Introduction

Understanding the dynamics and spread of plant pathogens is critical for the development of effective management strategies. Powdery mildews, a group of obligate parasitic fungi, exemplify this challenge, having a profound impact on a multitude of plant species, including economically significant crops and ornamentals such as wheat, grapes, blueberries, hops, peas, and strawberries. Recent advances in genetic research have shed light on the host specificity of these pathogens, suggesting a much more complex interaction with their hosts than previously understood (Vaghefi et al., 2022;Bradshaw et al., 2022aBradshaw et al., , 2024a,b;,b;Kusch et al., 2024). A comprehensive understanding of the taxonomy and evolutionary relationships of powdery mildews is essential, given their invasive behavior and potential to cause significant ecological and agricultural damage (Johnson et al., 1979;Bebber et al., 2014;Fuller et al., 2014;Kiss et al., 2020;Bradshaw et al., 2021).

Originating in Eastern North America, Erysiphe vaccinii Schwein. is a powdery mildew fungus that primarily infects the Ericaceae, showing a strong affinity for species of Vaccinium, notably blueberries (Braun & Cook, 2012). This pathogen poses a considerable risk to the blueberry industry by reducing crop yields and increasing reliance on fungicides. A noteworthy aspect of E. vaccinii, which was previously classified under the names 'Erysiphe vaccinii' and 'Erysiphe elevata', includes its infection of diverse hosts such as cultivated blueberries (Vaccinium corymbosum), Catalpa spp. (Bignoniaceae), and Eucalyptus spp. (Myrtaceae). Single-locus rDNA phylogeny previously grouped these pathogens into a highly supported clade named 'the Erysiphe vaccinii complex' (Tymon et al., 2022), highlighting their close genetic relationship and identical morphology despite infecting unrelated host species.

Dramatic growth in global production and trade of blueberries has been driven by mounting consumer demand and as blueberry production has expanded across the globe, so has its pests. In this study, we have detected E. vaccinii in blueberry-growing regions world-wide, marking its first reported occurrence outside North America (Bradshaw et al., 2024a) and signifying its status as an emergent pathogen. We uniquely combined a global collection of E. vaccinii from blueberry farms and herbarium specimens, some dating back over 150 yr. We sequenced six genetic loci from these specimens to shed light on the phylogeny, taxonomy, and population structure of this emerging threat. Our goals were to explore the diversity of the pathogen, its range of hosts, and the evolutionary dynamics of its spread over time and space. By integrating our findings into a 'living phylogeny' through an online phylogeny database (T-BAS), we aimed to simplify pathogen identification for both specialists and nonexperts, facilitating rapid response to outbreaks. Through this approach, we sought to better understand the spread of E. vaccinii and provide valuable information to the expanding blueberry industry to help better understand this pathogen and its world-wide impacts.

Materials and Methods

Morphological examinations

Morphological examinations were accomplished following Bradshaw et al. (2022b). Morphological examinations of the asexual morph of recently collected samples were accomplished by placing clear adhesive tape on powdery mildew colonies and setting the tape onto a slide containing a drop of water, or by doing hand sections and mounting them onto a slide with potassium hydroxide (KOH), Melzer's reagent (MLZ), or Congo red. Examinations of the sexual morph were accomplished by using a clean needle to mount chasmothecia onto a microscope slide containing water, KOH, MLZ, or Congo red. Photographs were taken of the slides using a compound microscope with an Olympus SC50 camera (Tokyo, Japan) and a Zeiss AX10 microscope.

Sequencing

DNA extractions were accomplished using the Chelex method (Walsh et al., 1991;Hirata & Takamatsu, 1996). The Polymerase chain reaction (PCR) was carried out for the Internal Transcribed Spacer rDNA (ITS) and Large Subunit rDNA (LSU) region using the primer pairs PM10/PM28R (Bradshaw & Tobin, 2020). If PCR was unsuccessful, a nested approach was applied using the primers AITS (Bradshaw & Tobin, 2020)/TW14 (Mori et al., 2000) followed by PM10/PM28R or AITS/PM11 (Bradshaw & Tobin, 2020) followed by PM10/PM2 (Cunnington et al., 2003). For the CAM, GAPDH, GS, and RPB2 region the primer pairs PMCAM1/PMCAM4R, PMGAPDH1/PMGAPDH3R, GSPM2/ GSPM3R, and PMRpb2_4/PMRpb2_6R were used (Bradshaw et al., 2022a). If these were unsuccessful for the GS and RPB2 regions, the following primers from Bradshaw et al. (2023) were used: EGS1/EGS2R and ERPB2_3/ERPB2_7R. For the TUB region the primers BTF5b/BTR7a (Ellingham et al., 2019) were used followed by ETUB2 and ETUB2R (Bradshaw et al., 2023). Often the GAPDH sequences were contaminated with Ampelomyces and as such the primers EGAPDH1/EGAPDH2 (Bradshaw et al., 2024b) were used if sequencing failed from PMGAPDH1/PMGAPDH3R.

Reference phylogenetic trees

A representative tree of taxa with a complete 7 loci dataset (except E. vaccinii f. eucalyptorum) from the E. vaccinii complex was generated from the concatenated ITS + LSU + CAM + GAPDH + GS + RPB2 + TUB sequences. In addition, singlelocus trees were constructed from all the newly generated sequences. Sequences were aligned and edited using MUSCLE in MEGA11 (Tamura et al., 2021). A GTR + G + I evolutionary model was used for phylogenetic analyses as it is the most inclusive model of evolution and includes all other evolutionary models (Abadi et al., 2019). The phylogeny was inferred using Bayesian analysis of the combined loci using a Yule tree prior (Gernhard, 2008) and a strict molecular clock, in the program BEAST v.1.10.4 (Suchard et al., 2018). A single MCMC chain of 10 7 steps was run, with a burn-in of 10%. Posterior probabilities were calculated from the remaining 9000 sampled trees. A maximum clade credibility tree was produced using TREEANNO-TATOR v.1.10.4 (part of the BEAST package). Stationarity was confirmed by running the analysis multiple times, which revealed convergence between runs. The resulting tree was visualized using FIGTREE v.1.3.1 (Rambaut, 2009). A maximum likelihood analysis was accomplished using RAXMLGUI (Silvestro & Michalak, 2012) under the default settings with a GTR + G + I evolutionary model. Bootstrap analyses were conducted using 1000 replications (Felsenstein, 1985).

A separate multilocus phylogeny was inferred for a total of 50 Erysiphe strains that were genotyped across seven loci (CAM, GAPDH, GS, ITS, LSU, RPB2, TUB). Multiple sequence alignments were generated for each locus using MAFFT, concatenated, and subjected to maximum likelihood phylogenetic inference using RAXML; support values for each node in the tree were calculated using 1000 bootstrap replicates. The concatenated maximum likelihood approach was compared to the Bayesian approach implemented in BEAST v.1.10.4 and to a multispecies coalescent model implemented in ASTRAL-III v.5.7.8 (Zhang et al., 2018). The best tree, alignments, and specimen metadata (e.g. host, locality) were uploaded to T-BAS (Carbone et al., 2017(Carbone et al., , 2019) ) to enable real-time phylogeny-based placement of unknown Erysiphe strains for any number of the seven loci in the reference tree. The reference tree is available for viewing and placement on the T-BAS guide page (https://guide-tbas.cifr.ncsu.edu/tbas).

Phylogeny-based placement

Multilocus sequence data for a total of 123 strains of E. vaccinii isolated from different hosts were phylogenetically placed into the Erysiphe reference tree using T-BAS and the EPA-NG v.0.3.8 algorithm (Barbera et al., 2019) implemented at CIPRES (Miller et al., 2015). In T-BAS, the placements are visualized and standardized using a Metadata Enhanced PhyloXML (MEP) format. This standardization allows placements and associated specimen attributes (e.g. host, locality) to be readily viewed, archived, and importantly analyzed within a phylogenetic context. For example, sequences can be filtered for a user-specified number of loci, and the corresponding multiple sequence alignments, associated metadata, and Newick tree can be downloaded for downstream analyses.

Population structure

Genetic admixture among formae was examined using STRUCTURE v.2.3.4 (Pritchard et al., 2000;Falush et al., 2003) and by using computational resources accessible via the CIPRES REST API (Miller et al., 2015). To reduce the potential for nonrandom missing data bias (Yi & Latch, 2022), the analysis was based only on 136 out of 173 strains with available sequence data for at least five of the seven reference loci (CAM, GAPDH, GS, ITS, LSU, RPB2, TUB). For each Erysiphe forma, estimators of population mean mutation rate (h) and pairwise nucleotide diversity (p) were calculated using the program SITES v.1.1 (Hey et al., 2018). Neutrality tests were performed to test for population size constancy using Tajima's D (Tajima, 1989), Fu and Li's D and Fu and Li's D* (Fu & Li, 1993), also implemented in SITES. Individuals were assigned to k = 10 possible clusters using PARALLELSTRUCTURE (Besnier & Glover, 2013) with three independent replicates for each k value. Membership probabilities were based on a MCMC burn-in of 50 000 steps followed by 100 000 sampling iterations. Optimal k values were examined using log e P(D) and delta K methods (Evanno et al., 2005) implemented in STRUCTURE HARVESTER v.0.6.93 (Earl & vonHoldt, 2011). Cluster membership results were visualized using histograms in outer rings of the phylogeny with placements results using the structure tool in DECIFR (https://tools. cifr.ncsu.edu/structure).

Network and haplotype inference

Network inference was based on parsimony and neighbor-joining methods implemented in TCS1.21 (Clement et al., 2000) andSPLITSTREE4 v.4.14.8 (Huson, 1998), respectively. In TCS, sequences were collapsed into haplotypes with gaps treated as missing data. Nodes were colored as a function of host and geographical location, and node size was proportional to haplotype frequency. The Neighbor-net algorithm implemented in SPLIT-STREE was used to further identify splits or bipartitions in the data, where the presence of multiple parallel edges between bipartitions indicates a history of recombination; in the absence of phylogenetic conflict, splits would be separated by a single edge. Recombination among formae was further tested using the pairwise homoplasy index (PHI) implemented in SPLITSTREE.

Multilocus isolation with migration

For each locus, a single recombination-free partition was extracted from the multiple sequence alignment using the four-gamete criterion to identify the largest interval of compatible single nucleotide polymorphisms, as implemented in the IMgc program (Woerner et al., 2007). The best time-ordered rooted topology of formae was determined by performing topology sampling runs with hyperpriors under an infinite-sites mutation model using IMA3 v.1.11 (Hey & Wakeley, 1997). Population parameter estimates of migration rates (2N e m), splitting times (t), and effective population size (N e ) were performed using the fixed best-rooted topology. Markov chain Monte Carlo (MCMC) runs were based on a burn-in of 1000 000 steps, 256 heated chains using a geometric heating scheme of parameters ha = 0.97 and hb = 0.8, 10 000 sampled genealogies, a mutation rate 1 9 10 À9 per base per generation, and a generation time of 1 yr due to the chasmothecia needing to perennate for ascospore discharge and germination (Salmon, 1900;Homma, 1937;Jailloux et al., 1998;Jarvis et al., 2002;Braun & Cook, 2012). Mutation rates can span several orders of magnitude ranging from 4.5 9 10 À7 mutations per base pair per generation in Blumeria graminis f. sp. tritici (Sotiropoulos et al., 2022) to 2.0 9 10 À8 in Schizophyllum commune (Baranova et al., 2015) and as low as 4.2 9 10 À11 in Aspergillus flavus ( Alvarez-Escribano et al., 2019). Because there are no reported estimates of mutation rates in Erysiphe we assumed a conservative estimate of 1 9 10 À9 per base per generation that is typical of many fungi (Edwards & Rhodes, 2021). In IMA3 simulations convergence in parameter estimates was based on two runs each with swapping rates > 0.9 and effective sample sizes in excess of 10 000. The best-rooted phylogeny was redrawn to include population parameter estimates and visualized using the IMFIG program (https://github.com/jodyhey/IMa3). All runs were performed using tools in DECIFR, working seamlessly with computational resources at CIPRES.

Results

Sample collection

During a comprehensive global survey of fresh material and herbarium specimens spanning Canada, China, Mexico, Morocco, South Africa, Peru, Portugal, and the USA, we have successfully sequenced 173 specimens of powdery mildew, all part of the E. vaccinii complex. These specimens were gathered from a wide array of host species including Catalpa bignoniodes, Catalpa speciosa, Epigaea repens, Eucalyptus camaldulensis, Gaultheria shallon, Gaylussacia baccata, Gaylussacia frondosa, Kalmia latifolia, various Vaccinium hybrids from the North Carolina State University (NCSU) blueberry breeding program, V. corymbosum, Vaccinium macrocarpon, Vaccinium myrtilloides, Vaccinium myrtillus, Vaccinium pallidum, Vaccinium parvifolium, and Vaccinium virgatum (Fig. 1; Supporting Information Table S1).

Economic cost of powdery mildew to the blueberry industry

The growers and Global Plant Health research colleagues participating in the study indicate that powdery mildew on blueberries was first identified on blueberry outside North America in Portugal in 2012, followed by detections in Honghe, China (2016), Hoedspruit, South Africa (2018), Viru, Peru (2019), Olmos, Peru, and Banna, China (2020), Baja, Mexico and Qujing, China (2021), and Barranca, Peru (2022) (Table S2). Although we reached out to growers and extension agents in Washington and Oregon, states where blueberries are widely cultivated and conditions are highly favorable for powdery mildew, no instances of infected blueberries were reported.

As of 2021 the global highbush blueberry planting area had reached 235 408 ha (Zang, 2022). Growers estimated the cost of powdery mildew control to range from $200 to $2250 per hectare (Table S2). As such, these figures suggest that powdery mildew could impose an annual cost of $47 to $530 million to the blueberry industry globally.

Morphology

Our analysis of powdery mildew specimens from outside eastern North America revealed a dominant asexual reproductive stage, marked by extensive conidiophore production (Fig. 2), throughout the growing season. Moreover, no sexual stage was observed on any blueberry specimens from these regions. This uniform presence of the asexual state suggests that a single mating type has been introduced globally from eastern North America. Notably, samples of E. vaccinii f. convertibilis displayed only asexual stages. Interestingly, North American samples within this clade were exclusively identified in greenhouse settings that were part of the NCSU blueberry breeding program. Analysis showed E. vaccinii f. vaccinii-corymbosi split into two subclades (Figs 3, 4, S2). One included all specimens from outside North America as well as a greenhouse specimen from the NCSU breeding program, all of which exhibited only the asexual state. This indicates that all of the strains proliferating globally and requiring significant management efforts (Table S2) propagate exclusively through asexual means. However, powdery mildew infecting blueberry farms in North Carolina exhibit only mild symptoms in midsummer followed by a prominent sexual reproductive phase that occurs late in the season. Although those farms report a high incidence of the disease (Table S2), there is only a minimal effect on crop yield. We hypothesize that the negligible impact on crop yield observed in North Carolina is likely attributed to environmental factors, and growing conditions (in California, China, Mexico, Portugal and Morocco blueberries tend to be grown New Phytologist (2025) www.newphytologist.com Ó 2025 The Author(s). New Phytologist Ó 2025 New Phytologist Foundation.

in hoop houses). It could also stem from the fungus manifesting earlier in the season with heightened virulence due to it only forming an asexual phase.

Phylogeny and population genetics

The multilocus reference phylogeny of Erysiphe is shown in Fig. 3. There was no significant difference in the topology of the tree inferred from three commonly used phylogenetic inference methods (Fig. S1). Phylogenetic, network, and multilocus isolation with migration analyses (Figs 3, 4, S1, S2) confirm the distinctiveness and host specificity of E. vaccinii, which led to our classification of this species into different formae based predominantly on host range (Bradshaw et al., 2024b). In the current study three different formae infecting blueberries have been identified, although Ó 2025 The Author(s). New Phytologist Ó 2025 New Phytologist Foundation. New Phytologist (2025) www.newphytologist.com New Phytologist Priority report Research 5 14698137, 0, Downloaded from https://nph.onlinelibrary.wiley.com/doi/10.1111/nph.20351 by NC State University Libraries, Wiley Online Library on [27/01/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

E. vaccinii f. principalis, was found only on specimens over 100-yr-old, and thus seems irrelevant to current blueberry cultivation.

Our analyses revealed 50 multilocus haplotypes, with a notable bottleneck effect in China (H0), Morocco (H42, H44), Portugal (H42), and Peru (H42, H43), indicated by single or shared Fig. 4 Phylogenetic and population structure inference. (a) The best maximum likelihood tree obtained from a multilocus phylogeny-based placement of 123 strains on the Erysiphe vaccinii reference tree. The numbers along the edges are node support values in the reference tree based on 1000 bootstrap replicates. Only strains with sequences for at least five of the seven reference loci (CAM, GAPDH, GS, ITS, LSU, RPB2, TUB) were retained for population genetic analysis and shown on the tree. Colorized rings around the tree are used to show different strain attributes. The inner ring displays 50 multilocus haplotypes followed by a ring for the different E. vaccinii formae. The two outer rings display optimal STRUCTURE admixture results of k = 2 and k = 7, based on delta K and log e P(D) methods, respectively; the corresponding plots of delta K and log e P(D) values with predefined k = 1-10 are shown below the tree with error bars showing SD per cluster (K ). Two genetically distinct clusters delineate the two outgroups: Erysiphe azaleae and E. vaccinii f. principalis. (b) Statistical parsimony haplotype networks. Two networks were inferred for the 50 multilocus haplotypes using a 95% parsimony connection limit: one large reticulating network and a smaller satellite network including the two strains from Rhododendron. The nodes in the network are replaced with pie diagrams depicting the frequency of haplotype sharing by different hosts. Six haplotypes occupy interior positions in the network and include strains from different hosts; 15 tip haplotypes are singletons and host specific. The E. vaccinii samples from Vaccinium corymbosum belong to least two distinct clonal lineages with evidence of recent clonal expansion. (c) SPLITSTREE haplotype network. A splits graph generated for the in group E. vaccinii strains showing five distinct bipartitions or splits in the data. The E. vaccinii that are not assigned to a host belong to two different bipartitions. The presence of multiple parallel edges between the five bipartitions indicates a history of recombination within formae but very limited recombination between formae on a modern time scale; in the absence of phylogenetic conflict, splits would be separated by a single edge. The bar is proportional to the weight of the split computed using the Neighbor-net algorithm. (d) Multilocus isolation with migration analysis for E. vaccinii formae. The phylogeny with the highest estimated posterior probability is represented as a hierarchical series of boxes with interior ancestor boxes connecting descendant populations where the width of boxes is proportional to the estimated N e . The 95% confidence intervals for each N e value are shown as dashed lines to the right of the left side of the corresponding population box. Gray arrows to the 95% N e intervals extend on either side of the right side of each population box. Splitting times, positioned at even intervals, are depicted as solid horizontal lines, with text values on the left in units of thousand years ago. In the phylogeny E. vaccinii f. vaccinii and E. vaccinii f. vaccinii-corymbosi are recently evolved populations and there is no evidence of significant gene flow among formae. Divergence time estimates assumed a generation time of 1 yr and a mutation rate of 1 9 10 À9 per base per generation. New Phytologist (2025) www.newphytologist.com Ó 2025 The Author(s). New Phytologist Ó 2025 New Phytologist Foundation.

0 3 2 1 4 9 0 0 H F FH01131045 10 0.0 100.0 NYSd14155 3 NYSd14 154 78.0 95 .0 83.0 73.0 FH011

31464 FH01 13146 7 FH01 1314 66 FH0 1131 463 FH0 113 146 2 FH0 094 195 7 90. 0 FH 009 412 01 10 0.0 FH 009 419 58 FH 00 94 19 56 FH 00 94 19 54 FH 00 94 19 55 PE 1A FH 01 12 22 63 97 .0 FH 01 13 14 74 30.0 26 .0 15 .0 P2 19 56 M1 P2 19 50 P2 19 52 P2 195 1 P21 957 P21 959 P21 958 0.0 P21 955 P219 54 P219 53 31.0 M3 M10 FH0094197 4 13.0 M6 0.0 FH00941972 M2 FH01131475 80.0 90.0 HAL 3481 F FH01122 257 34.0 FH011 22268 33.0 FH01 12226 6 NYS d141 53 CUP -A-0 0048 4 CUP F 168 9 20 .0 39. 0 64. 0 NY Sd1 414 6 DA OM 823 44 3.0 FH 011 222 56 DA OM 14 08 55 53 .0 FH 01 12 21 43 FH 01 12 22 33 DA OM 46 65 FH 01 12 22 60 1 FH 01 12 22 60 2 83. 0 43 .0 CU P S 16 59 FH 01 13 14 71 12.0 34.0 73 .0 FH 01 13 10 39 4.0 FH 00 94 19 61 WT U F 073 138 ET NY Sd1 415 0 FH0 112 224 8 NYS d14 147 DAO M 7160 2 37.0 FH01 1310 97 NYSd 14155 1 24.0 36.0 NCSLG 24435 FH00941 975 NYSd14149 FH01122231 8 5 2 2 2 1 1 0 H F 9 6 2 2 2 1 1 0 H F 33.0 FH01122251 37.0 CUP A 9398c 96.0 CCAS ASBF 2 0.0 TNS F 882224 CCAS ASBF 1 55.0 FH01 1310 59 BPI 5557 39 HAL 155 7 80 .0 HA L 348 3 F NR 297 2 76 .0 FH 011 310 58 NC SL G1 71 08 NR 26 27 NR 35 56 CU P A 93 98 a NR 26 35 NR 49 40 HA L 34 82 F CU P A 93 98 b BP I 55 57 40 WT U-F -72 44 1 90.0 89. 0 FH 009 419 76 FH 009 419 77 FH0 113 147 2 72.0 60.0 FH0 113 147 0 FH0 0941 973 FH01 1314 69 FH01 13146 8 FH009 41970 88.0 DS2022 065 DS2022050 30. 0 DS2022033 DS2022040 DS2023004 DS2022026 DS2022049 DS2022 027 DS202 2053 DS20 22068 DS20 2203 1 DS2 0220 32 DS2 022 041 DS 202 300 1 DS 202 203 6 DS 202 206 9 DS 20 22 03 8 DS 20 23 00 7 DS 20 22 03 4 DS 20 22 02 4 DS 20 22 05 1 DS 20 22 02 9 DS 20 22 03 9 DS 20 22 06 7 DS 20 22 07 0 DS 20 22 03 7 DS 202 206 6 DS 202 204 2 DS 202 203 0 DS2 023 002 DS2 0230 08 DS20 2203 5 DS20 22052 DS202 2047 DS2022 028 DS2022025 DS2022048 4 4 0 2 2 0 2 S D Species Erysiphe vaccinii f. epigaeae Erysiphe vaccinii Erysiphe vaccinii f. principalis Erysiphe vaccinii f. elevata Erysiphe vaccinii f. vaccinii-corymbosi Erysiphe azaleae Erysiphe vaccinii f. eucalyptorum Erysiphe vaccinii f. convertibilis Erysiphe vaccinii f. vaccinii (a) 0.001 (c) Species (b) (d) Erysiphe vaccinii f. vaccinii-corymbosi Erysiphe vaccinii f. vaccinii Erysiphe vaccinii Erysiphe vaccinii f. eucalyptorum Erysiphe vaccinii f. epigaeae Erysiphe vaccinii f. elevata Erysiphe vaccinii f. convertibilis Host Catalpa bignonioides Vaccinium myrtilloides Catalpa sp. Vaccinium pallidum Vaccinium corymbosum Vaccinium macrocarpon Chitalpa x tashkentensis No Host Listed Vaccinium sp. Vaccinium parvifolium Gaultheria shallon Vaccinium angustifolium Catalpa speciosa Eucalyptus urophylla x E. camaldulensis Eucalyptus camaldulensis Vaccinium virgatum Vaccinium angustifolium x Vaccinium virgatum Epigaea repens Vaccinium myrtillus Kalmia latifolia Vaccinium virgatum x Vaccinium virgatum 'Ira' Vaccinium darrowii x Vaccinium cylindraceum x Vaccinium corymbosum x Vaccinium elliottii Gaylussacia frondosa Rhododendron occidentale Rhododendron sp. K value K value Delta K Log likelihood probability K Mean LogeP(K) Delta K 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 -3000 -2500 -2000 -1500 -1000 100 200 300 400 79.0 KYR 133.0 KYR 188.0 KYR 257.0 KYR 325.0 KYR Ancestral Ne (thousands): 2.58

E ry si p h e va cc in ii f. va cc in ii-co ry m b o si E ry si p h e va cc in ii f. va cc in ii E ry si p h e va cc in ii f. e le va ta E ry si p h e va cc in ii f. co n ve rt ib ili s E ry si p h e va cc in ii f. e u ca ly p to ru m E ry si p h e va cc in ii f. e p ig a e a e 0 haplotype in each geographic region despite extensive sampling (Table S3; Fig. S2). The data indicate two primary introductions of E. vaccinii -E. vaccinii f. convertibilis to China, Mexico, and California, and E. vaccinii f. vaccinii-corymbosi to Morocco, Peru, and Portugal (Figs 3,4a,S3). This is further supported by significant negative values for Fu and Li's D and Fu and Li's D* for variation in TUB and CAM, indicative of population expansion (Table S4). Minimal genetic diversity was observed outside Eastern North America compared to high diversity within North America, suggesting origin of the pathogen in this region, which is consistent with the native range of its blueberry host.

The haplotype network based on host (Fig. 4b) indicates that the powdery mildews studied evolved from generalist to specialist in support of the specialization hypothesis (Futuyma & Moreno, 1988). Additionally, the SPLITSTREE graph (Fig. 4c) showed significant evidence of recombination within formae (P < 0.005) but limited recombination between formae. All formae, with the exception of E. vaccinii f. convertibilis, are represented by specimens dating back to the 1800s. Tests of isolation with migration did not detect any significant gene flow among formae (Fig. 4d). Gene flow among the different formae, if it exists, is both limited and ancient. The smallest effective population size estimates were for E. vaccinii f. vaccinii-corymbosi (5160-19 350 individuals) and E. vaccinii f. convertibilis (10 320-63 210). Overall divergence time estimates ranged from 79 to 325 thousand years ago (ka) indicating that formae evolved on an evolutionary time scale and before the development of agriculture (c. 12 ka). This underscores their ancient origins, suggesting they are not the result of recent evolutionary changes or modern agricultural practices.

Discussion

This study provides a comprehensive look into the evolution and spread of powdery mildews on blueberries, emphasizing its significant impact on the global blueberry industry. By leveraging a 'living phylogeny' through T-BAS (Carbone et al., 2019) and examining herbarium specimens dating back to the 1800s, we have facilitated the rapid identification of this emerging pathogen both for specialists and for nonexperts. This approach not only underscores the importance of historical collections in understanding pathogen evolution but also highlights the application of phylogenetic tools in understanding contemporary agricultural challenges.

The increased prevalence of powdery mildew across global blueberry farms over the last 5 yr has posed new challenges to the industry, notably in yield reductions and the increased need for fungicide applications. The observed absence of powdery mildew on blueberries in the Pacific Northwest (PNW), despite the region's conducive environment for powdery mildew (Bradshaw, 2020), suggests a limited current spread but anticipates a potential future risk. This could also be the case for Australia and central Mexico. It is important to note that in the PNW, agriculture conditions may not be as conducive as in other regions, as blueberries are not grown in tunnels here. This finding indicates the pathogen's possible expansion beyond its current distribution, urging continuous monitoring and research efforts to preemptively address its spread into new regions.

The insights gained from our study pave the way for future research into pathogen spread and invasiveness, and host-pathogen co-evolution. Conducting host range inoculation and virulence trials on the different blueberry formae, as well as comparative genomic analyses of the different formae, will be informative in unraveling the complexities of disease spread. By advancing our knowledge of these plant pathogens and employing phylogenetic approaches, we can enhance our ability to predict, monitor, and control the spread of powdery mildew.