New resources for the worldwide study of wheat powdery mildew populations

A dataset was compiled using a total of 399 isolates of B. graminis ff. spp. tritici and dicocci that were previously sequenced with Illumina short-read technology forming the basis of our analysis here (Table A in S2 Appendix). The dataset included the recently published Illumina short-reads of powdery mildew isolates from various parts of the world, e.g., Egypt, Russia, USA, and the Fertile Crescent [7,14,41]. The short reads produced in all these studies were used here and mapped to the genome of the reference isolate Bgt_CHE_96224 [13].

The assembled dataset included sequences of isolates from all continents, except Antarctica (Fig 1A). The majority of samples originate from the Fertile Crescent, which is the region of origin of both the wheat powdery mildew and its host [7]. The Principal Component Analysis (PCA) of the SNP data (see methods) clearly revealed distinct populations (Fig 1B). Most of the variance in the PCA can be explained by the first two principal components, loosely reflecting latitude and longitude, mostly for the Eurasian wheat powdery mildew isolates (Fig A in S1 Appendix). The powdery mildew populations in North and South America, along with the ones in East Asia seem to be separated. The three isolates from Australia group together with the USA powdery mildew isolates as described before [7]. The powdery mildew populations sampled in Europe, Egypt, the Fertile Crescent, central Russia and Kazakhstan overlap. This is in agreement with the hypothesis that there are traces of admixture between these populations, but also a gradient of geographic with genetic correlation [7].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Worldwide dataset of wheat powdery mildew shows varying levels of genomic diversity and clustering populations.

(A) A map with origin of the mildew isolates used in this study. Map outlines were produced with the R packages rnaturalearth and ggplot2, based on Natural Earth data (https://www.naturalearthdata.com/downloads/50m-cultural-vectors/). (B) PCA of the SNPs of 387 isolates of B. g. tritici. Isolates from tetraploid wheat (B.g. dicocci), rye (B.g. secalis) and rye/wheat hybrid Bgtl_THUN_12 were excluded to improve resolution of the PCA for the majority of isolates. (C) Admixture plot of the 399 isolates for K = 13. (D) Nucleotide diversity of the populations with at least eight isolates, based on eight randomly chosen isolates (see Table B in S2 Appendix) and analysing sequence windows of 10 kb. P-values represent the pairwise comparisons between neighboring pairs, while the stars show how significantly different the values are between each population and the ISR powdery mildew population.

https://doi.org/10.1371/journal.ppat.1013196.g001

ADMIXTURE analysis showed that K = 13 is the most probable number of ancestries according to cross-validation error (Fig B in S1 Appendix), with most populations harboring sequences from different ancestries, in some cases reflecting geographical proximity (Fig 1C). For Example, the TUR mildew population shows three main subpopulations. The one with the fewest isolates has a unique mostly unhybridized ancestry, the second one shares identity with ISR isolates and the third, largest one, appears to be a mixture between a EUR and RUS ancestry. In contrast, the EGY population shows only low admixture and two main ancestral subpopulations that are mostly unique with only low levels of admixture in the neighboring RUS, TUR, ISR and IRN mildew populations. Powdery mildew of tetraploid wheat (B.g. dicocci) has its exclusive ancestry at K = 13, even though it shares some genetic identity with isolates from the Fertile Crescent at K 2–12 (Fig C in S1 Appendix), which is expected due to geographical proximity. Additionally, it shares traces of ancestry with isolates of the EGY, RUS and CHN mildew populations (Fig C in S1 Appendix). For K = 13, the B. dicocci population looks more isolated with only traces found in the neighbor IRN powdery mildew population. Additionally, the previously described hybrid nature of the JPN powdery mildew population is visible up until K equals nine (Fig C in S1 Appendix) [7].

Furthermore, nucleotide diversity of these populations (with eight random isolates per population, Table B in S2 Appendix, in 10 kb windows) (Fig 1D) shows that the populations outside Europe, West Asia and Africa that are more isolated in the PCA (i.e., ARG, USA, JPN, CHN) have a statistically less diverse mean (e.g., ARG is statistically less diverse than CHN, and CHN is less diverse than diverse EGY). This can be explained by the isolated populations being further away from the region of origin with little genetic exchange. In this regard, they are similar to island populations, unlike the European – West Asian – North African populations, which hold the highest diversity even as individual populations. Moreover, as previously described [7], the isolates from Australia are genetically similar to the USA powdery mildew population.

A singletons analysis, identifying SNPs unique to only one isolate within a population, supported the results of the nucleotide diversity (Table B in S2 Appendix, Fig D in S1 Appendix). A mantel test indicated that there is no significant correlation between genetic distance and geography, while using most global isolates (366 isolates, Table C in S2 Appendix, Fig E in S1 Appendix). This could be attributed to trade which could allow for recombination between populations across continents. However, when using only the wheat powdery mildew isolates in the populations around the broader region of origin (166 isolates, powdery mildew populations TUR, IRN, ISR, RUS, KAZ, Table C in S2 Appendix, Fig E in S1 Appendix), which are not separated by any major geographic barriers, such as oceans or high mountains, there is a clear positive correlation (Fig E in S1 Appendix). This result supports a previous study that used a smaller dataset [7].

A Blumeria graminis pangenome representing worldwide diversity

We sequenced eight B. graminis isolates from around the world previously collected (Bgt_ARG_4_2, Bgt_USA_2, Bgt_JPN_CHIKA, Bgt_IRN_GOR_2, Bgt_CHN_52_27, Bgt_CHN_17_40, Bgd_ISR_211, Bgs_1459, Fig 1A and 1B, Table A in S2 Appendix) with PacBio long-read technology, resulting in chromosome-scale assemblies. For this study, we also included chromosome-scale sequences of three previously published B. graminis isolates (Bgt_CHE_96224, Bgtl_THUN_12, Bgt_ISR_7) [13,41–43]. We used a B. hordei genome (isolate Bh_DH14 from England [44], GCA_900239735.1) [12] only when needed, due its fragmentation [45,46]. While eight genomes belong to the forma specialis tritici, we also used one isolate of rye, triticale and wild emmer (tetraploid) wheat mildews in the pangenomic dataset (Table A in S2 Appendix), resulting in a pangenome of a total of 11 powdery mildew genomes with reference genome quality similar to the reference Bgt_CHE_96224. The quality of these genome assemblies was assessed using the CRAQ software, which showed high values of continuity (Table D in S2 Appendix, Methods). The assemblies had genome sizes between ~134Mb for the B.g. secalis isolate up to ~143Mb for the B.g. dicocci isolate, with the B.g. tritici genome sizes falling in between these numbers. (Table E in S2 Appendix). The actual genome sizes differ, largely due to highly variable repetitive sequences in the centromeric region. As in a previous study [13], centromeres were defined by the presence of the centromere-specific retrotransposon family RII_Fuji (Fig F in S1 Appendix), and described in more detail below.

BUSCO (Benchmarking Universal Single-Copy Orthologs) analyses showed high levels of completeness of all chromosome-scale genome assemblies, with values above 94%, 97% and 98% completeness for the Leotiomycetes, the Ascomycota and the fungi reference dataset (“Fungi Odb10”), respectively (Figs G and H in S1 Appendix). The different genomes contain between 7,563 and 9,454 genes (Fig I.A in S1 Appendix), with the effector genes constituting ~10% of the genes annotated, while the protein size distribution of the encoded effectors was consistent among isolates (Fig I.B in S1 Appendix).

All newly sequenced isolates have 11 chromosomes as was found in the Bgt_CHE_96224 reference isolate (Figs 2, J in S1 Appendix and Table E in S2 Appendix) [13].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Chromosomal synteny of the 11 B. graminis isolates in the pangenome.

The lines in colour indicate genomic regions of Minimum Alignment length of 100 kb. Each colour represents a chromosome from 1 to 11. The asterisks indicate the isolates from populations that were shown to be hybrids. The different shades of black and grey represent the different formae speciales. The red blocks indicate centromeres and the light blue ones effector genes.

https://doi.org/10.1371/journal.ppat.1013196.g002

Three of the newly sequenced isolates in this study have few small unassembled contigs (including Bgt_JPN_CHIKA, Bgt_CHN_17_40, and Bgt_CHN_52_27, see Table E in S2 Appendix). These contained parts of the mitochondrial genome, along with un-anchored repetitive sequences. There is strong sequence collinearity between isolates, along all chromosomes, except in the highly repetitive centromeric regions (Fig 2, see also below). A comparison with barley powdery mildew (B. hordei) showed much weaker synteny and sequence conservation, but we still found homology along all chromosomes (Fig K in S1 Appendix).

According to a phylogenetic network created with PhyloNet using 5,982 single-copy orthogroup proteins, B.g. secalis forms an outgroup together with the hybrid B.g. triticale isolate, which has a hybrid node that indicates potential recombination. A hybrid node (which implies the potential reception of genetic material from two parents instead of one common ancestor) is also observed for the Bgt_JPN_CHIKA isolate (Fig L in S1 Appendix), as expected from previous studies [7].

We constructed a pangenome graph (Fig M in S1 Appendix) which largely reflects the chromosomal collinearity shown in Fig 2. Most chromosome arms show occasional nodes with additional sequences, while most nodes are found in centromeric regions due to high sequence diversity (Fig M in S1 Appendix, see also below). Additionally, multiple sequence nodes are found in terminal regions of chromosomes, e.g., chromosomes 5, 7 and 11, (Fig M in S1 Appendix, Table F in S2 Appendix), indicating the presence of unique chromosome segments in some isolates.

Average Nucleotide Identity (ANI) analysis in all chromosomes showed varying levels of homology: all the 11 B.g. tritici isolates sequenced to chromosome scale share individually more than 99.45% homology with Bgt_CHE_96224. The B.g. dicocci isolate shares ~99.34%, the B.g. secalis ~98.75%, and the B.g. triticale ~99.44% average homology with the original reference B.g. tritici Bgt_CHE_96224 assembly (Table G in S2 Appendix). This result reflects the evolution and phylogenetic distances of the respective isolates as expected.

The sequence of the mitochondrial genome for the isolate Bgt_CHN_17_40 was assembled in one contig of ~102 kb (Fig N.A in S1 Appendix). The mitochondrial genome sequence shows high collinearity with the previously published B. hordei DH14 and Bgt_CHE_96224 mitochondrial sequences (Fig N.B in S1 Appendix, Table G in S2 Appendix) [12,47]. We annotated 168 open reading fames (ORFs) in the Bgt_CHN_17_40 mitochondrion genome, similar to the numbers we obtained when we annotated the previously published mitochondrial sequences from B. hordei (163 ORFs) and Bgt_CHE_96224 (156 ORFs) (Table H in S2 Appendix) [12,47].

We also reviewed the mating type of these genomes, by using blast search of the published mating type loci on the genomes (Table I in S2 Appendix), using Bgt_CHE_96224 (MAT 1-2-1: Bgt-3306 and SLA-2: Bgt-2805) [48], and the MAT 1-1-1 locus with the Bgt_CHN_52_27 gene BgtCHN52_27_00382. All the mating type loci were nearly identical between isolate and formae speciales with no more than one SNP difference. Both mating types were found in these 11 genomes, four isolates with MAT 1-1-1, and seven with MAT 1-2-1. The mating type locus is on chromosome 1. Sequences flanking the mating type loci are highly collinear between isolates of the same mating type, with the exception of occasional TE polymorphisms. In contrast, a region of ~200 kb surrounding the mating type genes do not align between isolates of different mating types.

Chromosomal and structural re-arrangements within B.g. tritici

An interchromosomal translocation was found between the isolate Bgt_JPN_CHIKA (belonging to the Japanese wheat powdery mildew population) and the isolates Bgt_USA_2 and Bgt_CHN_52_27, which belong to the respective parental populations of the hybrid Japanese wheat powdery mildew population (Fig 3A) [7]. This interchromosomal translocation affected a ~ 640 kb segment containing 55 genes that was moved from chr-05 of Bgt_CHN_52_27 to chr-01 of Bgt_JPN_CHIKA, starting at ~3.5Mb of chromosome 5. This segment corresponds to the left end of chr-01 in the Bgt_JPN_CHIKA genome. This translocated region is also in chr-05 in the Bgt_USA_2 genome (Fig 3A). Thus, we assume that it is a recent translocation in the Bgt_JPN_CHIKA genome (Fig O in S1 Appendix). Furthermore, another translocation was found between chr-05 of Bgt_CHN_52_27 and chr-02 of Bgt_JPN_CHIKA, which amounts to approximately 242 kb and includes 27 genes (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Examples of chromosome collinearity and re-arrangements within B. g. tritici.

(A) Example of synteny and translocations in the genomes of three isolates (two coming from the parental populations, USA and CHN, and the middle genome coming from an isolate of the hybrid population of JPN powdery mildew. (B) Example of synteny in the genomes of two isolates from the same population (CHN powdery mildew). The petrol lines on the chromosomes indicate genes. The triangle and the “NNNs” show the position of a sequence gap.

https://doi.org/10.1371/journal.ppat.1013196.g003

There are also re-arrangements between isolates coming from the same powdery mildew population (i.e., the Chinese powdery mildew isolates Bgt_CHN_52_27 and Bgt_CHN_17_40) (Fig 3B). In chr-11 of Bgt_CHN_17_40, there are two segments with possible re-arrangements between chr-01 and chr-02 of isolate Bgt_CHN_52_27. Despite high contiguity scores from CRAQ analysis for the new assemblies (Table D in S2 Appendix), we examined the breakpoints of translocations for the presence of sequence gaps for all the studied re-arrangements. We only found one gap next to a potential translocation breakpoint in chr-02 of Bgt_CHN_52_27 (position: 10,455,965 bp, Fig 3B), which is why a possible miss-assembly cannot be excluded here.

Such small structural re-arrangements between isolates are expected in a fungus with an annual sexual recombination cycle. However, more pronounced re-arrangements were found between different f. sp. isolates, as will be described hereafter.

Chromosome 11 shows high sequence diversity and the presence of potentially dispensable chromosome segments

In B.g. tritici, B.g. dicocci and B.g. triticale isolates, chromosomes 1–10 show higher sequence conservation (ANI median ~99.38%, Fig P in S1 Appendix), while conservation is lower between B.g. tritici genomes and Bgs_1459 (ANI median ~98.65%, Figs P and Q in S1 Appendix and Table G in S2 Appendix), reflecting their phylogenetic distance (Fig L in S1 Appendix). However, striking differences were found in chr-11: B.g. tritici, B.g. dicocci and B.g. triticale isolates show an ANI median value of ~ 99.16%, while the ANI median value was lower (median ~96.98%) in comparison with the Bgs_1459 chr-11(Figs 4, P in S1 Appendix, examples in Fig Q in S1 Appendix). Additionally, sequences on the right arm of chr-11 are less conserved than on the left arm (Fig 4A), both between B.g. tritici isolates and between B.g. tritici and B.g. secalis (Fig 4B and 4C).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Analysis of structural variation and sequence diversity on B. graminis chr-11.

(A) Synteny in chr-11 for the hybrid B. g. triticale genome and its two-parent f. sp. genomes. The petrol lines on the chromosomes indicate genes. The colour of the names of the isolates corresponds to the forma specialis (black = tritici, darker grey = triticale, lighter grey = secalis). (B) Sequence conservation along chromosome 11. The heat map shows comparisons of B. graminis isolates with the reference isolate Bgt_CHE_96224 in 50 kb windows. Bgtl_THUN_12 and Bgd_ISR_211 were excluded from this analysis. (C) The red line shows the average sequence conservation for all 50 kb windows among B.g. tritici isolates, while the gray line shows sequence conservation between Bgt_CHE_96224 and Bgs_1459. Note that sequence conservation is generally lower in the right chromosome arm. (D) Ratio of sequence conservation between B.g. tritici isolates and Bgs_1459. (E-H) Dot plot comparisons of chr-11 from B. graminis isolates. B.g. secalis and B.g. dicocci both have a segment at the left end (Chr11_LE, blue box) which is absent in B.g. tritici. Furthermore, a terminal segment of ~100 kb is absent in Bgt_CHN_17_40, suggesting it contains accessory genes.

https://doi.org/10.1371/journal.ppat.1013196.g004

Noteworthy are the termini of chromosome 11: as previously described, the B.g. triticale genome is a recombined mosaic of the B.g. tritici genome (~87,5%) and the B.g. secalis genome (~12.5%) [5,42]. In particular, in a previous study [42], the beginning of the left arm of chromosome 11 was found absent in B.g. tritici isolate Bgt_CHE_96224. Here, we studied this in more detail, and indeed found that the terminal ~400 kb in Bgtl_THUN_12 originated from B.g. secalis, while the rest of the chromosome shows stronger synteny with B.g. tritici.

We found that the B.g. secalis and B.g. dicocci isolates both have this segment at the left end (Fig 4E, chr11_LE), while it is absent in all B.g. tritici isolates. Conversely, an ~ 100 kb segment at the right end of chr-11 is absent in B.g. secalis when compared to B.g. tritici (Fig 4E, Chr11_RE). Interestingly, this segment is present in the B.g. dicocci isolate (Fig 4G-4H), suggesting B.g. dicocci might represent the ancestral state as it has both the segment on at the left and right end of chr-11. Furthermore, a terminal segment of ~100 kb on the right end of the chromosome is absent in Bgt_CHN_17_40 (Fig 4F, Chr11_RE). This segment is also absent in Bgt_CHN_52_27 (Fig R.A in S1 Appendix). Additionally, isolate Bgt_CHN_52_27 is missing a ~ 180 kb segment on the left side of chr-11 (Fig R.A in S1 Appendix), while Bgt_ARG_4_2 has an extra segment of about ~150 kb on the left side compared to Bgt_CHE_96224 (Fig R.B in S1 Appendix). We, therefore, propose that the termini of chr-11 contain genes, which are dispensable for general pathogenicity on Triticeae host, but may be involved in host specificity.

We also want to emphasize that sequence alignments are generally poor in centromeric regions in all chromosomes. Examples are shown in Fig 4E through 4H, where centromeres are the highly variable regions. This is likely due to high number of polymorphic insertions of the centromere-specific RII_Fuji retrotransposons (Figs 4E-4H, and F in S1 Appendix). Our previous study showed that centromeric regions are defined by absence of genes and high enrichment of RII_Fuji retrotransposons [13]. The poor alignments of centromeres suggest high RII_Fuji activity and thus high sequence turnover. This also explains the abundance of nodes in the pangenome graph in centromeric regions (Fig M in S1 Appendix, examples in Fig 5A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Analysis of sequence variation on chromosome 11.

(A) Comparison of pangenome graphs for chromosomes 1 and 11. Approximate location of centromeres are indicated. Note that the right arm of chromosome 11 contains a higher number of nodes and edges than the left arm and than chromosome 1 overall. Additionally, both ends of chromosome 11 contain multiple edges reflecting the presence/absence of large segments between isolates (see also Fig 4). (B) Copy number variation (CNV) along chromosomes 1 and 11. The x-axis indicates the position while the y-axis shows the number of B. graminis isolates that show copy number variation in a given gene. Circles indicate multiple copies while triangles indicate deletions. Genes were separated into effectors (red) and non-effector genes (gray). (C) Number of identified protein variants in 399 B. graminis isolates per gene, normalized for gene length. Data sets for effector and non-effector genes are shown separately, so are data for the left and right arms of chromosome 11. Boxes indicate the inter-quartile range (IQR) with the central line indicating the median and whiskers indicating the minimum and maximum without outliers, respectively. Outliers were defined as minimum – 1.5*IQR and maximum + 1.5*IQR, respectively. Whiskers indicate the standard error.

https://doi.org/10.1371/journal.ppat.1013196.g005

The right arm of chromosome 11 is enriched in effectors and shows high levels of sequence variation

The right arm of chromosome 11 stands out for its diversity with respect to gene-density and content of effector genes. While all chromosomes, including the left arm of chromosome 11 contain between 6.4 and 18% effector genes, the right arm of chromosome 11 has 31 effectors out of a total of 44 genes (70.5%). Additionally, the pangenome graph shows a high number of nodes and edges compared to other chromosomes (Fig 5). Moreover, overall gene density is roughly half that of other chromosomes (Fig 5). Note that for these calculations, we excluded previously defined centromeric regions [13], in order not to distort the numbers for shorter chromosomes.

Using Illumina sequence read coverage for the 399 mildew isolates, we identified gene presence/absence polymorphisms (see methods). Generally, presence/absence polymorphism are found along all chromosomes, with effector genes having more than twice the rate of deletions and duplications as non-effector genes (Figs 5B, and S in S1 Appendix). Still, the right arm of chromosome 11 has more than double the rate of presence/absence polymorphism than chromosomes overall (Fig S in S1 Appendix). Interestingly, genes on the right arm of chromosome 11 also have the highest numbers of protein variants across the 399 isolates (Fig 5C). For this analysis, we predicted protein sequences for all genes across the 399 isolates from SNP data derived from mapping Illumina reads to the Bgt_CHE_96224 reference genome. It is therefore possible that some isolates contain additional genes and gene variants that are not present in the reference genome.

We produced a genealogic tree of all the predicted candidate effector proteins found on the right side of the chromosome arm of the Bgt_CHE_96224, Bgd_ISR_211 and Bgs_1459 isolates (Fig T in S1 Appendix). We also included the effectors found in the exclusive non-B. g. tritici left chromosome arm segment. Most of these genes belong to the previously described large effector family E003 [13]. Furthermore, we included effector proteins from chr-06 that had homology with candidate effector Bgs1459–09640, a gene that is found on the left end segment in B.g. secalis. This group of effector genes has apparently seen an expansion specifically in B.g. dicocci (Fig T in S1 Appendix). Furthermore, a gene Bgs_09664 from the left end of chr-11 of B.g. secalis, groups with homologs from the right end of chr-11 in B.g. tritici (Fig T in S1 Appendix), indicating that the ancestral chr-11 had E003-family effectors on both ends, which underwent differential loss in the different formae speciales (with 28 E003-family effectors in chr-11 for Bgt_CHE_96224, 20 for Bgs_1459 and 33 for Bgd_ISR_211).

Finally, three additional smaller segments which originate from B.g. secalis in B.g. triticale were found in the ~ 1.6 Mb at the left end of chr-11 (Fig 4A). This segment contains a candidate effector gene that is only found in B.g. secalis and B.g. triticale (Bgs1459–09656 and BgTH12–07698 respectively), but is absent in B.g. tritici (Note A and Fig U in S1 Appendix).

We hypothesize that presence/absence of chromosome segments, differential gene losses and family expansions of candidate effectors in chr-11 may be among the factors that contribute to host specificity.

A variety of gene duplications and deletions in the B. graminis pangenome

The structural variations in chr-11 inspired us to further study presence and absence of genes in the pangenome. Here, we used short read sequence coverage data mapped on to the Bgt_CHE_96224 reference genome to identify possible gene copy number variants (CNVs). If coverage with sequence reads was particularly high for a gene in a given isolate (i.e., roughly multiples of the average sequence coverage), we took this as evidence for multiple copies of that gene. Using a normalized coverage cutoff of 0.1 < coverage < 1.7 (guided by gene coverage, see Fig V in S1 Appendix and Methods), we identified 66,856 individual CNVs for the various isolates (Figs 6A, W in S1 Appendix). We observed that the number of duplicated genes was statistically significantly higher in a few populations (e.g., EGY population) and between formae speciales (Fig 6A). For example, B.g. secalis, has approximately 2.5 times more gene duplications than the B.g. tritici isolates, while B.g. dicocci has approximately 1.5 times more. These results could be attributed to the larger genomic differences between the different formae speciales, but also to using Bgt_CHE_96224 as the reference genome, which creates a reference bias and possible underestimation of CNV frequencies. Furthermore, even within B.g. tritici isolates, we found that populations from the greater Fertile Crescent area were among the most diverse (Fig 6A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Duplications/Deletions of genes using the populations and the pangenome.

(A) Number of gene duplications per population. The asterisks signify statistical significance of difference between other populations and the TUR powdery mildew population. (B) PCA of effector gene presence/absence summary per isolate, including the B.g. secalis isolates (left) and without them (right). (C) Estimate of the size of the core genome of non-effector genes in B. graminis. A sample curve was generated by adding isolates incrementally and counting the number of genes conserved between all of them. Because isolates differ in their levels of similarity, all possible permutations were performed. Two sample curves were generated, one where the B.g. secalis isolate was excluded (red) and one where it was included (blue). The dashed purple lines indicate the extrapolated core genome size, assuming that the sample curve was asymptotic. (D) The same analysis for candidate effector genes. (E) Summary bar plot with sizes of core and accessory genomes for the isolates analyzed. Numbers of accessory effector and non-effector genes are indicated above and to the right of the bars, respectively. (F) Expansion or shrinkage of effector families in the pangenome. Number of effector genes per family in all 11 B. graminis isolates. The code “Bgt_” is omitted from all the B.g. tritici isolates here.

https://doi.org/10.1371/journal.ppat.1013196.g006

We identified 3,795 genes with indications for CNV in at least one isolate. Fixation rates of presence/absence polymorphism in individual populations were relatively low, ranging from TUR, where only two duplications and two deletions were fixed in the whole population to B.g. secalis, where 211 deletions and 142 duplications were fixed in all isolates.

We created a PCA from a matrix of sequence coverage of each effector gene per isolate for all isolates including the B.g. secalis ones (Fig 6B, left). The latter isolates clearly differ from the rest of the B. graminis isolates. If we exclude the B.g. secalis isolates, the PCA (Fig 6B) shows that even though most populations cluster separately (e.g., B.g. dicocci), they seem to overlap. Additionally, further examination revealed differences between populations and gene families (Fig X in S1 Appendix), especially in the case of effector genes (Fig X.A in S1 Appendix). Furthermore, B. graminis isolates other than B.g. tritici tend to have higher number of CNVs as expected for more distant isolates.

Estimation of the size of core genomes for effector and non-effector genes

We estimated the size of the core genome by determining the number of bi-directional closest homologs that are conserved across Blumeria isolates (see methods). We estimated the size of the core genome through a sample curve in which increasing numbers of isolates were added, and numbers of genes conserved in all were counted (Fig 6C). Because isolates differ in their level of similarity to one another, all possible permutations were made for each step of the sample curve (Fig 6C). Excluding different isolates from the sampling showed that inclusion of the B.g. secalis isolate 1459 led to lower core genome size estimates, which was not entirely surprising, since isolate Bgs_1459 represents a different forma specialis. Consequently, we did not include Bgtl_THUN_12 in this analysis as it is a hybrid between B.g. tritici and B.g. secalis.

From the resulting sample curves, we extrapolated the size of the core genomes. We emphasize that the extrapolations have to be taken with caution since the sample curves do not form ideal mathematically asymptotic curves. For this analysis, we treated effector and non-effector genes separately. As shown in Fig 6C, the sample curves indicate that the core gene number of non-effector genes of B.g. tritici and B.g. dicocci converges at approximately 7,228 genes, while the inclusion of the B.g. secalis isolate suggests a core genome of approximately 7,079 genes, which is a difference of less than 2%. In contrast, the core effector genome of B.g. tritici and B.g. dicocci comprises ~721 genes, while inclusion of B.g. secalis indicates a core genome of about 615 effectors, an approximately 15% lower number (Fig 6D). In the isolates analyzed here, the B. graminis core genome (i.e., genes conserved in all isolates) comprises 7,086 non-effector and 624 effector genes (Fig 6E). Candidate effector genes make about 8.9% of the core genome while they account for between 14.9% and 19.3% of the accessory genome (Fig 6E). The fact that inclusion of B.g. secalis lead to a lower estimate of the core genome is not entirely surprising, since isolate Bgs_1459 represents a different forma specialis.

Because the effectors showed high variability between isolates, we then searched the individual genomes to identify potential effector gene enrichments in specific isolates (Fig 6F). The B.g. secalis isolate 1459, even though it seems to have gene deletions in many of the effector families, also shows expansions of a few effector gene families (e.g., E027, E029, E042, E055, E080, E094, E114, E120, E147, Fig 6F). Furthermore, Bgtl_THUN_12 has an exclusive expansion of effector gene family E135, where Bgtl_THUN_12 has four effectors, while the other 10 isolates have only one (Fig 6F). Additionally, the isolate Bgt_JPN_CHIKA has a specific expansion of family E012 (Bgt_JPN_CHIKA has 17 members, while others have between 11 and 13) (Fig 6F). Also, for some isolates of the same powdery mildew populations (CHN), there can be expansions in one isolate for one family whilst not in the other (i.e., E001 and E005).

An interesting effector family is E014, with the identified avirulence gene AvrPm2 as a member [49]. The AvrPm2 gene (BgtE-5845) was absent only in two of the pangenome assemblies (Table J in S2 Appendix), i.e., isolates Bgt_CHN_52_27 and Bgtl_THUN_12, as was also shown previously for some of these isolates [49,50]. Family E014 has more members in some isolates (e.g., Bgt_ARG_4_2), while isolate Bgs_1459, has a reduction of gene number. Despite its smaller number of genes, Bgs_1459 has some effectors in this family that are unique and do not have homologues in the other isolates (i.e., Bgs1459–02312), as shown in a phylogenetic network of this family for all the pangenome isolates (Fig Y in S1 Appendix).

Unequal homologous recombination is a main driver of gene presence/absence polymorphisms

To study molecular mechanisms that lead to presence or absence of genes, we manually analyzed 128 insertions larger than 2 kb that were found in at least one of the 11 PacBio genomes. To ensure that orthologous loci were compared, we required that regions flanking insertion aligned over a length of at least 20 kb (see methods). Insertions that represented the same event in different isolates and those that either contain no genes, or genes that were wrongly annotated as TEs were removed, leaving a set of 14 high-confidence alignments with gene-containing insertions. Interestingly, seven of them involved predicted effector genes.

In ten cases, we could show that unequal homologous recombination (also referred to as “unequal crossing-over”) was the likely molecular mechanism. In nine cases, recombination was caused by TEs, five by small interspersed nuclear elements (SINEs) and four by long terminal repeats (LTRs) of retrotransposons. In these cases, TEs from the same family and in the same orientation flanking a gene presumably served as templates for unequal homologous recombination (example in Figs 7, Z in S1 Appendix). In our dataset, TE sequences ranging from 157 to 883 bp served a recombination templates. In all cases, we could narrow down the region where the recombination occurred, based on diagnostic sequence polymorphisms (Figs 7C, Z in S1 Appendix), in some cases down to a few bp (Fig Z.B in S1 Appendix). In one case, recombination occurred in the promoter of two neighbouring genes (Fig Z.C in S1 Appendix). Three additional cases had templates of 1–15 bp flanking the deleted regions. Here, it was likely that the deletion was caused by a double-strand break that was repaired via the single-strand annealing pathway [51].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Example for a gene presence/absence polymorphism caused by unequal homologous recombination.

(A) The Dot-plot alignment of the region carrying the polymorphism. Gene and TE annotation for the genome that carries additional sequence is shown above the plot with transcriptional orientation indicated by arrow heads. For easier visibility, the RSX_Yak TEs (red and blue) that served as recombination template are connected by shaded areas to the corresponding point in the plot. (B) Schematic of the molecular mechanism by which gene BgtE-20036 was deleted through unequal crossing-over. (C) Multiple alignment of the sequence that served as template. The recombinant sequence (which is found in the isolate that carries the deletion) is Rec_TE. Note that the Rec_TE sequence shares diagnostic SNPs with RSX_Yak-1 (red) in the 5’ region and with the RSX_Yak-2 (blue) in the 3’ region. The region where the recombination must have occurred is indicated.

https://doi.org/10.1371/journal.ppat.1013196.g007

For all 14 loci, we surveyed the Illumina data for all 399 isolates. We found that fixation levels of the presence/absence of the genes are usually below 50%, except for one locus (containing genes Bgt-51752 and Bgt-51753), where the absence of the genes was nearly or completely fixed in the ARG, IRN and Bgdic populations.

Transposable element insertions correlate with host specialisation and divergence of individual powdery mildew populations

Since TEs have been shown to play an important role in powdery mildew evolution [13] and in host-pathogen interactions [52,53], we analysed the TE composition in the B. graminis pangenome, as well as in seven of the wheat powdery mildew populations (ARG, CHN, B.g. dicocci, EUR, IRN, ISR and USA populations). All 11 isolates that are part of the pangenome have similar repeat content of approximately ~55% of the genome (Table K in S2 Appendix) and similar abundance of TE superfamilies (Fig AA in S1 Appendix). On average, ~ 21,000 full-length and classified TEs were annotated per isolate (Fig 8A). DNA (class 2) transposons make around 8.5% of annotated copies. SINEs (Short Interspersed Nuclear Elements) account for about 38% of the annotated TEs. LINEs (Long Interspersed Nuclear Elements) account for about 14%, while different LTR retrotransposon superfamilies account for more than 40% of annotated TEs. All TE families present in our library were found in all isolates, albeit at different abundance levels (see below). We found no evidence for TE families that are present only in certain isolates or groups of isolates.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 8. TE analyses in wheat powdery mildew.

(A) Annotated copies of TE types in the different isolates of the pangenome. (B) Number of all TEs found in various distance ranges downstream and upstream of genes for all the isolates of the pangenome, see legend of Fig 8A. The bars with the dotted lines refer to distance from effector genes, while the ones without any lines refer to the non-effector genes. (C) Boxplot of the number of TIPs per isolate for populations with eight or more isolates (isolates with less than 2000 TIPs are shown, extreme outliers were excluded, because they were interpreted as technical artifacts). The “X” symbol refers to the population that the statistical comparisons have been made, were the “ns”, “*”, “**”, “***”, and “****” show significant difference with said population. (D) Analyses of TE insertions (TIPs) in the different wheat powdery mildew populations normalized for sample size for each population. Depicted are the 17 TE families that showed most insertion polymorphisms in all populations. (E) Phylogenetic network of the alignment of reverse transcriptase (RT) proteins for most of the LINE TE elements, using consensus sequences from isolate Bgt_CHE_96224. The TEs in the circle include the three LINEs with the most TIPs in the populations tested.

https://doi.org/10.1371/journal.ppat.1013196.g008

We found an enrichment of Helitrons in the B.g. secalis isolate, which has 779 annotated copies, compared to the other isolates which have between 166 and 580 (Fig 8A). Moreover, the hybrid B.g. triticale isolate that originates from Europe has an intermediate number of Helitrons compared to its ancestors B.g. secalis Bgs_1459 and B.g. tritici Bgt_CHE_96224. We then analysed TE composition at different distances up- and downstream of genes (Fig 8B). The majority of genes are at a distance of 1,000 bp or more from the closest annotated full-length TE, with less than 3,000 TEs found closer than 500 bp to a gene on average (Fig 8B). Interestingly, TEs are closer to the effector genes, with particular enrichment of TEs around 1–2 kb distance from the effector genes.

We used the software detettore [54] to detect transposon insertion polymorphisms (TIPs) using the paired end information of Illumina read mappings of isolates to a reference genome. We ran detettore on mappings of 166 B. graminis isolates against the isolate Bgt_CHE_96224 (see methods and Table L in S2 Appendix). The populations closer to the region of origin, along with the B.g. dicocci population had the most TIPs (Fig 8C). The CHN powdery mildew population had the least TIPs, which can be attributed to the shorter Illumina reads, posing a bias, as shorter reads can miss TIPs due to smaller mapping space. Furthermore, the hybrid JPN population has the most TIPs. This could potentially be attributed to its hybridisation.

When looking at specific TEs, SINEs and LINEs were found to be the most mobile TEs in the various wheat mildew populations with the RSX_Bgt_Yhi and RII_Bgt_Amitabha_C families being the two with the most polymorphic insertions observed across populations (Fig 8D). The three populations in the Fertile Crescent, along with the hybrid powdery mildew population from Japan, showed the highest numbers of TIPs for most families. After LINEs and SINEs, LTR retrotransposons are the ones with most polymorphic insertions, particularly the families RLC_Bgt_Gamera and RLC_Bgt_CH (Fig 8D). The top three LINE families with most TIPs against Bgt_CHE_96224, were found across the genome (Fig AB in S1 Appendix). An interesting TE family, RLX_Bgt_RXB had higher numbers of TIPs in the B.g. dicocci population compared to all the other populations studied. This was also supported by the higher number of copies of this TE in Bgd_ISR_211 (672 copies) compared to the other genomes of the pangenome (which have fewer than 662 copies) (Table M in S2 Appendix).

Many factors such as evolutionary pressure or genetic drift could have led to this increase [55]. While it is tempting to speculate that the high number of lineage-specific TE insertions may also have driven the host-specificity of B.g. dicocci, more studies are needed to support this hypothesis.

Although both SINEs and LINEs generally showed high TIP numbers, only LINEs encode a conserved protein (reverse transcriptase) [25]. This allowed the study of their relative divergence time. For this, we used the predicted reverse transcriptase proteins from multiple LINE families of the Bgt_CHE_96224 isolate. The proteins were used for a multiple sequence alignment with ClustalW and construction of a phylogenetic network with PhyloNet and splitstree (Fig 8E). Some LINE proteins clustered together and were also the families with the most TE insertions (e.g., RII_Bgt_Amitabha_C, RII_Bgt_Itera and RII_Bgt_Aves). This indicates that the LINE cluster that includes these TEs (Fig 8E, in dotted circle) could have evolved more rapidly compared to other LINEs and could also include some of the most active LINE TEs in these genomes.

Fungal isolates collection

The whole-genome sequenced isolates were introduced into the collection by reviving the ascospores in chasmothecia [69] following the methods of [13] (Table A in S2 Appendix). The isolates that were acquired through chasmothecia were single-cell colony isolated to ensure that only one clonal individual was collected. Each isolate was propagated on ten-day-old freshly cut wheat leaves (cultivar “Kanzler”) in Petri dishes. After four days, we picked with a toothpick a single colony to pass to new fresh leaves. We repeated this twice to make sure we have isolated one individual and then proceeded to propagate it, in order to get many spores for sequencing. The dates of collection of the isolates vary between 1990 and 2019 and the isolates are listed in the Table A in S2 Appendix. For each region, isolates have been collected from various fields and cultivars from Triticum aestivum and T. turgidum. A list of the isolates, the region, the coordinates of collection, the year of collection and the wheat species/subspecies that they were collected from is shown in the Table A in S2 Appendix. As outgroups for some analyses the forma specialis B. g. secalis was used.

DNA preparation and whole-genome sequencing

The DNA extraction method that was used, was done with a chloroform- and CTAB based-protocol modified from Bourras and colleagues [70]. Illumina paired-end reads sequencing of 150 bp read length and an insert size of ca. 500 bp was performed to generate 1–6 Gb sequence data per isolate using NovaSeq 6000 technologies only for the isolates sequenced for long reads (see below). The rest of the isolates were sequenced before with the same extraction protocol and their accessions were found on SRA. The Illumina sequence data is accessible from the NCBI Short Read Archive (the project numbers are shown in Table A in S2 Appendix).

Read mapping and variant calling

The raw Illumina reads were trimmed for adapter contamination and better sequencing quality. For trimming, the software Trimmomatic V0.38 [71] was used with the following settings: illuminaclip = TRuSeq3-PE.fa:2:30:10, leading = 10, trailing = 10, slidingwindow = 5:10, minlen = 50. The reference genome of B.g. tritici Bgt_CHE_96224 version 3.16 was used to align the reads by using the short read aligner bwa mem v0.7.17-r1188 [72] with the following settings: -M. We used samtools v1.9 [73] for sorting and removing duplicate reads. In order to mark PCR duplicates in the alignment (bam) files, we used the MarkDuplicates module of Picard tools version 2.18.25-SNAPSHOT. The average genome-wide coverage (calculated as the number of mapped reads multiplied by average mapped read length and divided by the reference genome size) ranged between x10 to x60 for all the isolates. We finally used ‘samtools index’ to create indices for the bam files. The mapped reads were also further checked visually for their quality and GC% content, using the software qualimap2 and the multi-bamqc function [74]. Isolates that were found to be clonal in a previous study where excluded [7] leading to 399 isolates used for most analyses.

The mapped reads were used to make variant calling files using GATK v.4.1.2.0 [75–77]. We started by using the HaplotypeCaller with the options: --java-options “-Xmx4g”, -ERC GVCF, -ploidy 1. Afterwards, we used CombineGVCFs to combine the single gvcf files into one file and then, with GenotypeGVCFs we performed joint variant calls on the merged variant gvcf file using the option: --max-alternate-alleles 4. We used the GATK VariantFiltration function to hard filter SNPs with quality thresholds recommended by GATK [76,77] and used with other plant pathogenic fungi [78,79]. The following thresholds were used: QUAL < 450, QD < 20.0, MQ < 30.0, -2 > BaseQRankSum > 2, -2 > MQRankSum > 2, -2 > ReadPosRankSum > 2, FS > 0.1. Then, we filtered for a genotyping rate (>99.9%) and removed the indels. Finally, we kept only biallelic SNPs using vcftools: --max-alleles 2 (version 0.1.5) [80]. We also filtered the vcf files for Minor Allele Frequency < 0.1. The number of SNPs retained was 60,258.

Population genomics analyses

To select the isolates for long-read sequencing we performed various population genetics analyses. We used R (v4.3.3) [81], and the R packages tidyverse (v2.0.0) [82], ggplot2 (v3.5.1) [83], and stats (v4.3.3) [81] for many of the following analyses. Inkscape (v1.3.2) was used to compile many figures [84]. We first created a map with the coordinates of the samples using R packages: sf (v1.0-18) [85,86], rnaturalearth (v1.0.1) [87], ggspatial (v1.1.9) [88], and ggrepel (v0.9.3) [89]. Map outlines were based on Natural Earth data (https://www.naturalearthdata.com/downloads/50m-cultural-vectors/). Then, in order to create a PCA, we used PLINK (v2.00a3 64-bit, 17 Feb 2020) to make an eigenvec and an eigenval file. We used the file eigenval to make the PCA variance plot (Fig A in S1 Appendix), and then the eigenvec file to visualise the PCA with ggplot2 in R.

Admixture analysis was performed using the ADMIXTURE (v1.3) software [90]. With the help of the Snakemake (v9.12.0), we run ADMXITURE for K equals from 2 to 20, with 20 runs for each K and different seeds, with the parameter “--haploid = “*””. The results were visualised using the pong software [91] and the cross-validation error was visualised using ggplot2 in R.

Furthermore, we measured the nucleotide diversity of populations using vcftools (v0.1.16) with the parameter “window-pi” for 10 kb non-overlapping windows, and then visualised the results with the packages ggplot2, and ggpubr (v0.6.0) [92,93]. We used again vcftools to identify the singleton SNPs (SNPs found only in one isolate/individual within a whole population) once using all isolates and another time with only eight random isolate per populations.

Finally, we used packages vcfR (v1.14.0) [94] with the command “read.vcfR” to input the vcf file in R, then “vcfR2genlight” to convert the object, then StAMPP (v1.6.3) with the command “stamppNeisD” to find the genetic distances between isolates and between populations [95], we inserted the geographic coordinates and transformed them into distances with “dist” from the stats package [96], we then created a two dimensional kernel density heatmap with the command “kde2d” from the package MASS (v7.3-60) [97], with the support of the packages grDevices (v4.3.3) and graphics (v4.3.3) [81] to study the genetic and geographic correlation between the isolates. The package rcartocolor (v2.1.1) was used to find colour-blindness friendly colours used in many of the plots [98].

PacBio sequencing, gene and TE annotation

In addition to the Illumina sequencing, we long-read sequenced with the PacBio technology eight powdery mildew isolates (six B.g. tritici: Bgt_ARG_4_2, Bgt_USA_2, Bgt_JPN_CHIKA, Bgt_IRN_GOR_2, Bgt_CHN_52_27, Bgt_CHN_17_40, one B.g. dicocci: Bgd_ISR_211, and one B.g. secalis: Bgs_1459). There was 1x SMRT Cell 1M reads used per isolate, apart from isolate Bgt_CHN_17_40, where 8M reads with PacBio HiFi were used. The reads from the isolates were then initially assembled using SMRT Link software tools (for versions see in Table N in S2 Appendix) at the Functional Genomics Center Zurich (FGCZ) in Switzerland. The assembled contigs were then scaffolded into near chromosomes by aligning them to the reference genome of isolate Bgt_CHE_96224, Bgt_genome_v3_16 [13]. Each contig was placed and orientated by its best blastn hit (first: e-value, second: bit-score). Contigs were merged with stretches of 200 “N” insertions (sequence gaps). Unplaced contigs are shown in Table E in S2 Appendix. We evaluated the quality of the new genome assemblies using Clipping Reveals Assembly Quality (CRAQ) (v1.10) [99]. We also used the already sequenced and assembled genomes of two B.g. tritici (Bgt_CHE_96224, Bgt_ISR_7) and one B.g. triticale (Bgtl_THUN_12) isolate for the pangenome analyses (Table O in S2 Appendix). One recently sequenced genome in the European wheat powdery mildew population (isolate CHVD_042201) was not used [100], as it was not accessible at the time this analysis was done. The assembled genome data is accessible from NCBI (the project numbers are shown in Table A in S2 Appendix). The difference between these genome was evaluated using the OrthoANIu tool (v1.2), a standalone average nucleotide identity (ANI) calculator [101]. The mitochondrial genome, the coding sequences (CDS) and individual chromosomes were also evaluated in pairs of isolates using OrthoANIu.

The gene annotations were performed using maker (v2.31.10). The annotations were generated by using the Bgt_CDS_v4_23 proteins as a template. Maker was used with the prot2genome function and the repeats and transposons were found with maker internal TE_proteins, as well as the initial annotations in PTREP and nrTREP databases [102]. On a second maker round, the annotation was created based on the B. hordei gene annotations of isolates DH14 and RACE1 (GCA_900239735.1 & GCA_900237765.1). From the resulting genes in the second round, only genes with no annotation in their corresponding loci in the first round (non-redundant genes) were added to the final annotation (as extra genes). In order to assess the genome and assemblies and annotations we used BUSCO (v5.4.4) [103,104] along with some further analyses below. We looked at the size of chromosomes and created a phylogenetic network using Splitstree4 (version 4.19.1, 27 Jun 2023) [105], and 100 number of replicates for bootstrapping, with the orthogroups within B. graminis (Fig L in S1 Appendix) via Orthofinder (v2.5.5) and the parameter “-msa” [106]. The number of genes and their type (effector, non-effector etc) was visualised, along with the protein size distribution to further check for the quality of the annotation (Fig I in S1 Appendix).

TEs were further annotated by first using the TE database for Blumeria graminis nrTREP (version 25) [102] and the software RepeatMasker employed by EDTA (v.2.0.1) [107]. We then manually curated newly identified TEs and performed a final annotation with the latest Blumeria repeats deposited in nrTREP (Table P in S2 Appendix).

The mitochondrial genome sequences were identified by using blastn of the mitochondrial contig of the B. hordei DH14 mitochondrial genome to the genome assembly of isolate Bgt_CHN_17_40. The mitochondrial genome was in one contig. We then, used the software MITOS2 (v2.1.9) to annotate the one mitochondrial genome with number four genetic code (Mold, Protozoan, Coelenteral [108]. The annotated output was visualised using OrganellarGenomeDRAW (v1.3.3) [109]. We also used the software Mfannot [110] (online version) with the same genetic code to re-annotate three mitochondrial genomes of interest for comparison using the previously annotated Bgt_CHE_96224 mitochondrial genome [47] and the already assembled Bh_DH14 mitochondrial genome [12].

Synteny in the pangenome

In order to study any structural re-arrangements in the B. graminis pangenome, we looked at the synteny between the chromosomes and the various isolates with a Minimum Alignment Length set to 50 kb (for the individual chromosomal re-arrangement visualisation in Fig 3A and 3B) or 100 kb (for the overall chromosomal synteny in Figs 2, and 4A). For the synteny between the isolates Bgt_CHE_96224 and Bh_DH14 (Fig K in S1 Appendix), the Minimum Alignment Length was set to 10 kb, since these genomes are more dissimilar than the genomes within the B. graminis species. To perform these alignments, we used mummer (v4.0.0rc1) [111] between genome pairs through the command “GetTwoGenomeSyn.pl” from the software NGenomeSyn (v1.41) [112], using the appropriate minimum alignment lengths (as stated above) and the default other parameters. We then used “NGenomeSyn” command to visualise the synteny. We used the Paul Wong colour accessible palette the various isolates and populations across the rest of the study [113].

Analysis of sequence conservation between chromosomes

To study sequence conservation between isolates along chromosomes, 1 kb segments of the reference isolate Bgt_CHE_96224 were used in blastn searches against the corresponding chromosomes of all other isolates. Blastn hits were then filtered for collinearity, requesting that a given 1 kb segment had its top blastn hit within 5% of its position in the reference chromosome. Additionally, blast hits in reverse orientation were filtered out. With this, we filtered out non-specific matches stemming from repetitive sequences. No other filtering was used, because we did not have any top blastn hits shorter than 300 bp, and even between formae speciales (e.g., between Bgt_CHE_96224 and Bgs_1459), ~ 94% of top blast hits were longer than 500 bp and ~79% were longer than 900 bp. Sequence conservation between Bgt_CHE_96224 and a given isolate was then calculated as a running average over 50 of the colinear 1 kb segments. Overall sequence conservation across all B.g. tritici isolates was then calculated as the average of all 50 kb windows in the B.g. tritici isolates. This was used to produce the conservation plots shown in Figs 4C and Q in S1 Appendix. Finally, the ratio of sequence conservation between B.g. tritici isolates, and B.g. secalis was calculated for all windows, resulting in the plots in Figs 4D and Q in S1 Appendix.

Chromosome-scale dot plots were produced by running blastn searches of 1 kb segments of a chromosome of one isolate against the corresponding chromosome of another. Positions of the hits in both chromosomes were recorded and used for the dot plots shown in Fig 4.

Finally, to study phylogeny of candidate effectors from chr-11 and selected homologs from other chromosomes, we created a dataset with all the candidate effector identified proteins from the right chromosomal arm of Bgs_1459, Bgd_ISR_211, and Bgt_CHE_96224, along with the effectors on the left end region that is missing in the B.g. tritici isolates and homologous effectors of Bgs1459–09640 found in chr-06 after using blast (v2.7.1+) [114] on these three genomes to find any homologous protein. We excluded three proteins that had a size smaller than 50 amino acids (three were excluded). We added the unclassified candidate effector protein Bgs1459–00747 as an outgroup. We aligned the dataset with ClustalW (v.2.1) [115,116], and constructed a phylogenetic tree using the Bayesian inference with MrBayes (v.3.2.7a) [117,118] with settings nst = 6 and rates = invgamma. The Markov chain Monte Carlo (mcmc) analysis was run until the probability value decreased to under 0.02 with a sampling frequency of 10 and a burn-in of 25% of samples. We visualised the tree with Figtree (v1.4.4) [119] and we used colours from Paul Tol’s accessible palettes (https://sronpersonalpages.nl/~pault/).

Identification of specific candidate effectors

The study of some specific effectors of interest was made by first changing the annotation files gff to gtf format using gffread (v0.12.8) [120]. We then used STAR (v2.5.2a) [121] to map already published RNAseq data (namely SRR6410427, a replicate of RNAseq data of isolate Bgtl_THUN_12 on the Timbo triticale line 2 days post infection). Using IGV (v2.15.4, 12/08/2022) [122], we manually checked the expression and the annotation of specific genes BgTH12- and Bgs1459–0. After using blast (v2.7.1+) [114] to find the closest homologous proteins on the whole pangenome and on the B. hordei isolate DH14, we aligned them with ClustalW (v.2.1) [115,116] and visualised the alignment with Jalview (v2.11.1.4 + dfsg-3) [123]. A phylogenetic tree was constructed using the same method as described above.

Identification of gene duplications and deletions

Identification of potential gene deletions and duplications based on Illumina sequence read coverage from the 399 isolates (including B.g. secalis/triticale/dicocci isolates was done as previously described [124]. The strategy here was to infer gene deletions from a lack of mapped reads for a given gene in a given isolate, and duplications from higher than expected coverage. To establish formal thresholds to determine gene copy numbers, read coverage was calculated for all individual genes across all 399 mildew isolates used in this study. For each gene, it was statistically tested whether reads coverage across the 399 isolates showed uni-modal or bi-modal distribution. The test was done with the diptest R package (v0.77-2) and a p-value of 0.01 was used as cutoff for bimodal distributions. Positions of the modes were determined for each gene with the LaplaceDemon R package (v16.1.6). For the selection of genes with bimodal distribution, identified modes had to be below 0.2 and between 0.8 and 1.2. This was done to exclude genes that showed more continuous distributions (e.g., due to poor mapping of divergent variants or paralogs). Distributions and determination of thresholds are shown in Fig V in S1 Appendix. Using this analysis, we determined that sequence coverage below 0.1 was indicative of gene deletion, while coverage above 1.7 indicated duplication. For duplications, we did not distinguish between two or more that two copies.

Core genome size estimates

To account for genes that might have been missed in the annotation of a given genome, the Liftoff software (v1.6.3) was used (obtained from github.com/agshumate/Liftoff/) [125]. Each of the 11 genomes was used as a target onto which annotations of the other 10 genomes were mapped. After each Liftoff mapping, new genes (i.e., genes that do not overlap with previously annotated ones) were added to the existing gff file of a respective genome. For all mappings, the “-copies” option of Liftoff was used to allow identification of duplicated genes. The cumulative Liftoff mapping of the 11 genomes was done with an in-house perl script “pan_genome_liftoff_multiple_genomes”. The resulting gff files for all 11 genomes were used to produce coding sequence (CDS) fasta files using the gffread software (v0.12.7) obtained from ubuntu repositories (unbuntu.com).

To estimate the size of the core genomes (i.e., genes that were found in all isolates), the CDS of all isolates were used in bi-directional blastn searches to identify bi-directional closest homologs. This was done in a consecutive way in that the CDS of the reference genome Bgt_CHE_96224 were used in bi-directional blastn searches against CDS of all 10 other genomes. The blastn searches were run with the in-house perl script “blast_bi_directional_from_db_list”. The result was a presence/absence table for all genes across all 11 genomes. This was transformed into a simple table where presence/absence was encoded as 1 or 0 with the script “pan_genome_PAP_matrix_from_bi_di_homologs”. To approximate the size of the core genome, a sample curve was derived, adding increasing numbers of isolates to the reference genome and determining the number of conserved genes. Because isolates show different levels of similarity, for each sample step all possible permutations were tested. This was done with the in-house perl script “pan_genome_mk_core_genome_permutations”. Sample curve data was used to calculate the convergence value (i.e., the size of the core genome) using asymptotic regression in R. Estimation of the size of the core genome was done for candidate effector and non-effector genes separately. All perl and R scripts used for this analysis are available on Github (https://github.com/wicker314/).

Detailed analysis of loci containing gene presence/absence polymorphisms

For the analysis of mechanisms that lead to presence or absence of genes between isolates, we focused on insertions in regions where chromosomes align well between given pairs of isolates. We split the chromosomes of one isolate into non-overlapping segments of 1000 bp and used them as queries in blastn searches against the other isolate. We then searched for regions where the top blast hits were consecutive for at least 20 kb, but were interrupted by at least 2 windows that had no hits in isolate one (i.e., isolate 1 had additional sequence).

Regions where isolate 1 had additional sequence were then cross-matched with the annotation (gff files) to check whether the additional sequence carried genes. This search was run until at least 100 insertions were found (the search was stopped after 128). The candidate regions were the excised from the respective genomes (insertion region plus 20 kb flanking sequence). The excised regions were analyzed manually by dot plot (programme dotter, obtained from ubuntu.com). In multiple cases contained annotation artifacts (e.g., TE segments annotated as genes), or the same insertions were detected in multiple isolated. After these were removed, a set of 14 high-confidence insertions remained. These were searched for signatures such as overlapping sequences at the insertion breakpoint in the isolate that carries the deletion (i.e., the putative template for unequal recombination). The template sequences were annotated manually to identify the type of repeat that served as template and aligned with ClustalW to identify diagnostic SNPs that allowed narrowing don the recombination point.

Analyses of transposable elements for the pangenome

We analysed positions of TEs relative to genes with the TEGRiP pipeline (github: https://github.com/marieBvr/TEs_genes_relationship_pipeline) [126,127]. In order to find TE insertions which are polymorphic between isolates, we used the software detettore (v.2.0.3) [54,128] and a subset of isolates (Table L in S2 Appendix). We used ClustalW (v.2.1) [115,116] to produce the alignments of the TE consensus sequences of families of interest, and then changed the format to nexus and checked the alignment with ClustalX (v2.1) [116,129]. We used MrBayes (v.3.2.7a) with settings nst = 6 and rates = invgamma to infer trees [117,118]. The Markov chain Monte Carlo (mcmc) analysis was run until the probability value decreased to under 0.01 with a sampling frequency of 10 and a burn-in of 25% of samples. We visualised the tree using FigTree (v1.4.4) [119].

Code availability

All R and original Perl scripts used in this study were deposited in our GitHub repository (github.com/Wicker-Lab/Blumeria_pangenome). There, we provide all R scripts that were used to produce the graphs in the main and supplementary figures, as well as the underlying data used for the plots. For Perl scripts, we provide detailed explanations of the process how they were used. Large datasets such as variant call files (vcfs) were deposited at Zenodo (https://zenodo.org/records/18946991).