lower gene density than all the lower quartiles (P < 0.001; Fig. 3h). Repeat coverage increased with higher E1 values, with Q3 and Q4 showing significantly higher repeat coverage than Q1–Q2 (P < 0.05; Fig. 3i). We further assessed these relationships by grouping E1 values into 10 equal‐sized intervals. Across chromosomes other than Chr9A, gene density increased and repeat coverage decreased with increasing E1 (slopes = 1.68 and −6.55, respectively), whereas Chr9A showed the opposite trends (slopes = −3.43 and 2.59). The slopes differed significantly between chromosome groups for both features (P < 0.001). GC content showed only weak positive trends, with no significant difference between chromosome groups (Fig. S7K). Thus, on Chr9A, higher Ph560‐derived E1 values are associated with repeat‐rich and gene‐poor regions, opposite to the genome‐wide trend. This indicates that Chr9A has an unusual E1‐associated organization linked to its repeat‐rich architecture. We hypothesized that Ph560‐derived E1 might reflect compartment‐like organization. To test this hypothesis, we examined intra‐chromosomal Hi–C contacts and global gene expression levels. Because no epigenetic chromatin‐state data are available for P. hordei, we interpret these patterns conservatively as compartment‐like organization rather than definitive A/B compartment assignment. At 100 kb resolution, Chr9A and Chr9B both showed continuous chromosome‐scale Hi–C interaction patterns without abrupt discontinuities or translocation‐like signals, supporting the structural consistency of the assemblies (Figs 3j, S9A). Plaid‐like interaction patterns were also observed across other chromosomes represented by Chr1A/B examples (Figs S9B, S10A), indicating that compartment‐like interaction structure is present genome‐wide in Ph560. However, Chr9 contained numerous interaction blocks related to large repeat‐rich regions (Figs 3j, S9A). Across the genome, bins with the same E1 sign showed significantly higher detrended cis contact frequencies than bins with opposite E1 signs (Fig. 3k), consistent with moderate compartment‐like structure. To further explore whether Ph560‐defined E1 regions are associated with transcriptional activity across chromosomes, we analyzed RNA‐seq expression patterns from four representative isolates sampled from infected barley tissue at 7 d post inoculation. These isolates were collected spanning five decades, including two earlier collected isolates Ph491 (1990) and Ph518 (1970) and two more recently collected isolates Ph612 (2009) and Ph685 (2020). Expression values were based on DESeq2‐normalized counts, with biological replicates averaged within each isolate. Because only two isolates were included in each group, statistical comparisons between earlier‐ and more recently collected isolate groups were used to characterize expression distributions among genes rather than isolate‐level reproducibility across lineages. Across chromosomes other than Chr9 in the Ph560 hapA genome, gene expression differed significantly across E1 quartiles in all isolates (Fig. S11A). When bins are grouped into A‐like and B‐like classes based on the oriented E1 sign, genes in A‐like regions showed higher expression than those in B‐like regions in both isolate groups (Fig. S11C), indicating an exploratory association between Ph560‐defined E1 regions and transcriptional activity in the representative isolates analyzed. By contrast, Chr9A showed a weaker and less consistent pattern (Fig. S11B,D). In the earlier collected isolates, expression differences across E1 quartiles were not significant, whereas in the more recently, collected isolates the observed differences were primarily caused by reduced expression in the highest E1 quartile rather than by a consistent trend across bins (Fig. S11B). Similarly, direct comparison of A‐like and B‐like regions on Chr9A showed weak or nonsignificant expression differences relative to the genome‐wide pattern (Fig. S11D). Collectively, within the four representative isolates analyzed, Ph560‐defined E1 regions were associated with transcriptional differences across chromosomes other than Chr9, whereas this relationship was weaker and less consistent on Chr9A. Because candidate effector genes may be related to virulence change in fungal plant pathogens and were observed near repeat‐rich and structurally variable regions on Chr9, we next examined whether these genomic differences were associated with altered effector expression across isolates. Chr9A candidate effectors showed contrasting expression patterns between earlier and more recently collected isolate groups, with many genes displaying reduced expression in the more recently collected isolates (Fig. S12A–C). Most Chr9A candidate effectors were located close to the nearest TEs, either overlapping TEs or within 2 kb. When expression shifts between earlier and more recently collected isolates were stratified by TE‐distance class, TE‐proximal effectors showed greater expression shifts than more distant effectors (2–5 kb from the nearest TEs) (P = 3.98 × 10−2; Fig. S12D). These results indicate that repeat‐rich, structurally variable regions on Chr9A are associated with altered expression of nearby candidate effector genes. To further assess whether Chr9 exhibits unusual compartment‐like organization, we quantified chromosome‐level interaction strength based on detrended cis contacts between same‐sign and opposite‐sign E1 bins represented by the hapA genome. Although same‐sign interactions were generally found genome‐wide, Chr9A did not differ significantly from other Ph560A chromosomes in overall interaction strength (Fig. S10B). Similarly, the genomic span (Fig. S13A) and contiguous block structure of E1‐positive and E1‐negative regions on Chr9A (Fig. S13B) fell within the range observed across the genome. Together, these results indicate that Chr9A is primarily distinguished by repeat‐rich sequence composition and repeat‐associated structural variation, while the overall compartment‐like interaction patterns remain broadly comparable to those observed across the rest of the genome. Population genomics reflects lineage divergence and contrasting reproductive structure of a continental P. hordei population In Australia, P. hordei is the only known cereal rust pathogen that undergoes sexual recombination due to the presence of its alternate host, Star of Bethlehem (Ornithogalum umbellatum), in barley leaf rust affected regions of South Australia (Wallwork et al., 1992; Park, 2008). This provides a unique opportunity to investigate how contrasting reproductive modes, including sexual recombination and clonal expansion, have shaped the population structure and evolutionary history of barley leaf rust in Australia. We hypothesized that if both clonal expansion and sexual recombination have contributed to the evolution of the Australian P. hordei population, genetically distinct clusters should display contrasting signatures of these reproductive modes. To investigate this, we selected 41 P. hordei isolates representing pathotypes detected in annual Australian surveys over a 54‐yr period for analysis (Cotterill et al., 1995; Park, 2003). One isolate originated from the alternate host O. umbellatum (Ph577). Of the remaining 40, 37 were collected from infected crops and three were spontaneous virulence gain mutants identified in glasshouse testing (Table S1). We first sought to determine whether the Australian P. hordei population is composed of genetically distinct lineages. We performed genome‐wide SNP analyses using both haplotype assemblies of Ph560 to assess the consistency of clustering, co‐ancestry, and introgression patterns. PCA based on PC1–PC2 showed consistent grouping for most isolates across the two haplotype references, whereas Ph518, Ph577, and Ph640 showed haplotype‐dependent divergence in their positions (Fig. 4a,b; Table S1). K‐means clustering (k = 5) supported five distinct genetic clusters (Fig. 4a,b), which broadly corresponded with patterns inferred from ADMIXTURE (Fig. 4c,d), co‐ancestry matrices (Fig. S14), phylogenetic trees (Fig. 4e,f), and SplitsTree networks (Fig. S15). Ph518 remained phylogenetically distinct in higher‐order PCs (PC3–PC4) for both haplotype references (Fig. S16), supporting its genome‐wide genetic divergence relative to the other isolates. Population structure analyses of Puccinia hordei using single‐nucleotide polymorphisms (SNPs) called against both Ph560 haplotypes to assess the consistency of clustering, co‐ancestry, and introgression patterns. (a, b) PCA revealed similar major axes of variation. K‐means clustering (k = 5) to PC1 and PC2, which grouped individuals into five clusters. (c, d) ADMIXTURE analysis based on haplotype A (c) and haplotype B (d) at K = 3–5. Lines separate individuals derived from PCA defined clusters. (e, f) Phylogenetic inference of P. hordei isolates based on hapA (e) and hapB (f) references. Isolate collection years were indicated. Scales indicate number of nucleotide substitutions per site. To further examine whether lineage divergence was accompanied by chromosome‐specific diversification, we compared an earlier collected group represented by Clusters 1–2 (collected predominantly from the 1970s–1990s) with a more recently collected group represented by Clusters 3–5 (collected from the 2000s–2020s). Genome‐wide fixation index (F ST) analyses showed elevated differentiation on Chr9 in both haplotype references (Fig. S17A,D), and subsequent analyses focused on haplotype A because of the similar patterns for both haplotypes. Chr9A had the highest chromosome‐wide mean F ST among haplotype A chromosomes and was enriched for top 1% F ST windows, with 6.9% of outlier windows compared with 4.1% for the next highest chromosome (P = 0.02; Fig. S17E). Nucleotide diversity on Chr9A was not significantly different from the remaining chromosomes in the earlier collected group (Clusters 1–2, P = 0.19) but was significantly higher in the more recently collected group (Clusters 3–5, P = 2.08 × 10−5; Fig. S17F,G). These results suggest elevated recent diversification of Chr9A relative to the genome‐wide background in the P. hordei population. We next asked whether these genetically defined clusters reflect different reproductive modes. Under a predominantly clonal model, isolates within a cluster are expected to show low MLG diversity, high linkage disequilibrium, reduced SNP diversity, small pairwise SNP distances, tight clustering in PCA, and highly similar ancestry profiles (Milgroom, 1996; Grünwald et al., 2016; Guo et al., 2022). By contrast, clusters showing signatures consistent with sexual recombination are expected to have higher MLG diversity, weaker linkage disequilibrium, greater SNP diversity, and more reticulate relationships in network analyses. To test these expectations, we calculated MLG ratio (MLG counts/number of samples), genome‐wide linkage disequilibrium, numbers of polymorphic SNPs, mean pairwise SNP distances, co‐ancestry, and SplitsTree network structure for each PCA‐defined cluster, with generally consistent results obtained from both haplotype references (Fig. S18A,B). Because some clusters had modest sample sizes, these metrics were interpreted collectively as reproductive signatures rather than as definitive tests of reproductive mode for individual clusters. Clusters 2 and 5 showed the clearest signatures of clonal expansion. Both clusters exhibited low MLG ratios, elevated genome‐wide linkage disequilibrium, fewer polymorphic SNPs, and shorter mean pairwise SNP distances than the other clusters (Fig. S18A,B). Isolates in these clusters also grouped tightly in PCA plots (Fig. 4a,b), exhibited similar ADMIXTURE ancestry profiles at k = 3–5 (Fig. 4c,d), and showed minimal reticulation in SplitsTree networks (Fig. S15A,B) and high co‐ancestry values (Fig. S14B,D). Together, these independent patterns support predominantly clonal expansion in Clusters 2 and 5. By contrast, Clusters 3 and 4 showed contrasting patterns across these metrics (Fig. S18A,B), indicating a recombination‐associated diversification, in line with the only cereal rust pathogen undergoing sexual recombination on alternate host in Australia (Park, 2008). Cluster 1 showed intermediate values across several metrics, suggesting that this cluster may have experienced a mixed reproductive history (Fig. S18A,B). These results indicate that the P. hordei population in Australia has not been shaped by a single reproductive mode but instead comprises both clonally expanded lineages and genetically diverse lineages showing signatures consistent with recombination. Finally, we examined allele sharing among PCA‐defined clusters using f‐branch statistics and co‐ancestry matrices. In the haplotype A analysis, f‐branch statistics indicated strong allele sharing involving Cluster 5 and several other clusters, including Cluster 4 (f b ≈ 0.98), Cluster 1 (f b ≈ 0.9), and Cluster 2 (f b ≈ 0.6) (Fig. S14A). In the haplotype B analysis, the strongest signal involved Clusters 5 and 3 (f b ≈ 0.9; Fig. S14C). Co‐ancestry heatmaps also supported these relationships at the isolate level. Ph660 from the recombination‐associated Cluster 4 showed high co‐ancestry with the Cluster 5 clonal block because of the tight grouping with Cluster 5 isolates in the haplotype A analysis (Fig. S14B), whereas Ph518 and Ph640 from Cluster 3, also recombination‐associated, showed high co‐ancestry with Cluster 5 in the haplotype B analysis (Fig. S14D). These results suggest historical allele sharing or introgression involving Cluster 5‐related ancestry. Overall, the population genomic analyses support a model in which the Australian P. hordei population contains both clonally expanded lineages and more recombined lineages, with additional evidence of allele sharing among some clusters. Diverse lineages inferred from Chr9 haplotype k‐mer and mating‐type divergence Reproduction and mating‐type genes are important determinants of fungal population structure and reproductive biology. Using mating‐type genes from Pt and P st as queries, we identified both the P/R and HD loci in P. hordei. STE3.1 was located on Chromosome 1 in all four haplotypes and was highly conserved across isolates. The STE3.2 and STE3.3 homologs were located on Chr9A and Chr9B in Ph560 but were haplotypically swapped in Ph518, although their coding sequences remained identical between isolates. Orthology analyses grouped the P. hordei STE3 genes into the expected STE3.1 and STE3.2/3.3 clades, with Pt showing the closest relationship (Fig. S19A,B). By contrast, the HD locus on Chromosome 4 displayed clear allelic variation between haplotypes, with distinct bW‐HD1 and bE‐HD2 alleles identified in each haplotype. Orthology analyses similarly grouped these genes into the expected HD clades (Fig. S19C,D). Given the elevated structural divergence of Chr9 and the location of STE3.2/3.3 on this chromosome, we next performed Chr9‐specific k‐mer analysis to examine whether Chr9 carries lineage‐associated variation across the P. hordei isolates. The top 100 k‐mers showing the greatest variation across isolates were independently extracted from Chr9A and Chr9B and used to cluster isolates. In both Chr9A and Chr9B matrices, variable k‐mer clustering separated the isolates into two primary groups (Fig. 5a,b). Isolates in the upper portion of the heatmaps, corresponding to earlier collected isolates exclusively belonging to PCA Clusters 1 and 2, shared one set of variable k‐mer signals, whereas isolates in the lower portion, corresponding mainly to more recently collected isolates from Clusters 3–5, shared a different set of k‐mer signals (Fig. 5a,b; Table S1). These results indicate that Chr9‐derived variable k‐mers separate earlier and more recently collected isolate groups and capture lineage‐associated sequence variation on both Chr9 haplotypes. After clustering similar Chr9 k‐mers, a dominant k‐mer cluster was identified for each haplotype, comprising 49 k‐mers for Chr9A and 88 k‐mers for Chr9B (Fig. 5c,d; Table S5). De novo motif discovery among the most variable k‐mers from both Chr9A and Chr9B recovered a conserved motif resembling forkhead/winged‐helix DNA‐binding sequences (Fig. 5e), often associated with transcriptional regulators such as Fkh1 in Saccharomyces cerevisiae. Most k‐mers with significant matches only had one motif hit (Fig. 5f), suggesting these are discrete, nonrepetitive sequences, possibly located within regulatory domains. Chr9‐derived k‐mer clustering revealed lineage‐associated sequence variation broadly concordant with divergence in Chr9‐linked STE mating‐type genes in Puccinia hordei. (a, b) Distribution of sample‐wise k‐mers present across Chr9A and Chr9B. (c, d) The most variable (top 100) k‐mer features on each haplotype. To interpret the genomic content underlying these k‐mer differences, the most variable k‐mers were clustered by sequence similarity, resulting in a reduced representation matrix. Collapsed k‐mer clusters recapitulate the same two‐group structure seen in the full matrix and identify specific clusters with skewed distributions across isolate groups. Color key showing k‐mer sequence numbers. (e) Clustered k‐mer sequences are enriched in a motif related to forkhead/winged‐helix (FKH) motifs. (f) Motif matches per sequence. Most k‐mer clusters show a single dominant motif match per sequence. (g, h) Phylogenetic Trees of STE3.2 and STE3.3 coding sequences. STE3.2 on Chr9A (g) and STE3.3 on Chr9B (h) both yielded trees with two dominant clades (gray and yellow shaded) consistent with the k‐mer clusters. Colors represent isolates from five major PCA defined genetic clusters. Numbers indicate ultrafast bootstrap support percentages based on 1000 replicates. Phylogenies based on the Chr9‐linked mating‐type genes STE3.2 and STE3.3, particularly STE3.3, were broadly consistent with the Chr9 variable k‐mer grouping and separated earlier and more recently collected isolates at this locus (Fig. 5g,h). We further evaluated HD‐gene based phylogenies using both haplotypic copies of HD1 and HD2 genes (Fig. S20). In contrast to collection time separation observed for STE genes, the HD‐gene phylogenies showed patterns more concordant with the reproductive‐structure clusters inferred from MLG and linkage disequilibrium analyses. Isolates from each of Clusters 2 and 5 generally grouped together with limited within‐cluster divergence, consistent with their clonal expansion signature (Fig. S20). By contrast, isolates from Clusters 1, 3, and 4 showed more dispersed placement across several HD‐gene trees (Fig. S20). This shows that, compared with STE genes, HD‐locus variation is more concordant with the reproductive‐structure clusters inferred from MLG and linkage disequilibrium analyses, although these single‐gene trees should be interpreted cautiously. Genome‐wide CNV profiling reveals a duplication‐associated shift in recently collected isolates Prompted by the structural variation signatures detected in the haplotype comparisons, including elevated duplication on Chr9, we profiled genome‐wide CNVs across 40 P. hordei isolates sampled over a 54‐yr period (long‐reads based Ph518 excluded). We first summarized the number of genes affected by deletion or duplication in each isolate. To assess whether reference‐mapping bias could contribute to apparent CNV differences, we summarized mapping rate, properly paired read percentage, and callable genome proportion for each isolate against both Ph560 haplotype references. Callable genome proportion was defined as the percentage of reference bases with sequencing depth ≥ 5 after retaining alignments with MAPQ ≥20. Most isolates showed high and broadly comparable mapping performance across both haplotypes (Table S6). In Fig. 6(a), the leftmost 13 samples correspond to PCA Cluster 1, whereas the rightmost seven samples correspond to Cluster 2. These two clusters largely represent isolates collected during the earlier survey periods (Table S1) and showed deletion‐biased CNV profiles, with deletions generally accounting for > 70% of CNV‐affected genes in each isolate. By contrast, the more recently collected isolates from Clusters 3–5, positioned in the middle of Fig. 6(a), broadly showed a greater contribution of duplications to CNV‐affected genes, although a few exceptions were observed, such as Ph637 and Ph672. Genome‐wide copy‐number variation (CNV) landscape highlights structural divergence across Puccinia hordei isolates and potentially links Cyp51 amplification to fungicide insensitivity. (a) Total number of genes affected by deletions and duplications across P. hordei isolates, revealing widespread CNV activity across lineages. (b) PCA of CNV profiles shows lineage‐associated differences between earlier‐ and more recently collected isolates, with more recently collected isolates showing a greater contribution of duplication‐associated variation. Colors indicate previously assigned PCA clusters of P. hordei isolates. (c) Heatmap showing CNV directly affects Cyp51 across all the isolates. CNV values are shown as log2‐normalized. (d) Gene Ontology (GO) enrichment analysis of duplicated genes across isolates shows overrepresentation of terms related to metabolism, intracellular transport, protein complex organization, and transcriptional regulation. Circle size represents the number of genes per term, and color intensity indicates significance. (e) Functional validation of PhCyp51 in a yeast erg11 deletion strain confirming its ability to confer resistance to the demethylation inhibitor fungicide prothioconazole. BY4743 wild‐type (WT) background and empty vector were used as control. (f) Quantitative polymerase chain reaction (qPCR) showing elevated Cyp51 copy number in selected isolates. Relative Cyp51 genomic copy number in selected low‐copy and CNV‐gain isolates was quantified following the ΔCt (PhCyp51 relative to β‐tubulin). Small points represent six technical qPCR measurements per isolate, representative of two independent genomic DNA extractions. For statistical comparison, technical measurements were first averaged within each isolate to generate one isolate‐level mean ΔCt value. Low‐copy isolates Ph487, Ph491, and Ph570 were then compared with CNV‐gain isolates Ph561, Ph627, and Ph675 using a two‐sided Welch's t‐test (n = 3 isolates per group; P < 0.001). Boxes show the median and IQR, and whiskers extend to the most extreme values within 1.5× IQR. (g) Model of CNV on the Cyp51 locus. Schematic representation of the Cyp51 region showing gene positions, transposable element (TE) contents, and structural variant breakpoints (BND) and deletions (DEL) between Ph560 and Ph518. The locus comprises repeated homologous blocks containing Cyp51 and flanking genes possibly underwent tandem duplication followed by inverted duplication events. PCA based on gene‐level CNV profiles further showed that deletion and duplication patterns contributed differently to lineage separation (Fig. 6b). In the deletion‐based PCA, Cluster 1 formed the most distinct group, separating from most other isolates along PC1. By contrast, in the duplication‐based PCA, Cluster 5 was the most clearly separated group, shifting away from the other clusters along PC1. To examine chromosome‐level CNV patterns, we calculated average CNV log2 ratios per chromosome using 10 kb bins across all isolates for both haplotypes. The resulting heatmaps revealed lineage‐specific CNV signatures that were broadly consistent with the previously assigned PCA clusters (Fig. S21A,B). In particular, Cluster 5 isolates formed a distinct group in both haplotypes and showed widespread copy‐number gains across multiple chromosomes, with especially prominent gains on Chr9. Because TEs have been implicated in CNV formation in fungal pathogens (Stalder et al., 2023; Tralamazza et al., 2024), we examined the genomic context of CNVs. CNV losses were significantly closer to nearby TEs than duplications or neutral regions, with LINE/Jockey, LTR/Copia, simple repeats, and low‐complexity regions showing the strongest associations (Fig. S21C–N). GO enrichment analysis of genes overlapping high‐confidence CNV gains identified broad enrichment for metabolic and intracellular processes, including lipid metabolism, protein transport, carbohydrate transport, and protein targeting (Fig. 6d), indicating that CNVs may affect diverse cellular functions. Collectively, these results suggest a lineage‐associated shift in CNV composition, from deletion‐biased profiles in earlier‐collected isolate group toward increased duplication‐associated variation in more recent collected isolate group. We next examined whether CNV boundaries were associated with structural variant breakends (BNDs) which are known as CNV drivers (Stalder et al., 2023; Tralamazza et al., 2024). Among the randomly selected isolates examined, the more recently collected isolates Ph560, Ph584, Ph608, Ph612, Ph626, and Ph685 showed significantly shorter CNV‐BND distances than expected under chromosome‐preserving permutations (50 000 permutations; Benjamini‐Hochberg‐adjusted P = 0.0002–0.0363), whereas none of the earlier collected isolates differed significantly from the null expectation (Fig. S22). Representative genome‐wide CNV profiles, using Ph482 (1990) and Ph685 (2020) as representatives of earlier‐ and more recently collected isolates, respectively, further illustrated contrasting patterns between these groups. Earlier‐collected isolates showed widespread losses, whereas more recently collected isolates displayed gain‐biased CNV landscapes frequently associated with nearby BNDs (Fig. S23). Together, these analyses support a possible relationship between structural breakpoints and CNV formation in more recently collected lineages. Because CNV gains may alter gene dosage and transcriptional output, we next examined expression patterns of genes located within CNV gain regions using RNA‐seq data from the earlier collected isolate Ph491 and the more recently collected isolates Ph612 and Ph685. Across isolates, copy number gained genes did not show a clear increased expression pattern relative to neutral ones (Fig. S24A). However, when focusing specifically on genes gained in Ph612 or Ph685 relative to the baseline isolate Ph491, expression changes showed an overall shift toward increased expression, although decreased expression was also observed for some genes (Fig. S24B). To further test for dosage effects, we quantified the relationship between copy‐number difference and relative expression change. Genes with larger copy‐number gains tended to show greater expression increases, resulting in a significant positive relationship between copy‐number difference and expression fold change (linear model, P = 2.67 × 10−2; Fig. S24C). Collectively, these results indicate that CNV gains in the more recently collected isolates represented by Ph612 and Ph685 are associated with modest dosage effects on gene expression. CNV‐driven divergence in recently collected lineages is associated with fungicide insensitivity During investigation of CNV‐driven genomic variation, we identified Cyp51 as a key gene recurrently duplicated in a subset of isolates (Figs 6c, S23 and S25A). These isolates were restricted to recently collected lineages, including isolates from Clusters 4 and 5 (Figs 4 and 6a–c). In Ph560, five identical copies of Cyp51 were detected, including one on Chr9A and four tandemly arranged copies on Chr9B (Figs 6g, S25). By contrast, Ph518 contained only two copies, with no coding‐sequence differences relative to the Ph560 copies (Fig. S25C). Hierarchical clustering of CNV levels confirmed lineage‐specific Cyp51 duplication patterns (Fig. 6c), whereas earlier collected isolates predominantly showed neutral or deletion states at the locus (Fig. 6a–c). The duplicated Cyp51 copies were located within the CNV and BND abundant regions within Chr9 (Fig. S23). Consistently, these Cyp51 amplified isolates include the pathotype 5453 P‐ derived lineage from Cluster 5, of which Ph560 (Table S1) is the founding isolate. In planta studies under controlled conditions have shown that these isolates are insensitive to several DMI fungicides (Park et al., 2022). The founding isolate Ph560 from the clonal Cluster 5 was first detected in Western Australia in 2001 and subsequently spread across all major barley‐growing regions of Australia, together with closely related derivative pathotypes (R. F. Park, unpublished). To validate the CNV patterns, we quantified Cyp51 copy number by quantitative polymerase chain reaction (qPCR) using genomic DNA from representative CNV‐gain isolates (Ph561, Ph627, and Ph675) and CNV‐loss isolates (Ph487, Ph491, and Ph570). CNV‐gain isolates consistently showed significantly elevated Cyp51 copy number relative to low‐copy isolates (Fig. 6f). To determine whether increased copy number was associated with altered transcription, we examined Cyp51 expression in the two earlier collected isolates (Ph491 and Ph518, low‐copy) and two more recently collected isolates (Ph612 and Ph685, high‐copy) analyzed for RNA‐seq as aforementioned. Reads were mapped against the Ph560A reference, which contains a single Cyp51 copy (Ph560.Chr9A.97). Because all Cyp51 copies were sequence‐identical, total transcript abundance across copies could be aligned to this single locus of Ph560A and quantified directly. DESeq2‐normalized counts showed that high‐copy isolates exhibited substantially increased Cyp51 expression relative to low‐copy isolates (Fig. S26), supporting a positive relationship between copy number and transcript abundance. To test functional relevance, we cloned the PhCyp51 coding sequence (Table S7) into a galactose‐inducible expression vector and introduced it into the S. cerevisiae Δ erg11 mutant strain. In the absence of native ergosterol biosynthesis, PhCyp51 expression restored growth on galactose media and conferred resistance to the DMI fungicide prothioconazole, confirming functional compatibility and azole insensitivity in a heterologous system (Fig. 6e). No growth rescue was observed for the empty vector controls or when grown on non‐inducing glucose medium. Collectively, these results support a link between Cyp51 and fungicide insensitivity in P. hordei. To investigate the structural basis of Cyp51 duplication, we further examined the expanded Chr9 locus in the context of the genome‐wide association between CNV boundaries and nearby BNDs described above (Figs S22, S23). Short‐read mapping from older isolates such as Ph482, Ph488, and Ph489 against the Ph560A reference revealed clear coverage gaps flanking the duplicated Cyp51 region (Fig. S25A), consistent with structural divergence at the locus. Long‐read alignments from the older isolate Ph518 to Ph560B reference further identified large alignment gaps and BNDs near the boundaries of the duplicated region (Fig. S25B). The locus also contains simple/low‐complexity and unclassified repeats (Fig. 6g), consistent with the broader association between CNVs and repeat‐rich genomic regions observed genome‐wide (Fig. S21). The expanded region in Ph560 consisted of four highly similar c. 13.6 kb duplicated blocks, each containing a Cyp51 copy and conserved flanking genes (Fig. 6g). Additional BNDs were detected at internal deletion junctions within the locus. Together, these observations support a stepwise duplication model in which an ancestral single‐copy unit first underwent tandem duplication followed by inverted duplication of the duplicated segment, generating the four‐copy structure observed in Ph560B (Fig. 6g). The association of BNDs with these repetitive regions suggests that local repeat architecture may have facilitated recombination and CNV formation at the locus. Discussion This study presents haplotype‐resolved genome assemblies and population genomics in P. hordei, the causal agent of barley leaf rust, providing a comprehensive structural and functional map of its heterokaryotic genome. Using PacBio HiFi sequencing and Hi–C scaffolding, we reconstructed chromosome‐level assemblies for two pathogenically diverse isolates, revealing extensive structural conservation interspersed with haplotype‐ and lineage‐specific variation. For Ph518, although Hi–C data were unavailable, high‐coverage HiFi reads produced highly continuous contigs. Over 30 contigs reached near‐chromosome resolution, including Chr9, consistent with a recent report in P. triticina assembled without Hi–C data (Li et al., 2023). One limitation is that the use of Ph560 Hi–C data and homology‐based assembly may introduce haplotype errors. To further assess this possibility, we validated Ph518 Chr9 rearrangement positions using Ph518 self‐mapped HiFi reads. At corresponding Chr9 BND positions, we examined local HiFi read depth and counted high‐quality primary alignments spanning each position. All tested Chr9 BND positions were supported by breakpoint‐spanning Ph518 HiFi reads, and normalized local depth did not show a systematic approximately twofold increase or reduction that would indicate widespread local repeat collapse or redundant expansion. These results support the local integrity of the Ph518 Chr9 assembly, although they do not exclude smaller‐scale errors in highly repetitive regions (Table S8). Therefore, the detected divergence between chromosomes 9A and 9B, along with consistent large‐scale haplotype‐specific rearrangements and reduced synteny, indicated true allelic structural divergence rather than intermixing of haplotypes or phase switching. Compared to the previous fragmented and collapsed genome (Chen et al., 2019), our assemblies serve as a valuable resource for comparative genomics and further understanding of cereal rust biology. Comparative analyses showed broad chromosome‐scale synteny among haplotypes but also identified haplotype‐ and isolate‐associated structural differences, including duplications, translocations, and inversions, particularly on Chr9. We hypothesized that Chr9 might have undergone unique evolutionary processes, including repeat‐mediated sequence expansion and structural variation. To test this, we conducted comparative analyses across genome structure assisted with Hi–C and genomic content, focusing on both whole‐genome and Chr9‐specific patterns. Our analyses revealed that P. hordei Chr9 displays features of a core chromosome carrying accessory‐like regions, similar to those described in Cochliobolus heterostrophus (Condon et al., 2013) and the vascular wilt pathogen Verticillium dahliae (Klosterman et al., 2011; Faino et al., 2016). Both Ph560 and Ph518 provided strong evidence for the variable nature of Chr9, despite being isolated > 25 yr apart, supporting a biological basis rather than assembly noise in Ph518. Unlike canonical accessory chromosomes in fungi such as Z. tritici (Plissonneau et al., 2018), Fusarium oxysporum (Ma et al., 2010), and Leptosphaeria maculans (Grandaubert et al., 2014), which are often dispensable and lineage‐specific, Chr9 in P. hordei remains structurally conserved across isolates while containing abundant translocations, duplications, inversions, and noncolinear regions. Hi–C analyses of Ph560 urediniospores identified genome‐wide compartment‐like interaction patterns associated with E1 signal. Comparison of Ph560‐defined E1 regions with RNA‐seq data from four representative isolates sampled from infected barley tissue at 7 dpi showed an exploratory association with transcriptional activity. However, this relationship was weaker and less consistent on Chr9A. Importantly, chromosome‐level interaction strength and the genomic span of E1‐defined regions on Chr9A were not exceptional compared with other chromosomes. These results suggest that the distinctiveness of Chr9 is primarily associated with repeat‐rich genome composition and repeat‐associated structural variation, rather than with chromosome‐scale compartmentalization. Because E1 and RNA‐seq were generated from different isolates and developmental contexts, and because only four RNA‐seq isolates were analyzed, this analysis should be interpreted as an exploratory gene‐level association between Ph560‐defined E1 regions and expression patterns in representative isolates, rather than as direct evidence that chromatin state regulates transcription or as evidence for systematic population‐ or lineage‐level transcriptional differences. Broader isolate‐ and stage‐matched Hi–C and RNA‐seq data would be required to directly test whether these E1‐associated expression patterns are conserved across the wider P. hordei population. Chr9 also contains repetitive and gene‐sparse regions where structural breakpoints and candidate effector genes were observed nearby. Candidate effectors located near repeat‐rich regions on Chr9A also showed contrasting expression profiles between older and newly emerged isolates, with TE‐proximal effectors displaying the greater expression shifts. Although no direct virulence function was demonstrated, these results support an association between structural divergence and altered expression of nearby genes. Similar TE‐associated genome restructuring has been implicated in effector diversification and rapid adaptation in several fungal plant pathogens (Faino et al., 2016; Hartmann, 2022). Collectively, the elevated structural variation, together with the CNVs and gene expression patterns on Chr9, suggest that this chromosome represents a focal region of genome structural diversification in P. hordei. Population genomic analysis of P. hordei isolates collected in Australia over 54 yr (Luig, 1985; Cotterill et al., 1995; Park, 2003; R. F. Park, unpublished) identified five major genetic clusters using multiple complementary analyses, including PCA, ADMIXTURE, phylogenetic reconstruction, SplitsTree networks, and co‐ancestry analyses. MLG and linkage disequilibrium analyses could further infer reproductive structure within these clusters. Clusters 2 and 5 showed signatures consistent with predominantly clonal expansion, including low MLG diversity, elevated linkage disequilibrium, limited admixture, tight PCA grouping, and high within‐cluster co‐ancestry. By contrast, Clusters 3 and 4 showed contrasting patterns across these metrics and were identified as recombination‐associated clusters, whereas Cluster 1 showed intermediate features suggestive of a mixed reproductive history. These results support a mixed reproductive structure within the Australian P. hordei population involving both long‐term clonal propagation and recombination‐associated diversification. Phylogenies of HD mating‐type alleles broadly reflected the reproductive‐structure clusters inferred from MLG and linkage disequilibrium analyses, suggesting that the variation at these loci is associated with lineage differentiation in P. hordei, consistent with observations in other rust fungi (Holden et al., 2023; Luo et al., 2024). In addition, chromosome‐level F ST and nucleotide diversity analyses identified elevated differentiation and increased diversity on Chr9A in more recently collected lineages, suggesting Chr9 as a structurally dynamic region associated with ongoing genome diversification in P. hordei. We also demonstrate that CNV is a major component of population‐level genome divergence in P. hordei. Genome‐wide CNV profiling revealed lineage‐associated differences between earlier‐ and more recently collected isolates, with earlier‐collected isolates showing predominantly deletion‐biased profiles and more recently collected isolates showing a greater contribution of CNV gains. Recent studies in fungal pathogens have implicated TEs and structural breakpoints as important drivers of CNV formation (Stalder et al., 2023; Tralamazza et al., 2024). Consistent with this, CNV regions in P. hordei were associated with repeat‐rich genomic regions, particularly simple repeats and low‐complexity sequences. Moreover, CNV boundaries in several recently collected isolates occurred significantly closer to BNDs than expected by chance, supporting a possible relationship between genome rearrangement and CNV formation. Genes affected by CNV gains also showed a modest but significant dosage effect on transcription, indicating that copy number changes can contribute to altered gene expression at the genome‐wide level. Our study further supports Cyp51 amplification as a candidate mechanism contributing to fungicide insensitivity in a rust fungus. We identified lineage‐specific amplification of Cyp51 within the structurally dynamic Chr9 region in isolates previously shown to be insensitive to DMI fungicides (Park et al., 2022). Detailed characterization of fungicide sensitivity and resistance is ongoing to assess the prevalence and evolutionary dynamics of Cyp51 CNV across broader P. hordei populations and related rust species. Cyp51 copies were located within a structural‐variation hotspot on Chr9, and their copy‐number amplification was further supported by qPCR‐based copy‐number quantification and expression analyses. Cyp51 encodes sterol 14α‐demethylase, the target of DMI/triazole fungicides, and resistance can arise through increased Cyp51 expression or target‐site mutations (Ma et al., 2006; Cook et al., 2021; Rosam et al., 2021). The expanded Chr9B locus consisted of four highly similar duplicated blocks containing identical Cyp51 copies and conserved flanking genes, together with nearby BNDs and repeat‐dense regions. Both short‐read and long‐read alignments from earlier‐collected isolates supported structural divergence at the locus, consistent with a stepwise duplication history‐involving tandem and inverted duplication events. Importantly, high‐copy isolates showed substantially increased Cyp51 transcript abundance relative to low‐copy isolates, supporting an association between copy‐number amplification and elevated Cyp51 expression. Previous reports in P. striiformis f. sp. tritici showed that sequence polymorphisms in Cyp51 were associated with altered fungicide sensitivity (Cook et al., 2021). Our results integrate structural, transcriptional, and heterologous functional evidence, supporting that Cyp51 amplification may also contribute to fungicide insensitivity. Similar CNV‐associated azole resistance mechanisms have been described in Rhynchosporium commune and human fungal pathogens including Candida albicans and Cryptococcus neoformans (Stalder et al., 2023). Our findings therefore highlight CNV‐driven amplification as a potential mechanism contributing to fungicide‐associated adaptation in P. hordei. Together, our haplotype‐resolved assemblies and population genomic analyses show that genome structural variation contributes substantially to diversification in P. hordei. We identify Chr9 as a structurally dynamic chromosome containing repeat‐associated rearrangements and lineage‐associated CNVs and show that Cyp51 amplification is associated with recently collected lineages and is consistent with a contribution to fungicide insensitivity. These findings provide new insight into genome evolution and adaptation in cereal rust fungi. Competing interests None declared. Author contributions YD and RFP designed this study. YD and XY conducted the analyses and drafted the manuscript. XY and YD performed all experiments. PZ contributed to genomic DNA and RNA sequencing. MC prepared genomic DNA extraction and yeast plating. MH and THR prepared materials for Hi–C sequencing. XY and MC contributed equally to this work. Disclaimer The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations. Supporting information Fig. S1 Hi–C validation of the Ph560 haplotype‐resolved assembly. Fig. S2 Whole‐genome alignment of Puccinia hordei haplotypes to the Puccinia triticina reference. Fig. S3 Repeat composition and LTR‐retrotransposon divergence across Puccinia hordei haplotypes. Fig. S4 Pairwise sequence similarity among Puccinia hordei haplotypes. Fig. S5 Orthogroup enrichment and phylogenetic relationships among Puccinia species. Fig. S6 Chromosome‐scale collinearity among the four Puccinia hordei haplotypes. Fig. S7 Genome feature distributions and Chr9 repeat‐associated structural variation. Fig. S8 Whole‐genome self‐alignment of Ph560 haplotype A. Fig. S9 Hi–C interaction patterns and E1 profiles for Chr9B and Chr1B. Fig. S10 Hi–C‐based assessment of compartment‐like organization. Fig. S11 Exploratory association between Ph560‐defined E1 regions and gene expression. Fig. S12 Chr9A candidate effector expression shifts and TE proximity. Fig. S13 Genomic span and continuity of compartment‐like regions. Fig. S14 F‐branch and co‐ancestry analyses of PCA‐defined clusters. Fig. S15 SplitsTree networks of genetic relationships among Puccinia hordei isolates. Fig. S16 Higher‐order PCA of genome‐wide SNP variation. Fig. S17 Chromosome‐level population differentiation highlights elevated divergence on Chr9. Fig. S18 Reproductive signatures across PCA‐defined clusters. Fig. S19 Orthogroup classification of mating‐type genes. Fig. S20 HD‐gene phylogenies across PCA‐defined clusters. Fig. S21 Genome‐wide CNV distributions in the Ph560 reference genome. Fig. S22 Association analysis between CNV boundaries and structural breakends. Fig. S23 Genome‐wide CNV profiles and nearby structural breakends. Fig. S24 Expression patterns of genes in CNV‐gain regions. Fig. S25 Structural validation and sequence conservation of the Cyp51 locus. Fig. S26 Cyp51 expression in CNV‐gain and non‐CNV isolates. Methods S1 Supplementary methods. Table S1 Forty‐one Australian Puccinia hordei isolates collected between 1966 and 2017, along with PCA defined clusters and Chr9 k‐mer groups. Table S2 Putative secreted proteins predicted from the Ph518 and Ph560 haplotype assemblies. Table S3 Genome‐wide syntenic and nonsyntenic sequence categories identified by MUMmer/SyRI across six pairwise Puccinia hordei haplotype comparisons. Table S4 Comparison of syntenic and nonsyntenic sequence categories between Chr9 and the remaining chromosomes across six pairwise haplotype comparisons. Table S5 Collapsed Chr9 k‐mer clusters identified for both Ph560 haplotypes across Puccinia hordei isolates. Table S6 Short reads mapping rates and callable genome proportions against both haplotypes of Puccinia hordei Ph560. Table S7 Primers used in PCR studies of Cyp51. Table S8 HiFi read support and self‐mapping depth at Ph518 Chr9 breakpoints relative to Ph560. Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.