We selected 14 P. syringae strains that were isolated from the symptomatic tissue of watermelon, cantaloupe, and squash over the period of numerous years for shotgun sequencing, as detailed below. Most of these strains were described previously and altogether, comprised three different multilocus haplotypes located in phylogroups 2a and 2b of P. syringae (Table 1). Eight strains collected from bacterial leaf spot epidemics that occurred in Florida between 2013 and 2014 were sequenced. These strains were isolated from watermelon and squash grown in six different Florida counties during these epidemics and comprised haplotypes one and two (Newberry et al., 2018). Three strains included for sequencing comprised haplotype three and were isolated from cantaloupe and squash grown in Florida, Georgia, and California between 2000 and 2006 (Newberry et al., 2016). Finally, three additional strains that were isolated from symptomatic squash grown in Italy, Serbia, and China between 2005 and 2013 were also sequenced because they were found to be identical at four partial housekeeping gene sequences to one of the previously mentioned haplotypes (Hwang et al., 2005).

Bacterial strains were purified from single colonies and cultured overnight in nutrient broth. Genomic DNA was extracted using the CTAB-NaCl method (Ausubel et al., 1994), checked for quality using a NanoDrop 2000 (Thermo Scientific, Waltham, MA, United States) and gel electrophoresis, then quantified using a Qubit 3.0 fluorometer (Thermo Fischer, Waltham, MA, United States). Genomic libraries were prepared using a Nextera library preparation kit (Illumina Inc., San Diego, CA, United States) and the DNA was sequenced using the Illumina MiSeq platform at the Interdisciplinary Center for Biotechnology Research, University of Florida. The raw sequence data were subjected to adapter and quality trimming with Scythe¹ and SolexaQA, respectively (Cox et al., 2010). The quality-trimmed reads were then de novo assembled into contigs using the SPAdes Genome assembler (v3.5.0) with the “-careful” option to reduce mismatches in the assembly (Bankevich et al., 2012). The draft genome assemblies were submitted to the Prokaryotic Genomes Automatic Annotation Pipeline (PGAAP) pipeline and the Joint Genome Institute (IMG-JGI) server for annotation (Markowitz et al., 2012; Tatusova et al., 2016). The sequencing, assembly statistics, and collection information are presented in Table 1.

The strains sequenced in this study were investigated in context of P. syringae by including 42 additional genomes that were publicly available from the National Center for Biotechnology Information (NCBI) GenBank for analysis (Table 2). These genomes were selected to represent the diversity of phytopathogenic bacteria previously described within the species and are distributed among phylogroups 2a, 2b, and 2d (Berge et al., 2014; Bull and Koike, 2015). Representatives of phylogroup 2c, otherwise known as P. congelans, were not included in this analysis as they are not known to be phytopathogenic (Mohr et al., 2008). The genome assemblies were re-annotated with Prokka to generate GFF3 files (Seemann, 2014), which were then used as input for Roary pan-genome pipeline (v.3.6.1, Page et al., 2015). Orthologous groups were clustered using the CD-Hit and MCL algorithms with a BLASTp cut-off set to 95% along with the “-s” option to prevent the splitting of orthologous groups containing paralogs. The output of Roary was further analyzed using Scoary to test for associations between the presence/absence of orthologous groups and cucurbit-associated lineages (Brynildsrud et al., 2016). The population structure (as described below) was used to control for spurious associations in the estimation of probabilities and orthologous groups with a Bonferroni p ≤ 10⁻⁵ were reported.

Initial analysis of the population structure was conducted by calculating the average nucleotide identities between genomes using the MUMmer algorithm (Marçais et al., 2018), with the Python package pyani². Subsequently, a core genome alignment was constructed using the program Parsnp (Treangen et al., 2014). Locally collinear blocks (LCBs) of maximal unique matches shared across all genome assemblies were identified and aligned against the gold standard reference genome of P. syringae pv. syringae strain B728a (NC_007005.1). Single nucleotide polymorphisms (SNPs) located on LCBs < 200 bp or other regions of poor alignment were removed from the data set to generate a concatenated alignment of high-quality core genome SNPs. This concatenated SNP alignment was used to infer a maximum likelihood phylogeny using iQTree (v.1.6.4) with the Jukes-Cantor model of nucleotide substitution (Nguyen et al., 2015). Branch support was assessed using the ultrafast bootstrap method with 1,000 replicates (Minh et al., 2013) and the phylogenetic tree was visualized and annotated using FigTree (v.1.4.2³).

To analyze patterns of homologous recombination within P. syringae, the core genome alignment generated with Parsnp was used as input for analysis with fastGEAR using the default settings. The fastGEAR algorithm employs, BAPS, a Bayesian hierarchical clustering method to infer sequence clusters in an alignment (Corander et al., 2003). This was followed by a hidden Markov model (HMM), which was used to collapse clusters that share a common ancestry in at least 50% of the sites into lineages. The program detects “recent” recombination events between lineages using a HMM and the origin of the recombinant sequence is assigned to the lineage with the highest probability at that position. Similarly, “ancestral” recombination events that are shared by all strains which comprise a lineage are identified; however, the origin of these recombination events cannot be inferred due to their conserved nature (Mostowy et al., 2017). The statistical significance of recombination predictions was tested using a Bayes factor (BF) > 1 for recent recombination events and BF > 10 for ancestral recombination events. This analysis was conducted with two alternative genome alignments using completed reference genomes of P. syringae pv. syringae strains UMAF0158 (NZ_CP005970.1) and HS191 (NZ_CP006256.1) to validate the lineage prediction and proportion of gene flux between lineages. Finally, to examine the evolutionary history of P. syringae in the absence of recombination, the positions in the core genome alignment which corresponded to the predicted recent recombination events were removed and SNPs extracted using the program SNP-sites (Page et al., 2016). This recombination-filtered SNP alignment was then used to construct a maximum likelihood phylogeny as described above.

To characterize the functional impact of the recombinant genes leading to the emergence of phylogroup 2b-a, BLASTn (E-value ≤ 1e⁻⁵⁰) was used to map the recent recombination events predicted by fastGEAR back to the genome assemblies of three strains (13-140A, ZUM3584, HS191), representative of the different recombination profiles in the dataset. The Clusters of Orthologous Groups (COG) and Kyoto Encyclopedia of Genes and Genomes (KEGG) orthology IDs corresponding to recombinant and non-recombinant genes were subsequently extracted from the IMG annotations and a linear regression analysis was conducted to compare the composition of recombinant vs. non-recombinant core genes across the COG functional categories. Additionally, the KEGG orthology IDs were used to compare the distribution of recombinant genes across the KEGG BRITE functional annotations among the three different recombination profiles.

After noting evidence for extensive recombination affecting the hrp/hrc pathogenicity island of phylogroup 2b-a cucurbit strains in the fastGEAR output, amino acid sequence alignments for the 27 open reading frames that comprise hrp/hrc gene cluster, as described by Alfano et al. (2000), were constructed using a sub-set of 54 P. syringae genomes. Individual gene alignments were subjected to a phylogenetic analysis with FastTree 2 using the gamma time reversible model (Price et al., 2010). A gene was considered to be recombinant among the 2b-a lineages if it clustered into a monophyletic group with other of members of phylogroup 2a, rather than 2b, as predicted by the fastGEAR algorithm. To illustrate these relationships, a phylogenetic network was constructed with a concatenated alignment of the entire hrp/hrc gene cluster using the NeighborNet method and p-distance in SplitsTree 4 (Huson and Bryant, 2006).

Reference sequences were obtained from the public T3SE database⁴ and used to query the genome assemblies using BLASTn (E-value ≤ 1e⁻⁵). A given T3SE was considered to be present in a genome if the alignment with the reference sequence displayed at least 80% sequence identity over 80% of the query length. If genomic sequences orthologous to a given reference sequence were split between two contigs, then the sequences were concatenated to determine their presence in the assemblies. For each T3SE, an alignment of the subject sequence with the curated reference sequence was used to record putative frameshift mutations or other disruption of the coding sequence. The distribution of T3SEs was then used to construct a binary matrix, where 1 indicated the presence of a gene, 0.5 indicated a putative pseudogene, and 0 if the gene was absent. Hierarchical clustering of the binary matrix was performed with the Python package, hclust2⁵. A similar strategy described by Baltrus et al. (2011) was used to identify genes involved in the biosynthesis of the following phytotoxins: tabtoxin, phaseolotoxin, mangotoxin, syringomycin, syringolin, and syringopeptin.

The presence of plasmids was predicted for the strains sequenced in this study using plasmidSPAdes (Antipov et al., 2016). Assembled contigs were screened using Microbial Genome BLAST against the complete plasmid database to identify putative plasmid sequences. These contigs were subsequently screened for T3SEs and other virulence factors as described above. Genomic islands and other signatures of horizontal gene transfer (HGT) were detected using the Island Viewer 4 server (Bertelli and Brinkman, 2018). Genomic islands (GIs) were predicted using the reference strains P. fluorescens SBW25, P. protegens Pf-5, P. aeruginosa PAO1, and P. syringae pv. syringae B728a. The predicted GIs were analyzed for gene content and extracted from the genome assemblies to construct multiple sequence alignments using the Mauve software package (Darling et al., 2004).

The pathogenicity of most of the bacterial strains sequenced in this study was examined previously (Newberry et al., 2016, 2018). However, to make direct comparisons with strains collected from independent disease outbreaks, a sub-set of 10 P. syringae strains representative of the genetic diversity and geographic source of isolation in the collection were selected for further pathogenicity testing. A strain that was originally isolated from diseased proso millet, P. syringae pv. syringae HS191 (Gross and DeVay, 1976; Ravindran et al., 2015), was also included in these experiments due to its genetic similarity to the cucurbit strains examined here. Bacterial strains were cultured overnight on King’s medium B agar, suspended in a sterile MgSO₄^∗7H₂O solution (10 mM), and adjusted to ∼1 × 10⁸ colony forming units (CFU ml⁻¹) spectrophotometrically (OD₆₀₀ = 0.1). Two-week-old seedlings of watermelon cv. Troubadour (Harris Seeds, Rochester, NY, United States) and squash cv. Conqueror III (Seminis Vegetable Seeds) were sprayed with the bacterial suspension until run-off, incubated in a humidity chamber at 100% RH for 48 h, then placed under growth room conditions at 21°C and ∼70% RH with a 12 h photoperiod. Plants sprayed with the sterile MgSO₄^∗7H₂O solution served as a negative control. Two weeks after inoculation, the total proportion of necrotic/symptomatic leaf tissue was rated visually from 0 to 8 using a modified version of the Horsfall–Barratt scale (Horsfall and Barratt, 1945): 0 = no symptoms, 1 = 1–3%, 2 = 3–6%, 3 = 6–12%, 4 = 12–25%, 5 = 25–50%, 6 = 50–75%, 7 = 75–87%, 8 = 87–100%. Five replicates were included for each strain/host combination, and the experiment was conducted twice. A non-parametric analysis of variance (ANOVA) was used to test for differences among the treatments and all statistical analyses were conducted using JMP Pro 13 (SAS Institute, Cary, NC, United States).

Previous studies using MLSA described three phylogroups of phytopathogenic bacteria within P. syringae sensu stricto (Berge et al., 2014; Bull and Koike, 2015). Maximum likelihood reconstruction of the P. syringae core genome based on 246,510 polymorphic sites extracted from an alignment of 2.43 Mb revealed the presence of two well-supported phylogenetic groups in addition to those delineated through MLSA (2a, 2b, and 2d), designated here as 2b-a and Pav (Figure 1). The between-phylogroup average nucleotide identities (ANI) were primarily equal and ranged from 94.61 to 95.54% ANI. However, phylogroup 2b-a was intermediate of 2b (97.43% ANI), Pav (96.15% ANI), and 2a (96.00% ANI). Similar levels of sequence identity (between 98.16 and 98.98% ANI) were recorded within all P. syringae phylogroups (Table 3). Phylogroup 2b contained a number of strains collected from diverse monocot and herbaceous dicot plant species including the type strain, P. syringae pv. syringae ICMP3023^T, as well as other previously described pathovars such as P. syringae pvs. aptata ICMP459^PT, atrofaciens ICMP4394^PT, pisi PP1, and lapsa ICMP3947^PT. Phylogroup 2b-a branched as a sister group to 2b and contained many cucurbit strains that were isolated from plants grown in Florida, Italy, France, and Serbia between 2003 and 2014. All of these strains were of the same clonal lineage, except for ZUM3584, and shared a recent common ancestor with that of P. syringae pv. syringae HS191, which was isolated in Australia from diseased proso millet in 1969 and was more distantly related. The other novel P. syringae phylogroup inferred in this analysis, phylogroup Pav, consisted of strains associated exclusively with hazelnut decline that were classified primarily as P. syringae pv. avellanae (O’Brien et al., 2012) and branched between phylogroups 2b and 2d. The cucurbit strains located in phylogroup 2a branched from each other into two distinct lineages that were nested among other strains isolated from diverse plants and environments. A well supported split was observed between the clones collected in California and China (BS2121 and ZUM3984) and those in Florida and Georgia (03-19A and 200-1), which differed by less than 300 SNPs in the core genome alignment and represented distinct haplotypes within the lineage (Figure 1).

The sequence clusters identified by the Bayesian statistical clustering method were congruent to the phylogroups supported by the maximum likelihood phylogeny. However, fastGEAR collapsed phylogroups 2b and 2b-a into a single lineage, while the others remained distinct. The average proportion of recent recombination in the core genome of P. syringae phylogroups (± the standard deviation) was estimated to be 0.71 ± 1.13% for 2a, 1.44 ± 1.16% for 2b, 2.03 ± 0.06% for Pav, and 0.81 ± 0.20% for 2d. In contrast, the proportion of recent recombination in the core genome of phylogroup 2b-a ranged from 27.98 to 30.54%, with ≥98.69% of the recombinant sequences predicted to have originated from phylogroup 2a. Between 178 and 213 recent recombination events were dispersed across the core genome of 2b-a lineages, where ZUM3584 and HS191 displayed variation in distribution of recent recombination events in relation to the strains collected in Florida, France, and Serbia (Figure 2A). Although evidence of ancestral recombination was minimal or undetected among the 2a (0.49%), 2b (0.00%), and 2d (1.32%) lineages, 40.09% of the phylogroup Pav core genome was acquired through ancestral recombination events (Supplementary Figure S1). After removing 1,592,717 positions affected by recent recombination from the core genome alignment (64% of the original alignment), a second phylogenetic analysis was conducted using 81,421 recombination-free SNPs. Overall, the topology of the recombination filtered phylogeny was congruent to that of the unfiltered tree. However, phylogroup 2b-a appeared as a branch within phylogroup 2b, rather than as a distinct group and with considerably shorter branch lengths as compared to the unfiltered phylogeny (Figure 2B,C).

The recombinant regions predicted by fastGEAR corresponded to 3,094 coding sequences among strains 13-140A, ZUM3584, and HS191, of which 2,812 were assigned to a COG functional category. Likewise, 6,875 of 7,654 coding sequences were assigned to COG categories corresponding to regions of the core genome where no signal of recombination was detected. No significant difference in the effect of COG category on recombination was observed between strains (P = 0.15) and a strong linear relationship between recombinant and recombination-free core genes was noted across the general functional groups (R² = 0.886; Figure 3A). Although, a Grubb’s test for outliers indicated that no COG category was significantly over- or underrepresented with recombinant genes (data not shown), the top three categories skewed by recombination included amino acid transport and metabolism (E), cell motility (N), and extracellular structures (W). Functional categorization of recombinant core genes utilizing the KEGG BRITE database supported these observations. Approximately 25% of the classifiable genes impacted by recombination were ATP-dependent transporters associated with numerous metabolic pathways, followed by bacterial motility proteins and secretions systems, which each comprised approximately 7–8% of the recombinant genes in each strain (Figure 3B).

Examination of the specific secretion systems impacted by recombination indicated that many of the genes associated with the biosynthesis of type III secretion system (i.e., hrp/hrc gene cluster) were recombinant among strains 13-140A and ZUM3584, whereas no signals of recombination were detected in the hrp/hrc cluster of strain HS191. Phylogenetic analyses showed that 17 hrp/hrc genes clustered the cucurbit strains isolated in Florida, France, and Serbia into a monophyletic group with other members of phylogroup 2a and were therefore considered recombinant, while strain ZUM3584 carried 11 recombinant alleles (data not shown). A phylogenetic network constructed with a concatenated alignment of the entire hrp/hrc cluster (7,509 aa) was largely concordant with that of the core genome phylogeny. This analysis showed the phylogroup 2b-a cucurbit strains branching from other 2a lineages, while ZUM3584 was placed in a hybrid position in the phylogenetic network and HS191 clustered among members of phylogroup 2b (Figure 4).

The T3SE repertoires of the 56 genomes analyzed in this study are presented in Figure 5. The cucurbit strains carried between 17 and 21 potentially functional or disrupted effector genes, while an average of 13 T3SEs were present among other P. syringae lineages. Hierarchical clustering based on the presence/absence of effector genes revealed a loose correlation between the core genome evolution and effector repertoires, with most members of the primary P. syringae phylogroups clustering into distinct groups. The phylogroup 2b-a and 2a cucurbit strains carried similar effector profiles and clustered together along with other 2a lineages. The T3SE hopA1 was exclusive to most of the strains within this group and was noted to be the only gene present in the exchangeable effector locus of phylogroup 2b-a strains (except for HS191 which carried hopA2), along with its corresponding chaperone, shcA. The effector hopZ5 was present exclusively in the genomes of all but two cucurbit strains (13-509 and PS711) within P. syringae sensu stricto and displayed 98.27% amino acid sequence identity to the hopZ5 allele present in the P. syringae pv. actinidiae biovar 3 of phylogroup 1. This effector was linked to hopH1, present with a point deletion, resulting a frameshift mutation in the gene. Several effector genes found to be sparsely distributed across P. syringae were also identified among the cucurbit-associated lineages. Most of these genes were present on an integrative and conjugative element (ICE) or plasmid loci (see results below) and included avrRpt2 in strains 13-509A and PS711; hopAR1 in strains 200-1 and 03-19A; avrPto1, hopAU1, and hopX2 in strains 13-139B and 13-429; and hopAW1, which was common to all phylogroup 2a-cucurbit strains except 200-1 and 03-19A.

We also investigated the presence of phytotoxin biosynthetic gene clusters in the genome assemblies, which displayed a simpler distribution. The mangotoxin gene cluster was conserved across P. syringae. Evidence for the complete syringomycin, syringolin, and syringopeptin biosynthetic pathways was also present in most P. syringae genomes including the cucurbit strains examined here, although was notably absent from strains associated with diseases of woody hosts including all members of phylogroup Pav, P. cerasi 58^T, and P. syringae pv. papulans CFBP1754^PT. Additionally, phylogroup 2a cucurbit strains 03-19A, 200-1, BS2121, and ZUM3984 carried dcd2, which involved in phaseolotoxin biosynthesis, but lacked other components of this biosynthetic pathway (Supplementary Figure S2). Ravindran et al. (2015) previously noted that phylogroup 2b-a strain HS191 was negative for ice-nucleation activity and carried an ∼2 Kb truncation in the center of the ice-nucleation protein, inaZ. As the 2b-a cucurbit strains were previously found to be negative for ice-nucleation activity (Newberry et al., 2018), we investigated the presence of inaZ in the genome assemblies of these strains and found the same truncation (data not shown).

We investigated the pan-genome of P. syringae to identify orthologous groups (OGs) associated the cucurbit strains and therefore, potentially involved in niche adaptation. Among the 19,613 OGs that comprised the P. syringae pan-genome, only seven were significantly associated with both phylogroup 2a and 2b-a cucurbit strains (Bonferroni p ≤ 10⁻⁵). Among these, two displayed 100% specificity, indicating that these OGs were completely absent from the genomes of related P. syringae strains and included the T3SE hopZ5 and a hypothetical protein also present in the ICE locus. Several other OGs with known functions in virulence displayed significant associations and included the T3SE hopA1, its corresponding chaperone shcA, and the type VI secreted effector vgrG. The gene that displayed the strongest association with the cucurbit lineages was a hypothetical protein of 63 aa in length, adjacent to a putative iron-sulfur binding gene cluster (Table 4).

Analysis with Island Viewer 4 revealed that all cucurbit strains carried a predicted genomic island between approximately 90 and 125 Kb in size that was similar in structure to an ICE described in P. syringae pvs. actinidiae and phaseolicola of phylogroups 1 and 3, respectively (Pitman et al., 2005; McCann et al., 2013). The boundaries of the island were delineated by parA and xerC at opposite ends, each flanking a repeat region overlapping tRNA_lys, which serve as the attachment sites for the island (Pitman et al., 2005). The contigs corresponding to predicted ICE loci were subsequently extracted from the genomes of representative cucurbit strains (divided between one and five contigs per genome) and used to construct a multiple sequence alignment along with the complete ICE sequences of P. syringae pv. actinidiae (Psa) SR121 (Accession no. KX009066) and P. syringae pv. syringae HS191 (Figure 6). We were unable to confirm whether the predicted ICE carried by strains 13-429 and 13-139B was intact due to the presence of the pyoverdine biosynthetic gene cluster (∼60 Kb) in the center of the ICE region of these two strains and therefore, were not included in the alignment.

This analysis revealed that the putative ICEs displayed a mosaic structure characterized by differing mobile genetic elements and gene cassettes. It also revealed the overall synteny of the ICEs was conserved and that the core ICE genes displayed a high degree of homology (ranging from 96 to 99% average pairwise identity) with the ICE of Psa strain SR121. Similar to the ICE described in P. syringae pv. phaseolicola strain 1302A (Pitman et al., 2005), hopAR1 was carried in the ICE of strain 200-1 between a predicted helicase and pilL, while avrRpt2 was found at the same locus in strain 13-509A and HS191 carried a predicted type VI secretion protein (hcp1). Similarly, hopZ5 and a disrupted copy of hopH1 was present in the ICEs of all cucurbit strains except 13-509A, which carried hopC1 and an intact copy of hopH1 at the same locus. Both of these effector pairs were flanked by a predicted phage integrase and site-specific recombinase (xerD) that shared 97.05 and 93.13 aa identity, respectively.

The effector hopZ1 was also present in the ICE of all but two cucurbit strains (ZUM3584, 200-1) and was located approximately 10 Kb downstream of hopZ5/hopC1, near the boundary of the island. Interestingly, hopZ1 was also carried by strain HS191 and was linked to a similar integrase/recombinase gene cassette. However, in HS191, hopZ1 was present in the exchangeable effector locus, as indicated by the presence of queA and tRNA_leu, which delineate the boundary of this pathogenicity island (Supplementary Figure S3). Several other accessory genes and putative virulence factors were identified in ICEs. Strain ZUM3985 carried hopAW1 at a locus unique to the ICE found in this strain. It also carried a predicted calcium binding protein (1,691 aa) of the RTX toxin superfamily, which is a protein family commonly secreted via the type I secretion system (Linhartová et al., 2010). Likewise, a cassette of small peptides between 62 and 83 aa in length with a predicted papB domain, which is involved in the regulation of adhesin biosynthesis (Xia et al., 2000), was conserved in ICE of several strains including 13-140A, ZUM3584, 200-1, HS191, and SR121 (Figure 6).

Analysis with plasmidSPAdes provided no evidence of plasmid sequences in phylogroup 2b-a cucurbit strains, whereas several were assembled among strains within phylogroup 2a. These results were consistent with the size of genome assemblies, which averaged 5.92 and 6.23 Mb for the 2b-a and 2a cucurbit strains, respectively (Table 1). A 52.42 Kb contig was assembled in strains 200-1 and 03-19A which displayed 93% nucleotide identity with 66% query coverage to the complete plasmid sequence from strain HS191 (NZ_CP006257.1). This plasmid carried the effectors hopC1 and hopH1, which marked the second copy of hopH1 for these two strains, in addition to the disrupted allele present in the ICE. A putative virulence plasmid of 16.51 Kb in size was assembled in strains 13-429 and 13-139B which harbored four accessory effector genes including hopAU1, hopAW1, hopAF1, and avrPto. The top hit for this contig in the NCBI complete plasmid database was plasmid PP3 (NZ_LT963405.1) of P. syringae pv. avii strain CFBP3846 and displayed only distant homology (95% nucleotide identity and 39% query coverage). Strains BS2121 and ZUM3984 also carried two putative plasmids, for which the top BLAST hits included the complete plasmids pCC1557 (NZ_CP007015.1) and pB13-200A (NZ_CP019872.1). However, no apparent virulence factors were identified in these contigs. A summary of the origin of T3SEs among representative cucurbit strains is presented in Figure 7C and a compiled list of putative plasmid sequences and BLAST results are available in Supplementary Table S1.

The results from two independent pathogenicity experiments produced similar results and showed significant differences in disease severity between the treatments (P < 0.0001). Phylogroup 2b-a strains 13-140A, 13-C2, and ZUM3854 induced expanding lesions and foliar blighting on watermelon and squash, with mean severity ratings ranging from 5.60 to 7.50 among the three strains. In contrast, the squash strains 13-509A and PS711 produced only superficial lesions on both hosts and mean severity ratings that did not significantly differ from that of the millet strain, HS191 (mean severity ratings ranging from 0.70 to 2.40). Two effector pairs differentiated the virulent from weakly virulent strains within this group and included hopZ5 and the disrupted copy of hopH1 among the virulent strains, while the weakly virulent strains carried hopC1 and an intact copy of hopH1. Within phylogroup 2a, no significant differences in disease severity were observed among strains BS2121, ZUM3984, and 13-139B, which produced moderate to high levels of disease severity on both hosts. While strains 03-19A and 200-1 also induced expanding lesions and foliar blighting on squash (between 5.00 and 7.00 mean severity), they were weakly virulent on watermelon (between 1.40 and 2.00 mean severity). Four T3SEs were unique to the squash limited strains of phylogroup 2a and included hopC1, an intact copy of hopH1, hopAR1, and hopM1 disrupted by a frameshift mutation. Likewise, hopAW1, hopZ1, and an intact hopM1 allele were unique effectors common among the strains virulent to both watermelon and squash (Figure 7).