Population Analyses Reveal Two Host-Adapted Clades of Pseudoperonospora cubensis, the Causal Agent of Cucurbit Downy Mildew, on Commercial and Wild Cucurbits

Downy mildew pathogens are obligate oomycetes in the family Peronosporaceae that cause disease in many plant species and are an understudied but important group of organisms with over 700 described species that affect landscape vegetation and economically important crops (Thines 2014). These pathogens have significant epidemic capacity due to long-distance airborne dispersal of spores (Granke et al. 2014), aggressiveness on their respective hosts (Gascuel et al. 2015; Savory et al. 2011), and ability to quickly develop fungicide resistance (Gent et al. 2008; Blum et al. 2011). It is estimated that 17% of all fungicides in the world are applied for control of downy mildew diseases (Gisi and Sierotzki 2008).

Pseudoperonospora cubensis causes major epidemics wherever cucurbit crops are grown. Economic losses due to P. cubensis have been reported in many parts of the world including Europe, the Middle East, and Asia (Cohen et al. 2015). The United States has suffered particularly dramatic losses since 2004 when the pathogen re-emerged after a population shift rendered resistant cucumber varieties and commonly used fungicides ineffective (Holmes et al. 2015). Approximately $20 million dollars in losses occurred in 2004 and since then, disease management requires frequent fungicide applications. The pathogen remains at high risk for developing resistance to fungicides and overcoming host resistance (Savory et al. 2011). P. cubensis can produce up to 7 × 10⁴ sporangia per cubic centimeter when it sporulates on the abaxial surface of the leaf (Lebeda and Cohen 2011). Sporangia can travel approximately 1,000 km via air currents allowing P. cubensis to migrate from field to field throughout a season as cucurbit hosts are planted and temperatures become conducive for infection (Ojiambo et al. 2015).

Downy mildew pathogens such as P. humuli (hop), Bremia lactucae (lettuce), and Peronospora belbahrii (basil) typically cause disease on only a few plant species usually within the same genus (Crandall et al. 2018). However, other species, such as P. cubensis, have a very broad host range. P. cubensis is capable of infecting over 60 different species of plants in the family Cucurbitaceae from very diverse genera (Runge and Thines 2009; Wallace and Quesada-Ocampo 2015; Wallace et al. 2014). Differences in virulence of P. cubensis on diverse hosts have been described, with isolates of the pathogen being classified into different races and pathotypes based on the ability to infect a set of host differentials (Thomas et al. 1987). However, the obligate nature of downy mildew pathogens makes species, race, and pathotype classification systems based on virulence on differential hosts laborious, highlighting the need of genetic studies to better understand the evolution and host specificity of these organisms (Mitchell et al. 2011; Rahman et al. 2017, 2019; Runge and Thines 2010; Runge et al. 2012; Summers et al. 2015a; Withers et al. 2016).

Population genetics studies have shown genetic differentiation of P. cubensis isolates is associated with host and geographical location. Quesada-Ocampo et al. (2012) found differentiation between isolates from Europe and North America, as well as isolates from cucumber and other cucurbit hosts using multilocus analysis. Genetic differences among isolates from cucumber and cantaloupe, and isolates from squash were also reported by Summers et al. (2015a, b) using genotyping by sequencing (GBS). Quesada-Ocampo et al. (2012) also reported increased genetic diversity in certain regions of the United States, such as Georgia and North Carolina. Nonetheless, because these studies used opportunistic instead of structured sampling (Quesada-Ocampo et al. 2012) or had limited sampling (Summers et al. 2015b), they were not able to determine if the observed geographic structure was due to differences in host availability in the regions sampled. In addition, only commercial hosts have been analyzed in these studies and the contribution of wild cucurbit hosts as a source of inoculum and/or as a diversifying factor of P. cubensis is largely unknown. Lebeda (1992) determined that most of the 56 accessions of wild cucurbits evaluated in laboratory inoculations were susceptible to the pathogen. Runge and Thines (2009) also found that the perennial Bryonia dioica is a host for P. cubensis in Europe. Reports in the United States revealed that wild cucurbits such as Momordica balsamina, M. charantia, and Cucurbita foetidissima are also hosts of P. cubensis (National Plant Data Team 2020; Wallace and Quesada-Ocampo 2015; Wallace et al. 2014).

Wild cucurbits could be involved in the overwintering survival of the pathogen, serve as green bridges for dispersal, provide sites of sexual reproduction, act as reservoirs, and/or contribute to the genetic diversity of the pathogen (Cohen et al. 2015; Holmes et al. 2015; Ojiambo et al. 2015), but no extensive studies have been conducted thus far to examine these aspects in detail. Therefore, to establish the role of wild and commercial cucurbit hosts in the structuring of P. cubensis populations, we analyzed isolates obtained from standardized field plots planted with commercial and wild cucurbits with previously developed microsatellite markers (Wallace and Quesada-Ocampo 2017). Specifically, we aimed to determine if P. cubensis populations are genetically structured by host genus (Cucumis, Cucurbita, Citrullus, Momordica, and Lagenaria), host species (Cucumis sativus, Cucumis melo, Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Citrullus lanatus, Cucurbita foetidissima, Momordica charantia, and Momordica balsamina), geographical location (East, Central, and West North Carolina), and/or time of production (summer and fall).

MATERIALS AND METHODS

Field plots, isolate collection, and DNA extraction.

P. cubensis isolates were collected from three replicated fields in three regions in North Carolina: the Cunningham Research Station in Kinston (East), the Piedmont Research Station in Salisbury (Central), and the Mountain Research Station in Waynesville (West). Each of these locations are in geographically distinct regions of the state with differences in weather patterns, soil type, and surrounding cucurbit production (Supplementary Fig. S1, Supplementary Table S1). North Carolina is a primary producer of all major cucurbit crops including cucumber, cantaloupe, squash, pumpkin, gourds, and watermelon and it is in the geographical range of several wild cucurbits that are naturally infected by P. cubensis (Wallace et al. 2014, 2016) and other downy mildew pathogens (Wallace et al. 2016) (Supplementary Fig. S2).

Each field contained one plot per host with 10 plants of each commercial cucurbit crop including Cucumis sativus ‘Straight 8’ (cucumber), Cucumis melo ‘Hale’s Best Jumbo’ (cantaloupe), Cucurbita pepo ‘Table Ace’ (acorn squash), Cucurbita maxima ‘Big Max’ (pumpkin), Cucurbita moschata ‘Waltham Butternut’ (butternut squash), and Citrullus lanatus ‘Micky Lee’ (watermelon). The field also contained plots with five plants of each wild cucurbit including Momordica charantia (bitter melon), Momordica balsamina (balsam apple), Cucurbita foetidissima (buffalo gourd), and Lagenaria siceraria (bottle gourd). The field was planted in raised beds with black plastic mulch and disease was monitored weekly. Fifteen symptomatic leaves were collected per plot in plastic bags at two time points over 2 years: summer (July to August) and fall (September to November) of 2013 and 2014. Disease was confirmed in the laboratory by observation of sporulating lesions under a dissecting microscope. Individual sporulating lesions were excised and stored in microcentrifuge tubes at −80°C and these single-lesion samples were considered isolates for downstream analyses as previously described in Quesada-Ocampo et al. (2012).

Infected tissue was frozen in liquid nitrogen and disrupted with 425 to 600 µM acid-washed glass beads (Sigma Life Sciences) and 2.3 mm Zirconia/Silica beads (BioSpec Products, Inc. Bartlesville, OK) in an Omni Bead Ruptor 24 (Omni International, Inc., Kennesaw, GA). Sodium dodecyl sulfate-polyacrylamide gel electrophoresis extraction buffer was used to extract DNA and was purified using phenol-chloroform extractions and ethanol washes adapted from previous work (Ahmed et al. 2009). DNA quantity (ng/µl) and quality (260/280) were measured with Nanodrop ND 1000 spectrophotometer and NanoDrop 2.4.7c software (NanoDrop Technologies Inc., Wilmington, DE).

Product amplification, fragment analysis, and genotyping.

DNA from each P. cubensis isolate was amplified with 10 microsatellite markers (Supplementary Table S2) identified and characterized in Wallace and Quesada-Ocampo (2017). PCR products labeled with different fluorescent dyes were pool-plexed and diluted 20-fold in 96-well plates. Samples were prepared for the 3730xl DNA Analyzer (Applied Biosystems, Foster City, CA) by adding HiDi Formamide and Liz600 size standard to each well. Samples were submitted to the North Carolina State University Genomic Science Laboratory for fragment analysis. Raw data generated by the 3730xl DNA Analyzer were individually evaluated by calling peaks and binning them with the Geneious Microsatellite Plug-In (Kearse et al. 2012) as previously described in Wallace and Quesada-Ocampo (2017). After genotyping, all isolates with >20% missing data were removed from analyses, resulting in a total dataset of 385 isolates with an average of 4.6% missing data. All genotyping data used for analyses are available at https://doi.org/10.6084/m9.figshare.11389824.v1.

Population structure analyses.

Bayesian clustering and the admixture model with correlated allele frequencies were used in the program STRUCTURE 2.3.4 (Pritchard et al. 2000) to determine the optimal number of genetic clusters within the P. cubensis isolates, and to assign all individuals to the genetic clusters (n = 385). A burn-in value of 300,000 was used with 500,000 Monte Carlo Markov chain (MCMC) repetitions. Five runs were performed for each K (K= 1 to 40). Structure Harvester (Earl and vonHoldt 2012) was used to determine the optimal number of genetic clusters (K). Population structure figures sorted by the proportional membership (Q) of an isolate in a cluster were generated to visualize the distribution of clusters in predefined categories of host (species, genus, commercial, and wild), location (west, central, and east), and time (summer or fall, year).

Pairwise comparisons of F_ST values from populations were generated to determine population differentiation across K, host, location, and time categories using GenAlEx (Peakall and Smouse 2006). P values and significance for pairwise F_ST values were calculated with 999 permutations. Only categories with at least eight individuals were included for these analyses following the methods of Quesada-Ocampo et al. (2012) and Naegele et al. (2016). For analyses by K, an isolate was considered to belong to a K grouping if it had genetic membership of 60% or higher in a particular K as described in previous studies (Naegele et al. 2016). Genetic differentiation was interpreted to be low (<0.10), moderate (0.10 to 0.20), or high (>0.20) consistent with the guidelines in Hartl and Clark (2006).

Genetic diversity and linkage disequilibrium analyses.

Locus-based statistics such as the number of multilocus genotypes, Shannon-Wiener index (Shannon 2001), Stoddart and Taylor’s index (Stoddart and Taylor 1988), Simpson’s index (Simpson 1949), and Nei’s unbiased gene diversity (Nei 1978) through the function “locus_table” were generated using the R package Poppr (Kamvar et al. 2014). Only categories with at least eight individuals after clone correction were included for analysis. Metrics of genetic diversity across K, host, location, and time categories were calculated with the “poppr” function on the dataset. The index of association (I_A) and standardized index of association (rbarD) (Brown et al. 1980) with 999 resamplings were also calculated from a clone-corrected dataset to determine the extent of linkage disequilibrium by quantifying recombination among microsatellite loci and detecting association between alleles. The I_A and rbarD are expected to be zero if isolates are freely recombining, which is consistent with random mating. The indexes are greater than zero if isolates are not freely recombining, consistent with a population not randomly mating due, for example, to selfing or asexual reproduction. The rbarD accounts for the number of loci tested and it is considered a more robust statistic (Agapow and Burt 2001).

RESULTS

Bayesian clustering identified two genetic clusters (K1 and K2) in the 385 P. cubensis isolates (Fig. 1, Supplementary Table S3). Most isolates had majority membership (≥60%) to only one genetic cluster, with only six isolates having no majority membership to either and designated as unassigned. These six isolates had almost equal membership to the K1 and K2 genetic clusters and were collected from Cucumis melo (three), Citrullus lanatus (one), Momordica charantia (one), and Cucurbita foetidissima (one) in different locations, years, and sampling times.

Download as PowerPoint

Host was the main factor driving population structure in P. cubensis isolates. Occurrence of K1 and K2 majority membership isolates (≥60%) as determined by Bayesian clustering was examined in the different host categories (Fig. 2). This analysis revealed that K2 isolates were more frequently found in host species including Cucumis sativus, Cucumis melo, and the wild host Lagenaria siceraria, while K1 isolates were more frequently found in Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Citrullus lanatus, and wild hosts Momordica charantia, and Momordica balsamina. Interestingly, the wild species Cucurbita foetidissima harbored K1 and K2 isolates with similar frequency, while Cucurbita pepo, Cucurbita moschata, Momordica charantia, and Momordica balsamina isolates only had K1 membership and Lagenaria siceraria isolates only had K2 membership (Fig. 2A).

Download as PowerPoint

When isolates were examined by cucurbit host genus, K2 isolates were more frequently found in Cucumis and Lagenaria, while K1 isolates were more frequent in Cucurbita, Citrullus, and Momordica. Unassigned isolates were found in all genera except for Lagenaria with a majority being found in Momordica (Fig. 2B). Examining isolates by the category host type (commercial, wild) revealed that K1 and K2 isolates as well as unassigned isolates are found in both commercial and wild cucurbits (Fig. 2C). Host trends remained true when isolates were analyzed within location and year, and combinations of those categories.

Analysis of location and time categories did not reveal population structure in most cases, with K1 and K2 isolates occurring at comparable numbers among location, time, and year categories (Fig. 3A, B, and C). However, all regions had higher occurrence of K1 isolates in the fall sampling, and the east region had higher occurrence of K2 isolates in the summer sampling (Fig. 3D).

Download as PowerPoint

Levels and significance of population genetic differentiation between isolates grouped by host, location, time categories, and combinations of those were examined by calculating pairwise F_ST values. Analysis of host by location revealed that host drives population differentiation (Supplementary Fig. S3). The highest genetic differentiation occurred between isolates from Cucumis sativus and Cucurbita pepo, Cucurbita moschata, Momordica charantia, and Momordica balsamina regardless of location. High differentiation was also observed between isolates from Lagenaria siceraria and Cucurbita pepo, Cucurbita moschata, Citrullus lanatus, and Momordica charantia in all locations examined. High to medium genetic differentiation was detected between isolates from Cucumis melo and Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Momordica charantia, and Momordica balsamina, and between isolates from Cucumis sativus and Cucurbita maxima, Cucumis sativus and Citrullus lanatus, and Cucurbita maxima and Lagenaria siceraria depending on location. There was medium to low genetic differentiation between isolates from Cucumis melo and Citrullus lanatus, and Lagenaria siceraria and Momordica balsamina depending on location. There was low genetic differentiation among Cucumis sativus, Cucumis melo, and Lagenaria siceraria isolates in all locations, and among Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Citrullus lanatus, Momordica charantia, and Momordica balsamina isolates across all locations. Isolates from Cucurbita foetidissima had low differentiation to isolates across all host groups and locations. There were no obvious trends of genetic differentiation between isolates from commercial versus wild hosts, or isolates of a particular genus, and levels and significance of genetic differentiation were specific at the host species level. There were some cases of genetic differentiation by host and location. Isolates from Cucumis sativus were more differentiated from isolates from Cucurbita maxima in the central and western part of the state compared with the east. Conversely, isolates from Cucumis melo were more differentiated form isolates from Cucurbita pepo in the eastern part of the state compared with the central and western regions (Supplementary Fig. S3). Because results of pairwise F_ST analysis could be due to underlying genetic cluster K occurrence by host, genetic differentiation between isolates with majority membership (≥60%) in genetic clusters K1 or K2 was calculated and found to be high and significant (K1 versus K2 pairwise F_ST = 0.317, P ≤ 0.001).

Further analysis of P. cubensis isolates using minimum spanning networks (MSN) based on Bruvo’s genetic distance also showed two major groupings of multilocus genotypes (MLGs) that followed the trends observed in the Bayesian clustering and pairwise F_ST analyses (Fig. 4; Supplementary Fig. S4). Most MLGs from Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Citrullus lanatus, and the wild hosts Momordica charantia, and Momordica balsamina grouped into a separate group, labeled as clade 1, while most MLGs from Cucumis sativus, Cucumis melo, and the wild host Lagenaria siceraria grouped into a discrete group, labeled as clade 2. MLGs from Cucurbita foetidissima were found in both groups.

Download as PowerPoint

Genetic diversity estimates for isolates grouped by host, location, time categories, and combinations of those were also examined. Since the major factor resulting in population stratification of P. cubensis was host, genetic diversity estimates are shown based on cucurbit species of origin (Table 1) as well as based on K majority membership (≥60%) as defined by Bayesian clustering analysis (Table 2). Our analyses showed that isolates from Cucumis sativus and Cucumis melo had the highest values for MLGs and the Shannon-Wiener (H) and Stoddart and Taylor’s (G) indices of MLG diversity (Table 1). However, the highest values of expected MLGs (eMLGs) occurred in isolates from Cucumis melo, Cucumis sativus, and Cucurbita foetidissima. Isolates from Cucurbita maxima, Cucurbita foetidissima, Cucumis melo, and Momordica charantia had the highest values of gene diversity (Hexp). Genetic diversity estimates for isolates grouped by K showed higher values of all diversity estimates for K2 isolates, except for the MLG and Hexp values, which were higher for K1 isolates (Table 2).

TABLE 1. Genetic diversity and index of association (I_A) estimates for Pseudoperonospora cubensis isolates grouped by host of origin

	All individuals									Clone-corrected
Host species	Na	MLGb	eMLGc	Hd	Ge	Lambdaf	Hexpg	IAh	rbarDi	IAh	rbarDi
Cucumis sativus	80	72	9.83	4.22	61.53	0.98	0.32	0.63*	0.073*	0.53*	0.060*
Cucumis melo	67	64	9.92	4.13	59.85	0.98	0.41	0.93*	0.106*	0.86*	0.098*
Cucurbita pepo	48	40	9.58	3.61	32.91	0.97	0.36	0.05	0.006	−0.02	−0.002
Cucurbita maxima	51	45	9.60	3.71	33.78	0.97	0.48	1.69*	0.203*	1.53*	0.183*
Cucurbita moschata	59	46	9.50	3.71	34.46	0.97	0.36	0.09	0.013	0.01	0.002
Citrullus lanatus	27	24	9.52	3.12	20.83	0.95	0.39	0.08	0.009	−0.02	−0.002
Momordica charantia	13	13	10	2.56	13	0.92	0.41	0.31	0.036	0.31	0.036
Cucurbita foetidissima	24	23	9.83	3.12	22.15	0.95	0.44	1.38*	0.160*	1.31*	0.152*
Lagenaria siceraria	9	9	9	2.20	9	0.88	0.19	0.46	0.067	0.46	0.067
Total	385	309	9.84	5.54	159.21	0.99	0.46	1.43	0.166	1.26	0.146

^a N: number of individuals.

^b MLG: number of multilocus genotypes.

^c eMLG: number of expected MLG at the smallest sample size based on rarefaction.

^d H: Shannon-Wiener index of MLG diversity (Shannon 2001).

^e G: Stoddart and Taylor’s index of MLG diversity (Stoddart and Taylor 1988).

^f Lambda: Simpson’s index (Simpson 1949), corrected for population size.

^g Hexp: Nei’s unbiased gene diversity (Nei 1978).

^h I_A: index of association for all individuals and for clone-corrected data (Brown et al. 1980). An asterisk indicates statistical significance (significant with P ≤ 0.001) compared with 999 resamplings (Agapow and Burt 2001).

ⁱ rbarD: standardized index of association for all individuals and for clone-corrected data (Brown et al. 1980). An asterisk indicates statistical significance (significant with P ≤ 0.001) compared with 999 resamplings (Agapow and Burt 2001).

View as image HTML

TABLE 2. Genetic diversity and index of association (I_A) estimates for Pseudoperonospora cubensis isolates grouped by majority membership (≥60%) to K1 or K2 genetic clusters

	All individuals									Clone-corrected
Genetic cluster	Na	MLGb	eMLGc	Hd	Ge	Lambdaf	Hexpg	IAh	rbarDi	IAh	rbarDi
K1	216	162	126.40	4.82	69.02	0.98	0.39	0.23*	0.028*	0.08	0.009
K2	163	141	141	4.85	106.70	0.99	0.32	0.69*	0.079*	0.58*	0.066*
Total	385	309	140.86	5.52	155.28	0.99	0.46	1.46	0.170	1.17	0.136

^a N: number of individuals.

^b MLG: number of multilocus genotypes.

^c eMLG: number of expected MLG at the smallest sample size based on rarefaction.

^d H: Shannon-Wiener index of MLG diversity (Shannon 2001).

^e G: Stoddart and Taylor’s index of MLG diversity (Stoddart and Taylor 1988).

^f Lambda: Simpson’s index (Simpson 1949), corrected for population size.

^g Hexp: Nei’s unbiased gene diversity (Nei 1978).

^h I_A: index of association for all individuals and for clone-corrected data (Brown et al. 1980). An asterisk indicates statistical significance (significant with P ≤ 0.001) compared with 999 resamplings (Agapow and Burt 2001).

ⁱ rbarD: standardized index of association for all individuals and for clone-corrected data (Brown et al. 1980). An asterisk indicates statistical significance (significant with P ≤ 0.001) compared with 999 resamplings (Agapow and Burt 2001).

View as image HTML

Since P. cubensis isolates showed some differences in occurrence of K1 and K2 genetic clusters in certain location and time combinations (Fig. 3D), we also examined genetic diversity of isolates within those groupings (Table 3). The values for Hexp were higher in the fall than in the summer for all locations. Values for MLG and Shannon-Wiener index (H) were higher in the fall for the central and eastern locations, but higher in the summer for the western location. The Stoddart and Taylor’s index (G) had higher values in the western and central locations in the summer, but higher in the fall for the eastern location. The eMLG values were higher in the summer for the central and eastern locations, but higher in the fall for the western location. The value for Simpson’s index was similar in all locations and times (Table 3).

TABLE 3. Genetic diversity estimates for Pseudoperonospora cubensis isolates grouped by location and time

	All individuals
Location/time	Na	MLGb	eMLGc	Hd	Ge	Lambdaf	Hexpg
West
Summer	57	52	36.81	3.89	44.26	0.97	0.42
Fall	45	43	39.17	3.75	42.08	0.97	0.45
Central
Summer	80	71	37.43	4.19	58.32	0.98	0.45
Fall	90	78	36.59	4.25	55.47	0.98	0.47
East
Summer	40	40	40.00	3.68	40.00	0.97	0.42
Fall	73	66	37.12	4.12	53.68	0.98	0.45
Total	385	309	37.84	5.54	159.21	0.99	0.46

^a N: number of individuals.

^b MLG: number of multilocus genotypes.

^c eMLG: number of expected MLG at the smallest sample size based on rarefaction.

^d H: Shannon-Wiener index of MLG diversity (Shannon 2001).

^e G: Stoddart and Taylor’s index of MLG diversity (Stoddart and Taylor 1988).

^f Lambda: Simpson’s index (Simpson 1949), corrected for population size.

^g Hexp: Nei’s unbiased gene diversity (Nei 1978).

View as image HTML

Since population differentiation by host was high, we estimated the I_A and rbarD, and calculated statistical significance to gain insights of mating and recombination in P. cubensis isolates. Isolates from Cucurbita maxima, Cucurbita foetidissima, Cucumis sativus, and Cucumis melo had the highest values for I_A and rbarD before and after clone correction and all values were significant, which indicates nonrandom mating with no evidence of recombination (Table 1). In contrast, the I_A and rbarD values before and after clone correction for isolates from Cucurbita pepo, Cucurbita moschata, Citrullus lanatus, Momordica charantia, and Lagenaria siceraria were lower and not significant, indicating random mating and evidence of recombination (Table 1).

Population analyses also revealed high differentiation by genetic cluster K. When I_A and rbarD values were calculated for P. cubensis isolates with majority membership (≥60%) to K1 or K2 before clone correction, results were significant for K2 and K1 isolates, indicating nonrandom mating with no evidence of recombination. Because I_A and rbarD values were high in K2 isolates and low in K1 isolates, the test was also performed with clone-corrected data. Values of I_A and rbarD were high and significant for K2 isolates, suggesting nonrandom mating with no evidence of recombination, but results were lower and not significant for K1 isolates, indicating random mating and evidence of recombination (Table 2).

DISCUSSION

Population analyses using Bayesian clustering, pairwise F_ST values, and minimum spanning networks revealed two distinct clades of P. cubensis isolates preferentially infecting particular cucurbit species regardless of location and time of sampling. In North Carolina, clade 1 (K1) isolates preferentially infect Cucurbita pepo, Cucurbita maxima, Cucurbita moschata, Citrullus lanatus, and wild hosts Momordica charantia, and Momordica balsamina, while clade 2 (K2) isolates preferentially infect Cucumis sativus, Cucumis melo, and the wild host Lagenaria siceraria.

Host resistance is the most desirable method for disease control, and the existence of clade 1 and clade 2 P. cubensis isolates has implications for disease resistance breeding. Because P. cubensis is an obligate pathogen, breeders and seed companies often rely on natural inoculum to evaluate breeding lines and populations for disease resistance (Call et al. 2012a, b; Wang et al. 2019). Nonetheless, the occurrence of clade 1 and clade 2 isolates in a particular area may vary due to local cucurbit production contributing to each inoculum type (Ojiambo et al. 2015). In our study, we saw that isolates from Cucumis sativus versus Cucurbita maxima and Cucumis melo versus Cucurbita pepo were more differentiated from one another in particular regions of the state, which likely reflects local cucurbit production. Thus, connecting clade host specificity by region with breeding efforts to ensure clade does not become a confounding factor when evaluating breeding materials, would be critical to refine phenotyping efforts. Seasonality of clades also needs to be considered. In our study we showed that clade 1 isolates were more frequent in the central region in the fall, while clade 2 isolates were more frequent in the eastern region in the summer. Monitoring of P. cubensis at the state, host, and seasonal level as done in this study and in previous work (Naegele et al. 2016) would provide insights on the population dynamics of clade 1 and clade 2 isolates to have pathogen-informed disease management strategies (Rahman et al. 2017).

Our findings also have important implications for disease management of cucurbit downy mildew using chemical control. Currently, management heavily relies on frequent fungicide applications with effective products (Holmes et al. 2015). Efficacy trials to evaluate fungicides are typically performed in cucumber and recommendations are extended to other cucurbits (Goldenhar and Hausbeck 2019; Keinath et al. 2019). However, it has been reported that fungicide efficacy is variable by state and by year (Keinath et al. 2019), which could be explained by differences in clade occurrence. In addition, currently growers are recommended to initiate sprays to protect all cucurbit crops once a cucurbit downy mildew report has been made in neighboring states to the Cucurbit Downy Mildew IPM pipe (CDM IPM PIPE) regardless of the host (Ojiambo et al. 2011). If clade 1 and clade 2 P. cubensis isolates have differences not only in host preference but also in fungicide sensitivity, crop-specific fungicide recommendations and guidelines for when to initiate sprays would need to be generated for the most effective disease and fungicide resistance management.

This study was also able to determine that wild cucurbits are reservoirs of P. cubensis. Interestingly, clade 1 and clade 2 also showed host preference for species of wild cucurbits, with clade 1 occurring more frequently in Momordica charantia and Momordica balsamina, while clade 2 was more frequently found in Lagenaria siceraria. Both clades were able to infect Cucurbita foetidissima, a perennial weed (Bemis et al. 1978), which may have implications for overwintering. The hop downy mildew pathogen, P. humuli, is systemic and able to overwinter in its perennial host, Humulus lupulus (Royle and Kremheller 1981). Because commercial Cucumis species usually become infected earlier in the season than commercial Cucurbita spp., C. foetidissima may pose a threat for the disease to occur earlier than anticipated. This could be supported by the fact that a previous spatiotemporal study examining aspects of cucurbit downy mildew outbreaks found that there were up to 5% discrepancies of reported outbreaks in comparison with the known northward advance of the epidemic wave front in the eastern United States (Ojiambo and Holmes 2011). Those unexplained reports could have been due to the green bridge provided by year-round greenhouse production (Naegele et al. 2016), but also because wild cucurbits such as C. foetidissima are influencing host availability and therefore outbreaks in various areas. Currently, the CDM IPM PIPE does not account for overwintering scenarios (Ojiambo et al. 2011), which has highlighted the need for additional biosurveillance tools for P. cubensis that rely on inoculum detection and complement disease reports (Granke and Hausbeck 2011; Granke et al. 2014). However, each has advantages and disadvantages that need to be understood to develop robust alert systems (Rahman et al. 2017; Ojiambo et al. 2015; Crandall et al. 2018).

In our study, the host adaptation displayed by clade 1 and clade 2 of P. cubensis is at the species level and not genus level as found in other work for Cucumis versus Cucurbita isolates (Summers et al. 2015a, b), since isolates from both clades seem to be able to infect the wild host Cucurbita foetidissima. The first report of two P. cubensis clades was made by Runge et al. (2011) where they described a new phylogenetic lineage associated with recent cucurbit downy mildew epidemics in the United States and Europe, but no host-clade association was found. Quesada-Ocampo et al. (2012) found differences in Bayesian genetic clusters occurring in cucumber versus other cucurbits, but did not include samples from commercial hosts such as watermelon or wild hosts, and were not able to separate host from geographic effects. Clades identified in our study are in line with groupings described by Quesada-Ocampo et al. (2012), Runge et al. (2011), and Summers et al. (2015a, b), but do not completely coincide with lineages described by Thomas et al. (2017). Thomas et al. (2017) analyzed polymorphisms in nine isolates based on DNA sequencing data and reported two P. cubensis lineages, with lineage 1 containing isolates from Cucurbita pepo, Cucurbita moschata, and Citrullus lanatus and lineage 2 isolates from Cucumis spp. and Cucurbita maxima. Our analysis showed that clade 2 isolates can infect C. maxima, but at much lower levels than clade 1; thus, placement of C. maxima isolates in lineage 2 by Thomas et al. (2017) may be due to a small sample size.

The genetic differentiation between clade 1 and clade 2 isolates is significant based on Bayesian clustering, pairwise F_ST values, and minimum spanning networks analyses, which suggests a reproductive barrier between clades. Pseudoperonospora cubensis has been reported to be heterothallic with mating types showing host preference (Cohen et al. 2013). Cohen et al. (2013) reported that out of 303 isolates from Israel, most cucumber (94.3%) and cantaloupe (58.5%) isolates belonged to the A1 mating type, while most of the Cucurbita maxima (96.7%), Cucurbita moschata (87.3%), and Cucurbita pepo (97.6%) isolates belonged to the A2 mating type. Thomas et al. (2017) determined with nine P. cubensis isolates from the United States, that two isolates from cucumber, one from cantaloupe, and one from pumpkin from lineage 2 also belonged to mating type A1, while two isolates from butternut squash, two from acorn squash, and one from watermelon from lineage 1 also belonged to mating type A2. Interestingly, in our study, index of association analyses revealed random mating with evidence of recombination in clade 1 consistent with a sexually reproducing population, and/or heterothallism, while clade 2 showed nonrandom mating with no evidence of recombination consistent with selfing or asexual reproduction (Agapow and Burt 2001; Burt et al. 1996). Thomas et al. (2017) made a similar observation where they found evidence of recombination in lineage 1 but not in lineage 2.

The limited genetic evidence for recombination between clade 1 and clade 2 isolates could be explained by the expansion of a single clonal lineage in P. cubensis (clade 2) or a speciation event between clades with established pre- and/or postzygotic barriers. Since both clades are able to infect several shared cucurbit hosts, although at a lower level, host may not be the only reproductive barrier resulting in such high genetic differentiation. Intra and interclade crosses along with parent and progeny genotyping and phenotyping would be needed to clarify this. Previous studies have performed crosses between P. cubensis isolates originating from different hosts but oospore formation, viability, and infectivity in those studies seem to be low (Cohen and Rubin 2012; Cohen et al. 2011). Phytophthora infestans is a heterothallic oomycete with rare but confirmed sexual reproduction and recombination between the A1 and A2 mating types, but self-fertile clonal lineages have also been described (Smart et al. 2000; Zhu et al. 2015). Further studies are warranted to determine if currently described A1 and A2 P. cubensis mating types are still mating in nature, undergoing recombination, and producing viable, infective progeny, which has not been corroborated to date (Cohen et al. 2013; Thomas et al. 2017).

If clade 1 and clade 2 are distinct species, they could each have a different mating system (homothallism versus heterothallism) and possibly undergo hybridization events (Corredor‐Moreno and Saunders 2020; Rouxel et al. 2013; Stukenbrock 2013). Hybridization has been reported between different Phytophthora species (Boccas 1981; Burgess 2015). Pseudoperonospora cubensis is believed to be heterothallic due to production of viable oospores when the two mating types come into contact (Cohen et al. 2013), however, oospores can be formed in oomycetes due to exposure to the hormone from an opposite mating type even from a different species (Shattock et al. 1986; Tomura et al. 2017). If clade 2 was a homothallic species and clade 1 heterothallic, it could explain evidence of recombination in one clade but not the other, but this hypothesis needs to be tested in future studies. However, the obligate nature of downy mildews and host specificity of the clades is an obstacle for experiments to determine true mating events by assessing oospore recombination versus just oospore formation in response to the opposite mating hormone versus a hybrid oospore from two species. Despite clade 1 and clade 2 isolates having some shared hosts, only six isolates had equal membership to genetic clusters K1 and K2 and limited genetic evidence for recombination between clades was observed. These isolates could be clade 1 × clade 2 hybrids. Unfortunately, the lack of phase information for genetic markers used in this study prevents a formal test to determine clade-specific haplotypes to explore hybridization.

The cucurbit downy mildew epidemic of 2004 in the United States was a historical event due to its devastation, economic impact, and long-lasting changes in cucurbit agriculture across the eastern United States (Beans 2018; Holmes et al. 2015). This epidemic has prompted numerous studies trying to understand pathogen populations and improve disease management strategies (Naegele et al. 2016; Quesada-Ocampo et al. 2012; Runge et al. 2011; Summers et al. 2015b; Thomas et al. 2017). In this study we demonstrate that wild and commercial cucurbits are hosts of P. cubensis in the United States, that host is the main factor structuring P. cubensis populations, and that P. cubensis has two distinct, host-adapted clades at the cucurbit species level, with clade 1 showing random mating and evidence of recombination and clade 2 nonrandom mating and no evidence of recombination. Determining the level of differentiation between clade 1 and clade 2 P. cubensis regarding host specificity, virulence, morphology, fungicide sensitivity, and mating system will be valuable in refining disease management recommendations.

ACKNOWLEDGMENTS

We thank all members of the Quesada lab for their valuable help.