Resurgence of cucurbit downy mildew in the United States: Insights from comparative genomic analysis of Pseudoperonospora cubensis

Abstract Pseudoperonospora cubensis, the causal agent of cucurbit downy mildew (CDM), is known to exhibit host specialization. The virulence of different isolates of the pathogen can be classified into pathotypes based on their compatibility with a differential set composed of specific cucurbit host types. However, the genetic basis of host specialization within P. cubensis is not yet known. Total genomic DNA extracted from nine isolates of P. cubensis collected from 2008 to 2013 from diverse cucurbit host types (Cucumis sativus, C. melo var. reticulatus, Cucurbita maxima, C. moschata, C. pepo, and Citrullus lanatus) in the United States were subjected to whole‐genome sequencing. Comparative analysis of these nine genomes confirmed the presence of two distinct evolutionary lineages (lineages I and II) of P. cubensis. Many fixed polymorphisms separated lineage I comprising isolates from Cucurbita pepo, C. moschata, and Citrullus lanatus from lineage II comprising isolates from Cucumis spp. and Cucurbita maxima. Phenotypic characterization showed that lineage II isolates were of the A1 mating type and belonged to pathotypes 1 and 3 that were not known to be present in the United States prior to the resurgence of CDM in 2004. The association of lineage II isolates with the new pathotypes and a lack of genetic diversity among these isolates suggest that lineage II of P. cubensis is associated with the resurgence of CDM on cucumber in the United States.

using fungicides. Successful breeding efforts in the 1960s led to availability and wide deployment of resistant cucumber cultivars, and this rendered CDM on cucumber to be of minor concern requiring very limited application of fungicides to control the disease. However, in 2004, CDM resurged on cucumber, and P. cubensis overcame host resistance due to the dm-1 gene that had been effective for more than 40 years (Criswell, Call, & Wehner, 2010). Severe disease epidemics were reported in the eastern United States in 2004 and 2005, leading to complete crop failure in many fields (Holmes, Wehner, & Thornton, 2006). The disease now occurs annually in the United States, and fungicides need to be applied during the season to effectively control the disease on cucumber.
Considerable efforts have been undertaken to establish possible causes of the resurgence of CDM in the United States (Colucci, 2008;Holmes, Ojiambo, Hausbeck, Quesada-Ocampo, & Keinath, 2015;Ojiambo, Gent, Quesada-Ocampo, Hausbeck, & Holmes, 2015;Quesada-Ocampo et al., 2012), in Israel (Cohen et al., 2015), and in parts of Europe (Kitner et al., 2015;Lebeda, Pavelková, Sedláková, & Urban, 2013). It has been hypothesized that the resurgence of CDM in the United States was due to the introduction of a new pathotype or a new cryptic species of the pathogen (Colucci, 2008;Runge, Choi, & Thines, 2011). A new genetic recombinant from sexual reproduction has also been suspected as a possible cause for the resurgence in other parts of the world (Cohen & Rubin, 2012;Runge et al., 2011). During the past two decades, changes in the virulence structure of P. cubensis have been reported around the globe. For example, new pathotypes have been reported in several countries (Cohen et al., 2015) and a change in population dynamics of the pathogen has been reported in the Czech Republic (Kitner et al., 2015;Lebeda et al., 2013). In addition, both mating types of P. cubensis have been found in Israel, China, the United States, and several other countries (Cohen & Rubin, 2012;Cohen et al., 2013Cohen et al., , 2014Cohen et al., , 2015Thomas, Carbone, Cohen, & Ojiambo, 2017). The global resurgence of CDM has been reviewed (e.g., Cohen et al., 2015;Holmes et al., 2015;Ojiambo et al., 2015), and understanding the genetic basis for host specialization within P. cubensis was identified as an important component in providing insights in the resurgence of CDM around the globe (Cohen et al., 2015).
Unlike most other downy mildew pathogens, P. cubensis is able to infect a wide range of cucurbits (Runge et al., 2011), and the virulence of the pathogen has been classified into pathotypes based on its differential compatibility with different cucurbit hosts. A study examining the physiological specialization of P. cubensis using isolates from Japan, Israel, and the United States conducted in 1987 identified the existence of five different pathotypes (Thomas, Inaba, & Cohen, 1987) using thirteen differential host types namely Cucumis sativus, C. melo var. reticulatus, C. melo var. conomon, C. melo var. acidulus, Citrullus lanatus, Cucurbita maxima, C. pepo, C. moschata, Benincasa hispida, Luffa acutangula, L. cylindrica, Momordica charantia, and Lagenaria siceraria.
Isolates capable of infecting Cucumis sativus and C. melo var. reticulatus were classified as pathotype 1, while those capable of infecting C. melo var. conomon in addition to the former hosts were classified as pathotype 2. Isolates capable of infecting all Cucumis spp. were classified as pathotype 3, while all isolates capable of infecting C. lanatus in addition to the Cucumis spp. were designated as pathotype 4. All isolates capable of infecting Cucurbita spp. in addition to Cucumis spp. and C. lanatus were classified as pathotype 5. Since then, up to 10 different pathotypes have been identified (Cohen et al., 2015). While these studies indicate a unique level of interaction between P. cubensis and cucurbit host types, the genetic basis underlying host specialization within this pathosystem is still unresolved.
Following the resurgence of CDM in 2004, a change in the virulence pattern of P. cubensis exhibited by increased aggressiveness on cucumber and a broader host range was reported in the United States (Colucci & Holmes, 2007;Thomas, Carbone, Lebeda, & Ojiambo, 2017). A phylogenetic study conducted in Europe showed that P. cubensis is composed of two cryptic species (Runge et al., 2011).
Clade 1 (cryptic species 1) was associated with isolates in the United States collected prior to the CDM resurgence, and clade 2 (cryptic species 2) was associated with isolates from East Asia and Europe and also included one isolate from the United States collected after the 2004 resurgence. Based on the association of the post-resurgence isolate from the United States with clade 2, it was hypothesized that the resurgence of CDM in the United States could have been due to migration of new strains, possibly from East Asia (Runge et al., 2011).
However, only one isolate from the United States collected after the CDM resurgence was included in the study by Runge et al. (2011), and this lack of data precludes gaining key insights into population-level processes that may have been initiated and subsequently progressed after the resurgence of CDM in 2004. More recent studies have documented the existence of A1 and A2 mating types of P. cubensis in different parts of the world (Cohen et al., 2015) including the United States (Thomas, Carbone, Cohen, et al., 2017). In the Czech Republic, the A1 mating type that has an affinity for Cucumis species was associated with CDM epidemics before 2009, while the A2 mating type that frequently occurs on other cucurbit host types has been associated with disease epidemics since 2009 (Lebeda et al., 2014). Lebeda et al. (2014) provided some indirect insights into host specialization within P. cubensis and the possible association of cucurbit host types with specific clades of the pathogen.
The evolutionary origin of P. cubensis and the relatedness to its sister species, P. humuli that infects hop, have been widely investigated (Choi, Hong, & Shin, 2005;Göker, Voglmayr, Riethmüller, & Oberwinkler, 2007;Kitner et al., 2015;Mitchell, Ocamb, Grünwald, Mancino, & Gent, 2011;Runge et al., 2011). Choi et al. (2005) conducted a phylogenetic analysis of P. cubensis and P. humuli using the nuclear ribosomal internal transcribed spacer (nrITS) region and showed a high level of sequence similarity between the two species. Based on this observation and the high degree of morphological similarity between the two species, P. humuli was reduced to a taxonomic synonym of P. cubensis. However, a subsequent study by Sarris et al. (2009) based on amplified fragment length polymorphism fingerprinting and ITS sequence variation in P. humuli and P. cubensis from C. sativus in Europe reported that the two species formed two distinct clusters. Mitchell et al. (2011) examined variation in three nuclear and two mitochondrial loci and reported that P. humuli from Europe and the United States formed a distinct clade compared to P. cubensis isolated from herbarium samples from South Korea. In that study (Mitchell et al., 2011), one P. humuli isolate from Korea clustered with P. cubensis isolates from Asia, whereas P. humuli isolates from Europe and the United States formed a separate cluster ancestral to P. cubensis. Some cohesion between the two species was observed in the cox2 region of two isolates obtained from Cucurbita pepo in North Carolina and a P. humuli isolate from South Korea (Mitchell et al., 2011), and between P. humuli and P. cubensis isolates from non-cucumber host types (Kitner et al., 2015). Generally, these studies indicate that although P. cubensis and P. humuli are morphologically and physiologically similar, they exhibit quantifiable physiological and genetic differences that support their retention as two separate species. However, a more extensive population genetic analysis is needed to determine whether these two species are specialized on their respective hosts.
The overall goal of our study was to examine patterns of genomewide variation underlying host specialization in P. cubensis and to provide insights into the evolutionary processes that may have driven the resurgence of CDM with increased virulence on cucumber in the United States in 2004. To achieve this goal, this study focused on two specific objectives: (1) to examine genomewide variation in P. cubensis and determine whether it is associated with host specialization, and (2) to determine whether distinct evolutionary lineages or individual multilocus genotypes are associated with pathotypes and mating types of P. cubensis.

| Reference isolates
Nine single-lesion isolates of P. cubensis obtained from cucumber, cantaloupe, pumpkin, butternut squash, acorn squash, and watermelon from different regions of the United States from 2008 to 2013 were used as reference isolates in this study (Table 1). These isolates were maintained in the laboratory by propagating them on their respective host types using a detached leaf assay. Briefly, adaxial sides of detached leaves of 4-week-old plants were placed on moist paper towels in clear acrylic boxes, and the abaxial side of the leaf was inoculated with sporangial suspension at a concentration of 2 × 10 4 sporangia/ml. Inoculated leaves were then incubated in a growth chamber at 21/18°C under 12-hr/12-hr light/dark cycle and were reinoculated once sporangia were produced in abundance. The assay was repeated as necessary until the isolates were ready for use in the subsequent experiments.

| Mating type determination
The mating type of the reference isolates was determined by coinoculating each isolate with a known A1 and/or A2 tester strain as described by Cohen and Rubin (2012). The A1 tester strain was originally isolated from cucumber in North Carolina in 2012, while the A2 tester strain was originally isolated from butternut squash in Israel.
Equal volumes (1:1) of sporangial suspension (2 × 10 4 spores/ml) of the tester strain and each isolate were mixed together, and the mixed suspension was then used to inoculate cantaloupe or cucumber leaves. Cantaloupe and cucumber are a favorable host substrate that supports maximum production of oospores (Cohen & Rubin, 2012).
The first leaves of 3-week-old cucumber and cantaloupe leaves maintained in the glasshouse were used for inoculations. The adaxial side of detached leaves was placed on moist paper towels in clear acrylic boxes, and the abaxial side was spot-inoculated on at least 20 different spots with 10 μl of the sporangial mixture. Care was taken to dispense the inoculum suspension at vein junctions especially for cucumber to ensure maximum infection (Thomas, Carbone, Cohen, et al., 2017). Inoculated leaves were incubated in a growth chamber at 21°C at 50 to 60% relative humidity under 12-hr/12-hr light/dark cycle. At 7 to 10 days postinoculation, leaf disks measuring 11 mm diameter were randomly cut from infected leaves and clarified for 24 hr in ethyl alcohol-acetic acid solution (3:1 v/v). Clarified leaf disks were washed three times in de-ionized water and examined for the presence of oospores using a compound microscope (×100 magnification). If oospores were produced when the unknown isolate was co-inoculated with an A1 or A2 tester strain, then that isolate was considered to be of the A2 or A1 mating type, respectively.
T A B L E 1 Source, description, and characterization of reference isolates of Pseudoperonospora cubensis collected across the United States to study the genetic basis of host specialization

| Pathotype determination and designation
An internationally recognized host differential set (Lebeda & Widrlechner, 2003)  and Momordica charantia, was used to determine the pathotype of the isolates. Poinsett 76, a cucumber cultivar with dm-1 gene, was also included as control to determine isolates that were capable of causing disease on previously resistant cucumber cultivars. The compatibility of the isolates with the host differential set was determined using a detached leaf assay. First and second true leaves from four-week-old host differentials placed with the adaxial side on moist paper towels in clear acrylic boxes were spray-inoculated with sporangial suspension at a concentration of 5 × 10 3 sporangia/ ml. Inoculated leaves were then incubated in a growth chamber at 21/18°C under 12-hr/12-hr light/dark cycle. Disease severity was recorded on a per-leaf basis by assessing the percentage of leaf area infected 5 to 7 days after inoculation (Thomas, Carbone, Cohen, et al., 2017). Two leaf disks measuring 9 mm in diameter were cut from each leaf at 7 DPI and shaken well in microcentrifuge tubes with 500 μl of 50% ethanol to dislodge the sporangia (Cohen, Meron, Mor, & Zuriel, 2003). Sporangial count was then determined using a hemocytometer, and sporangia production was expressed as spores per cm 2 leaf area. A spore concentration of about 5 × 10 3 spores/cm 2 and a disease severity rating of ≥50% were considered as compatible (+) reaction, and an absence of sporulation or very low sporulation (~3 × 10 3 spores/ml) and <20% disease severity were considered as incompatible (−) reaction. Pathotype designation was based on the pattern of compatible and incompatible reactions of each isolate with the differential host set (Cohen et al., 2003(Cohen et al., , 2015Thomas et al., 1987).

| DNA extraction and sequencing
Seven days after inoculation, leaves inoculated with individual reference isolates were rinsed with sterile distilled water using a Preval

| Comparative whole-genome sequence analysis
A publicly available draft genome assembly of P. cubensis, MSU-1 (67.9 Mbp) isolate originally isolated from cucumber, was used for reference-guided comparative genome analyses (Savory et al., 2012).
A complete mitochondrial genome assembly (38,553 bp) of P. cubensis (Lu, Hu, & Wang, 2015) was used as the reference for aligning mitochondrial reads generated from the reference samples.
A pipeline in Mobyle SNAP Workbench that was developed (Aylor, Price, & Carbone, 2006;Monacell & Carbone, 2014;Price & Carbone, 2005) previously for read filtering and variant discovery ( Figure 1) was used in this study. Raw sequence reads were errorcorrected using Quake (Kelley, Schatz, & Salzberg, 2010) and filtered based on a Phred-quality score >20 using the FASTX-Toolkit (Gordon & Hannon, 2010). Filtered read pairs were interleaved into a single FASTQ file using the shuffleSequences.pl script in Velvet version 1.2.10 (Zerbino & Birney, 2008). Corrected and filtered reads were aligned to the MSU-1 reference genome (Savory et al., 2012) and the reference mitochondrial genome (Lu et al., 2015) using the Burrows-Wheeler Aligner (Li & Durbin, 2009 file output from GATK was then used to conduct population genomic analyses using PLINK (Purcell et al., 2007), VCFTOOLS (DePristo et al., 2011), and HAPLOVIEW (Barrett, Fry, Maller, & Daly, 2005), and to conduct phylogenetic analyses using RAxML (Stamatakis, 2006). Population genomic analyses were based on the ten longest contigs ranging from 59 to 106 kbp in size. Estimates of Tajima's D statistic, allele frequency, linkage disequilibrium (LD), heterozygosity, and F ST were based on a sliding window analysis across the contigs. An ancestral recombination graph (ARG) was reconstructed using the branch and bound method implemented in Beagle (Lyngsø, Song, & Hein, 2005) and was based on a window of 100 single nucleotide polymorphisms (SNPs) from a con- we examined all available segregating sites found in the ten longest genome scaffolds for reconstructing the underlying phylogeny and for estimating population parameters.

| Coalescent analysis
The ancestral history of P. cubensis and P. humuli was inferred using Genetree (Griffiths & Tavaré, 1994) implemented in Mobyle SNAP workbench. Coalescent analysis was used to infer rooted gene genealogies showing the relative ages of mutations, haplotypes, and clades. Genealogies for segments of the mitochondrial and nuclear genomes were inferred separately. Specifically, nuclear gene genealogies were inferred for each of the ten longest contigs using windows of ~100 SNPs from the start, middle region, and end of each contig. All SNPs from the mitochondrial genome (~38 kbp) were used for genealogical analysis. Compatibility among pairs of SNPs was used to determine the largest nonrecombining partition of compatible SNPs for each contig using the CladeEx program (Bowden, Price, & Carbone, 2008). The population mean mutation rate was based on Watterson's θ estimator calculated for each analysis window (Griffiths & Tavaré, 1994). The relative probabilities of all rooted genealogies were calculated by performing 1 × 10 6 simulations of the coalescent assuming panmixis and constant population size that was tested a priori using the Tajima's D statistic. Three independent coalescent runs with different starting random number seeds were performed to select the best-rooted gene genealogies for inference of ancestral relationships.

| Mating type determination
Of the nine reference isolates tested, four isolates produced oospores when they were crossed with a known A2 mating type tester isolate and were assigned to the A1 mating type. The remaining five isolates produced oospores only when they were paired with the A1 tester isolate and were designated as the A2 mating type. Two isolates collected from cucumber (A11 and 08A1) and one isolate from cantaloupe (2013B17) and pumpkin (2013C3) belonged to A1 mating type.

| Population genomics
Population genomic analyses were based on 3,356 SNPs spanning the ten largest contigs across the nine sequenced genomes. Binary Alignment/Map (BAM) files for alignments of each of the ten sequenced genomes to the MSU-1 reference genome have been submitted to the NCBI sequence read archive (Accession Nos. SAMN06210292, SAMN06210293, SAMN06210294, SAMN06210295, SAMN06210 296, SAMN06210297, SAMN06210298, SAMN06210299, SAMN0 6210300, SAMN06210301). Population parameter estimates were based on a 1-kbp sliding window analysis. Tajima's D statistic was zero for windows spanning many genomic regions, which is consistent with neutrality. However, there was a signature of population size expansion or positive selection (Tajima's D > 0) (Figure 2) in some regions that were identified as a hypothetical protein or protein kinase.
Plots of coefficients of determination (R 2 ) revealed distinct LD blocks in seven of the 10 contigs, which suggests recent genetic exchange and recombination in P. cubensis (Figure 3). Approximately 37% of the SNPs were heterozygous across both lineages, based on evidence from paired-end reads. In addition, 93% of the SNPs were heterozygous among lineage II isolates.
An ancestral recombination graph (ARG) indicated extensive recombination among inferred single-locus haplotypes of P. cubensis and P. humuli. There was evidence of recombination among lineage I haplotypes of P. cubensis. However, mutation alone separated haplotypes comprising lineage II, as was observed for 27 of 29 regions examined.
Four of the ten contigs provided evidence for a hybrid origin of lineage II. For example, the ARG inferred for a DNA segment on contig 7 indicates that one or more crossovers involving haplotypes H2, H8, and H10 followed by mutation gave rise to the clade that comprises haplotypes H12 and H1 (Figure 4) (Figure 4).

| Phylogenetic analysis
The mitochondrial phylogeny based on 392 SNPs showed that lineage I of P. cubensis shared a recent common ancestor with P. humuli ( Figure 5). Nuclear phylogenetic analyses based on 3,356 SNPs provided evidence of two distinct evolutionary lineages of P. cubensis that are associated with specific hosts, mating types, and pathotypes ( Figure 6). Lineage II isolates were associated with cucumber, cantaloupe, and pumpkin, while lineage I isolates were associated with squash and watermelon. The lineage of P. cubensis was also found to be associated with mating types and pathotypes. All isolates in lineage II belonged to the A1 mating type and pathotypes 1 or 3, while all isolates in lineage I belonged to the A2 mating type and pathotype 4 or 5. Allelic variation was low within lineage II isolates (π = 0.005) but high within lineage I isolates (π = 0.059).

| Coalescent analysis
Mitochondrial gene genealogies were based on 336 SNPs and eleven haplotypes that spanned the largest nonrecombining block in the mitochondrial genome of P. cubensis and P. humuli. Coalescent analysis revealed three distinct evolutionary lineages: P. cubensis lineage I, P. cubensis lineage II, and P. humuli (Figure 7). Although the three lineages are separated by many fixed polymorphisms, lineage I of P. cubensis shared a more recent common ancestor with P. humuli rather than with its conspecific lineage II. A total of thirty gene genealogies were inferred for the ten longest nuclear contigs. All six gene genealogies inferred for contigs 1 and 2 showed that lineages I  F I G U R E 5 Best maximum likelihood phylogenetic tree based on variation in the mitochondrial genome (38.5 kbp, total size; 392 SNPs) for nine isolates of Pseudoperonospora cubensis and one isolate of P. humuli. Isolates of P. cubensis belonging to lineages I and II and an isolate of P. humuli are indicated on the tree, with the tree scale indicating the base substitutions per site and II of P. cubensis shared a more recent common ancestor that was distinct from P. humuli (Figure 8). This topology was supported in the majority (20 of 30) of nuclear gene genealogies. In some regions that spanned contigs 3, 4, 6, 9, and 10, mutation rates were low (π = 0.04) and P. cubensis lineages were indiscernible (Figure 9), while in other regions, mutation rates were higher (π = 0.27) and there was evidence of recombination (or shared haplotypes) between the two lineages of P. cubensis ( Figure 10). Concordance between the nuclear and mitochondrial gene genealogies was observed for two regions on contigs 3 and 8 ( Figure 11).

| DISCUSSION
Genomewide variation in P. cubensis has provided evidence for the existence of two distinct evolutionary lineages specialized on different cucurbit host types. Lineage I was associated with isolates infecting squash and watermelon, while lineage II was associated mainly with isolates infecting Cucumis spp. Pairing assays of the reference isolates with mating type tester strains showed that all lineage I isolates belonged to mating type A2, while all lineage II isolates belonged to mating type A1. In addition, lineage II was comprised of isolates belonging to pathotypes 1 and 3, whereas lineage I comprised isolates belonging to pathotypes 4, 5, and 6. Although Previous studies using multilocus phylogenetic analyses of variation in cox2, ypt1, and nrITS in several similar strains also provided evidence of two clades within P. cubensis (Kitner et al., 2015;Runge et al., 2011). Evolutionary lineages I and II identified in the current study were found to be congruent with clades 1 and 2 (Runge et al., 2011), respectively, based on a phylogenetic analysis of sequence data obtained from NCBI ( Figure 12). In addition, Kitner et al. (2015) in the Czech Republic observed that all isolates sampled before 2009 were associated with Cucumis species, while a broader host range was associated with a subset of isolates collected after 2009. A recent study on the genetic variation between P. cubensis and P. humuli in the United

States used principal component analysis and separated P. cubensis
isolates into two groups, with one group being composed of isolates from Cucumis spp., and the other group comprising isolates from Cucurbita spp. (Summers et al., 2015). In the present study, lineage I was composed of isolates from Cucurbita spp. and Citrullus spp., while lineage II was also composed primarily of isolates from Cucumis spp.
These results are also similar to those reported by Lebeda et al. (2014), where one clade was composed of isolates primarily from Cucumis spp., and the other clade was comprised of isolates from a broader host range. The mechanism controlling host preference in P. cubensis is still unknown, and understanding how these lineages are associated with different host types may provide insights into the potential mechanism(s) that control host preference within the P. cubensis-cucurbit pathosystem.
In the present study, phylogenetic analysis showed low allelic variation among lineage II isolates compared to lineage I isolates. The lower genetic diversity in lineage II, comprising of four isolates collected from different hosts across a wide geographic area, is indicative of a recently introduced population, a bottleneck or a selective sweep within P. cubensis. Lineage I, comprising of five isolates belonging to pathotypes 4, 5, and 6, showed a high level of genetic diversity.
Isolates belonging to pathotypes 4 and 5 had previously been in the United States (Thomas et al., 1987) prior to the resurgence of CDM in 2004. In the United States, CDM were usually observed on squash, pumpkin, and watermelon prior 2004 (Holmes et al., 2006), while disease on cucumber were very minimal prior to 2004 and required very limited fungicide sprays for effective disease control. The latter was largely due to the dm-1 resistance gene that had been widely deployed in cucumber since 1960s (Criswell et al., 2010). In present study, only isolates within lineage II were highly compatible with Poinsett 76, a cucumber cultivar that has the dm-1 gene. Isolates specialized on Cucumis spp. also belonged to pathotypes 1 and 3 that were previously known to exist only in Asia (Thomas et al., 1987). The low allelic variation coupled with the ability to infect cucumber with the dm-1 F I G U R E 6 Best maximum likelihood phylogenetic tree based on the nuclear allelic variation in contig 1 (106 kbp; 680 SNPs) for nine isolates of Pseudoperonospora cubensis rooted with P. humuli. Isolates of P. cubensis belonging to lineage I and lineage II are shown on the tree for which the tree scale indicates the base substitutions per site F I G U R E 7 Rooted mitochondrial gene genealogy showing the distributions of mutations of Pseudoperonospora cubensis and P. humuli in the entire mitochondrial genome (38 kb, 336 SNPs). The numbers indicated along the branches indicate the number of mutation events. The analysis is based on the highest probability tree from three independent runs using 1 × 10 6 coalescent simulations. There are three distinct evolutionary lineages: P. cubensis lineage I (LI), P. cubensis lineage II (LII), and P. humuli. Lineage I of P. cubensis and P. humuli shared a common ancestor in the past. The timescale is in coalescent units, and the direction of divergence is from top (past) to the bottom (present) F I G U R E 8 Rooted nuclear genealogy depicting the distribution of mutations of Pseudoperonospora cubensis and P. humuli using 100 contiguous SNPs from one end of contig 1, with mutations numbered along the branches. The analysis was performed using 1 × 10 6 coalescent simulations, and the best tree was based on three independent runs. Lineage I and lineage II of P. cubensis have diverged from a common ancestor, which coincides with the speciation event separating P. humuli and P. cubensis. The timescale is in coalescent units, and the direction of divergence is from top (past) to the bottom (present) F I G U R E 9 Rooted nuclear gene genealogy showing the distribution of mutations of Pseudoperonospora cubensis and P. humuli using 100 contiguous SNPs from one end of contig 4. Mutations are numbered along the branches. The analysis is based on 1 × 10 6 coalescent simulations, and the best tree was generated based on three independent runs. The genealogy clearly separates the speciation event between P. humuli and P. cubensis (many fixed polymorphisms) in the distant past (top) from population divergence events (few polymorphisms) within each species on a more recent timescale (bottom). Alleles of lineage I and lineage II of P. cubensis are shared between isolates, and the distinction between the lineages is not apparent at this segment of the nuclear genome F I G U R E 1 0 Rooted nuclear genealogy showing the distribution of mutations of Pseudoperonospora cubensis and P. humuli using 100 contiguous SNPs from the first part of contig 9. The analysis is based on 1 × 10 6 coalescent simulations, and the best tree was generated based on three independent runs. Lineage I and lineage II of P. cubensis diverged from a recent common ancestor, which coincides with the speciation event separating P. humuli from P. cubensis. The timescale is in coalescent units, and the direction of divergence is from top (past) to the bottom (present) F I G U R E 1 1 Rooted nuclear gene genealogy showing the distribution of mutations of Pseudoperonospora cubensis and P. humuli based on 100 contiguous SNPs from one end of contig 3, with mutations numbered along the branches. The analysis is based on the highest probability tree from three independent runs for 1 × 10 6 coalescent simulations. There are three distinct evolutionary lineages, and lineage I of P. cubensis and P. humuli shared a common ancestor in the past, which in turn shared a common ancestor with P. cubensis lineage II. Previous studies (Cohen & Rubin, 2012) have suggested an association between cucurbit host types and mating types of P. cubensis.
There have also been reports of both mating types being able to infect the same host species suggesting that mating type and virulence might be genetically unlinked (Cohen et al., 2015). In the present study, all lineage II isolates were of the A1 mating type, while all lineage I isolates were of the A2 mating type. However, the association of lineage and cucurbit host type was not very specific. Indeed, both lineage I and lineage II isolates were able to cause infection on susceptible varieties of cucumber, cantaloupe, and giant pumpkin. This suggests that the association between host types and mating types may be a reflection of the differences in the lineages of P. cubensis rather than differences mediated through host types. In the United States, the distinctiveness between lineages of P. cubensis is more evident, and this may be due to a reproductive isolation due to the absence of lineage II prior to the resurgence of CDM in 2004. Both mating types of P. cubensis were reported recently in the United States, and a distinct association between mating type and cucurbit host types was documented (Thomas, Carbone, Cohen, et al., 2017). The distinctiveness between lineages of P. cubensis may diminish if recombination becomes more widespread, which could result in the formation of hypervirulent strains with a wider host range (Runge et al., 2011). Although both mating types have now been reported in the United States, the extent of sexual recombination among P. cubensis field populations is unknown. Coalescent analysis based on SNP variation in the nuclear genome indicated genetic mixing between lineages I and II of P. cubensis. This was also evident from LD plots generated for the ten longest contigs. Distinct LD blocks indicative of genetic recombination were present in seven of the ten contigs examined. Gene genealogies inferred for separate contigs were incongruent, which further supports a history of genetic exchange.
Recently, Kitner et al. (2015) reported that over 60% of P. cubensis isolates collected in the Czech Republic after 2009 had two heterozygous positions in the nrITS region irrespective of the host of origin, which they postulated as evidence for sexual recombination.
Coalescent-based analysis of SNPs from mitochondrial genomes revealed a close affinity between P. humuli and lineage I of P. cubensis. With the exception of a few nuclear regions, which supported this grouping, the majority of nuclear gene genealogies examined supported the monophyly of the two species. One possibility is that P. humuli and lineage I of P. cubensis initially diverged from a common ancestor (nuclear gene genealogies) and thereafter evolved host specificity on a more recent timescale (mitochondrial gene genealogies), while retaining the ancestral signature of speciation in some genomic regions. Ancestral recombination graph reconstruction indicated a hybrid origin of P. cubensis lineage II isolates with lineage I of P. cubensis and P. humuli as putative parents. However, only a single isolate of P. humuli was examined in this study, and several isolates of P. humuli are needed to draw robust conclusions. In addition, the potential for interspecific hybridization between two species has not yet been investigated. It will be interesting to sample P. humuli and P. cubensis isolates where both species coexist and examine whether there is evidence of ongoing genetic exchange between the species, as has been hypothesized in other studies (Choi et al., 2005;Kitner et al., 2015;. Previous studies have identified two wild cucurbit relatives, Bryonia dioica and Sicyos angulatus, as hosts compatible with both P. humuli and P. cubensis . Susceptible varieties of hop in the United States can also be infected by P. cubensis (Mitchell et al., 2011). There could be more undocumented host types that are compatible with the two Pseudoperonospora species, and a co-infection of such hosts may facilitate genetic exchange between the two species. Given the findings F I G U R E 1 2 Best maximum likelihood phylogenetic tree based on variation in the cox2 region of Pseudoperonospora cubensis. Lineage I and lineage II of P. cubensis are shown on the phylogenetic tree. Sequence data for the cox2 region of isolates labeled as 24230, U201, 22238, U196, 23231, NY429, NY430, U190, D320, and NY442 (from Runge et al., 2011) were obtained from the NCBI database