Adaptation without Dominance in Pseudomonas syringae Pathovars

Abstract

Understanding local adaptation of phytopathogens has significant practical and economic implications. The opportunistic pathogen Pseudomonas syringae exemplifies this challenge, causing regular epidemics in diverse host plants. Many pathogenic microbes, including P. syringae, are divided into intraspecific lineages, or pathovars, based on their host-of-isolation. However, whether pathovar classifications reflect adaptation of the pathogen to the host (local adaptation) or a competitive advantage of the pathogen in the host (local dominance), often goes untested. In this study, we performed in vitro growth assays and factorial controlled infections to test whether a suite of five P. syringae pathovars are locally adapted to, and/or locally dominant in, their hosts-of-isolation. We found evidence of local adaptation in three of five pathogens, only one of which was also locally dominant. Several strains performed as well or better than the locally adapted strain in that strain’s host-of-isolation, consistent with cost-free generalism. Thus, pathovar designations do not reliably delineate pathogenic phenotypes. Moreover, we found that in vitro growth was not predictive of in planta growth. To contextualize phenotypes, we compared pathogen gene content, identifying unique phytotoxins, secreted effectors, and general virulence factors. In all, we found that local adaptation is common but not universal, and that locally adapted strains are not necessarily constrained from performing competitively in multiple hosts. Thus, neither host-of-isolation nor in vitro performance is reliable for strain classification. Our findings highlight the vast intraspecific variation in P. syringae, and the coexistence of multiple successful adaptive strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00248-025-02675-1.

Introduction

Ecological specialization is a central driver of taxonomic diversity [1–3]. An early step in this process is differential adaptation of populations within a species [4], which can lead to local individuals having greater fitness in sympatric relative to parapatric hosts, or local adaptation [5]. Conversely, generalism occurs when individuals maintain the ability to grow in many hosts [6], with no host-specific advantage. In theory, niche breadth is constrained by genetic tradeoffs that impose costs of adaptation [2, 7–9], such that improvement in one environment requires decline in another [8, 10–12]. These costs of adaptation are expected to burden generalists by constraining their performance across hosts, and specialists by restricting optimal performance to a single host. Thus, a jack-of-all-trades generalist should be master-of-none, unable to outperform a specialist in the specialist’s host (cost of generalism). At the same time, specialists should suffer poor performance on alternate hosts (cost of specialization). Experiments with microbes have found that costs of generalism readily evolve [13–16], but examples of cost-free generalism, where microbes selected in multiple hosts perform as well as those selected in the individual hosts, are also common [17–20]. Likewise, for microbes evolved in the presence of a single host or selective pressure, costs of specialization are often [21–26] but not necessarily [27–29] detected. A mechanistic explanation for these seemingly contradictory findings is that not all beneficial mutations involve performance tradeoffs, and not all fitness declines are associated with adaptation (e.g., mutations neutral in the selective environment can impose declines in alternative environments). Thus, host adaptation occurs via varied genetic trajectories.

Microbial phytopathogens reduce crop yields and cause major economic losses that are intensifying as global food demand tracks population growth [30–33]. However, predicting and controlling plant disease remains challenging for several reasons, including high intraspecific genetic diversity amongst pathogens, a minimal understanding of the spatial scale at which populations differentiate [34, 35], and the fact that classification methods have long been inquorate [36–42]. A focus on plant pathology, and the fact that in planta phenotypes can sometimes better distinguish strains than in vitro phenotypes (e.g., Xanthomonas [43]; Agrobacterium [44]; and Pseudomonas [45]), led to the convention of classifying strains based on host-of-isolation [36, 46]. Accordingly, many species have been sub-divided into pathovars (pv.), or pathogenic lineages with distinct host ranges (Erwinia amylovora [47]; Ralstonia solanacearum [48]; Xanthomonas campestris [49]). Using host-of-isolation as the defining character implies that the strain grows especially well in (is locally adapted to) that host; however, where a strain is detected depends on many factors in addition to host compatibility, including the strain’s competitiveness across hosts and against other strains. Moreover, local adaptation requires multiple experiments to confirm, and new microbial taxa are being discovered increasingly rapidly [50], such that local adaptation often goes untested in practice, and its prevalence remains an open question.

The bacterium Pseudomonas syringae provides an excellent system in which to test this question. The P. syringae species complex is divided into more than 60 pathovars that span 13 canonical phylogroups [51, 52], yet several cases reveal incongruence between pathovar designation and host specificity. First, greenhouse infections have demonstrated that some strains have generalist host ranges [53], and first reports have identified pathovars capable of causing outbreaks on alternative hosts, e.g., pv. tomato (or Pst) on kale [54] and pv. tabaci on coffee [55]. Second, pathogenic varieties of P. syringae have been isolated from non-host sources including snow, rain, and irrigation waters [56]. Finally, host-of-isolation is only poorly associated with core genome content [52, 57, 58], indicating a disagreement between phenotypic and genomic methods of classification. Since P. syringae regularly causes major disease epidemics in diverse hosts (cantaloupe [59]; horse chestnut [60]; kiwifruit [61]; and cherry [62]) a deeper understanding of its host associations is needed for predicting and managing future outbreaks.

Though local adaptation is often thought to predominate in nature, host–pathogen relationships can take several, non-mutually exclusive forms. Traditionally, local adaptation is inferred where a pathogen performs better in its host-of-isolation than other hosts (Fig. 1a). If local adaptation is widespread, pathogens should also be locally dominant, outperforming competitors in their host-of-isolation (Fig. 1b). Thus, a strain exhibiting both local adaptation and local dominance would suggest a cost of specialization, as that strain grows competitively in only one host. Alternatively, a pathogen may perform better in its host-of-isolation without specialization (e.g., via beneficial mutations that are neutral in alternative hosts). This pathogen would still grow best in its host-of-isolation, but may dominate in other hosts due to the lack of underlying performance tradeoffs. A cost of generalism is suggested where a generalist is outcompeted by a locally adapted strain in that strain’s host-of-isolation. Conversely, cost-free generalism would be inferred where a generalist dominates (outcompetes) the locally adapted strain.

Fig. 1.

Schematic of local adaptation vs. local dominance. Local adaptation (a) is inferred when a pathogen performs better in its host-of-isolation than alternative hosts. Local dominance (b) is inferred when a pathogen outperforms other pathogens in its host-of-isolation

At the molecular level, pathogen adaptation involves the gain, loss, or alteration of genetic elements that promote infection. Strains carry phytotoxins, adhesins, motility factors, and multiple effectors, or secreted virulence factors that benefit the pathogen by manipulating the host or surrounding microbes to facilitate infection [63–65]. Effectors represent ‘double edged swords’ that promote infection but may also trigger host immune responses [66–68]. In an evolutionary ‘arms-race’, pathogens evolve to evade effector recognition, and hosts counter-adapt to restore recognition and mount resistance responses [69–72]. Conventional wisdom is that pathogens have the advantage in the arms-race, due to their shorter generation times and larger populations [73–76]. Supporting this, strong in planta selection has driven loss-of-function in many pathogen effectors: X. axonopodis escaped pepper recognition via transposon insertion in avrBs2 [77], Pyricularia oryzae evaded wheat through mutations in PWT3 [78], and Leptosphaeria maculans virulence in Brassica napus is associated with effector loss [79]. Yet, reports of local maladaptation in phytopathogens indicate that the host is sometimes ahead (P. syringae to Arabidopsis thaliana [80]; Microbotryum violaceum to Silene latifolia [81]; and Microbotryum carthusianorum to Dianthus carthusianorum [82]. These findings underscore that pathogen adaptation is dynamic and shaped by an interplay of evolutionary pressures that can tip the balance in favor of either player.

Herein, we used a reciprocal transplant design and comparative analyses to quantify local adaptation and local dominance across five P. syringae pathovars. We prioritized phylogenetically diverse, epidemiologically relevant strains to maximize the likelihood of finding local adaptation: three well-characterized laboratory strains (DC3000, ES4326, and 1448 A) and two more recent isolates (A9 and NP29). We measured growth in three non-host (liquid media) environments to assess performance in the absence of a host, and conducted reciprocal infections to measure in planta performance across all host–pathogen combinations, estimating growth rate (r) and carrying capacity (K). To assess local adaptation, we compared performance of each strain in the host-of-isolation vs. all other hosts, inferring local adaptation when growth was significantly higher in the host-of-isolation (i.e., Fig. 1a). To test for local dominance, we compared performance of each strain to all other strains on a given host, inferring local dominance when the local strain had a significant competitive advantage (i.e., Fig. 1b). To distinguish local adaptation from any genotype being a generally better performer, we compared each strain’s overall performance across hosts. Additionally, we analyzed pathogen genome characteristics and gene content to contextualize growth phenotypes and identify the potential molecular bases of adaptation. This design enabled a direct test of the often untested assumption that P. syringae pathovars are locally adapted to their hosts-of-isolation.

Methods

Experimental System

We selected virulent P. syringae strains, including pathovars of bean, radish, and tomato, and an A. thaliana isolate (Table 1). As the original cultivars were unavailable, we sourced phenotypically similar cultivars from the same geographic regions. A9 was isolated from tomato in Yolo County, California in 1996 [83], and DC3000 was isolated in Guernsey Island, UK in 1961 [84]. Race 1 strains like A9 displaced race 0 strains like DC3000 as the dominant tomato isolates in the early 2000 s [85–87]. We matched the tomato pathovars with cultivars from northern California (A9) and Guernsey Island (DC3000). ES4326, isolated from radish in Wisconsin in 1965 [88], was later reclassified as P. cannabina pv. alisalensis, the dominant Brassica pathovar [89]. We matched ES4326 with a common 1960 s Midwest USA cultivar of radish. Race 6 strains are known to dominate bean infections [90]. 1448 A is a race 6 strain isolated from bean in Ethiopia in 1985 [90], which we matched with an east African bean cultivar. NP29, isolated from A. thaliana in Michigan in 2002 [91], belongs to an Operational Taxonomic Unit (OTU5) that dominates A. thaliana [92]. We matched NP29 with RRS10, a representative Midwest A. thaliana ecotype, due to its similarity to NP29’s host-of-isolation [93].

Table 1.

Genomic characteristics of pathogen strains

Pathovar	Strain	PG	Host	Length	% GC	Genes	Singletons
tomato	A9	1	Tomato cv. Brandywine	6,295,740	58.8	5,744	429
tomato	DC3000	1	Tomato cv. Guernsey	6,538,260	58.3	5,858	404
maculicola	ES4326	5	Radish cv. Cherry Belle	6,606,571	58.3	6,054	1,022
phaseolicola	1448A	3	Bean cv. African Premiere	6,112,448	57.9	5,573	540
–	NP29	2	A. thaliana ecotype RRS10	5,992,538	59.3	5,146	453

Estimates of CFU/mL by Optical Density

For both in vitro and in planta experiments, we needed to compare cell quantities across strains at various 600 nm optical densities (ODs). To enable these comparisons, we confirmed that CFU/mL did not differ across strains at equivalent ODs. Overnight cultures were grown from single colonies, pelleted, and resuspended in sterile buffer to ODs of 1, 0.1, and 0.01. Cultures were then diluted for spread-plating and incubated at 28 °C for two days before colony enumeration. CFU/mL was calculated by multiplying the number of colonies by the dilution factor and dividing by volume plated (Fig. S1). A one-way ANOVA failed to find an effect of strain on CFU/mL (F = 0.568, p = 0.687), suggesting equivalent cell quantities at a given OD.

Growth in vitro

Bacteria were cultured on King’s Medium B (KB) at 28 °C, shaking at 175 rpm. For in vitro growth, we used two typical cultivation media with distinct nutrient profiles, KB and Lysogeny Broth (LB), plus one novel medium created in-house. We made Arabidopsis Broth (AB) from mixed A. thaliana plants collected across the Midwest, USA. Plants were washed with ethanol and reverse osmosis (RO) water, then macerated and filtered through a 0.2 µm mesh, and finally autoclaved twice for 60 min. Thus, AB contained the nutrients and chemical elicitors in the A. thaliana host. Single colonies were grown overnight, diluted 1:200, incubated at 28 °C in a Tecan plate reader, and OD was measured every 15 min for 18 h. All combinations of bacteria and environments were included in each plate for a total of three replicate measures. Growth was analyzed using the R package growthcurver [94].

Growth in Planta

The in planta experiments included all host–pathogen combinations in each of four fully randomized blocks. Each block was treated as a biological replicate, and blocks were conducted sequentially over time due to the scale of the experiment. Hosts were synchronized to the same developmental stage by planting bean and radish 12 days after A. thaliana and tomato, and inoculation time was set by the age of A. thaliana (21 day-old-seedlings). Seeds were sown into autoclaved potting soil in four-inch 18-cell flats, watered as needed by soaking with RO water, and supplemented with Jack’s 15–30-15 fertilizer one week prior to infection. Growth chambers were set to a 16-h photoperiod at 22 °C, and flats were rotated within and between shelves every 48 h to minimize random effects of position.

Overnight cultures were grown from colonies, diluted 1/100, grown for four to six hours, and normalized to an OD of 0.2. Plants were spray-inoculated until dripping wet, dried for 12 h, then randomized for the experiment. To prevent cross-contamination, plants were arranged in a checkerboard pattern with every other cell left empty. A sterile buffer mock treatment was included; day one samples served as baseline (uninfected) bacterial levels for each host. Infections reached significantly higher titers than mocks (unpaired t-test, p < 2.2e-16). We harvested two randomly selected leaves (whole rosettes for A. thaliana) per infection at one, three, and five days post-inoculation. Samples were washed in ethanol and RO water (30 s each), and six (six-mm) hole punches were used for total DNA extraction [95]. Bacterial titers were quantified with qPCR targeting oprF [96] after primer specificity was confirmed (single amplicon per strain and none from hosts). 10 μL reactions were run in 384-well plates using PowerUP SYBR MasterMix (cat. A25741). Cell quantity was calculated using the standard curve method: log10(cell quantity) = (CT-b)/m, with no differences in standard curves per strain (two-way ANOVA: F = 0.313, p = 0.868). K was defined as the peak titer; r was estimated by fitting linear models to bacterial quantity over time (initial vs. K), and extracting the slope (r) and time coefficient p-value. Significant growth (p < 0.05) was observed in 19 out of 25 infections; the rest approached significance (p < 0.09; Table S1).

Statistical Analyses

Statistical analyses were done in RStudio v2023.12.0 + 369 [97]. For both experiments, we used two-way ANCOVA on log transformed data with plate as covariate and plant and pathogen as fixed effects, including their interaction (e.g., K ~ environment * pathogen + plate). In all cases, we confirmed assumptions of ANCOVA, including linearity of regression slopes and a lack of an interaction between the covariate “plate” and either grouping variable “plant” or “pathogen” (p > 0.05). For the in vitro experiments, the two-way ANCOVA was followed with post hoc contrasts, i.e., a Tukey’s Honest Distance test on all pairwise comparisons. For the in planta experiments, we followed the two-way ANCOVA with planned contrasts to test for 1) performance in the host-of-isolation vs. all other hosts for each pathogen, 2) performance of each pathogen vs. all other pathogens for each host, and 3) performance of each pathogen vs. all other pathogens across all hosts.

Genome Assemblies

DC3000, ES4326, and 1448 A already had high-quality genomes, but the published A9 and NP29 genomes each had over a hundred contigs. To facilitate more accurate genomic comparisons, we generated additional short- and long-reads to combine with the existing reads and built new assemblies for A9 and NP29. Paired-end sequencing was done at the UChicago Core on a Miseq with a v2 kit (2 × 150 bp), and long-read sequencing was done in-house on a MinION Flongle R9.4 flowcell (courtesy of Meren Lab). Hybrid assemblies were generated with Unicycler v0.4.8 and gcc v6.2.0, polished with pilon v.1.22 [98], quality controlled with BUSCO v5.8.2 [99], and annotated with Bakta v1.9.3 [100]. The NP29 assembly produced one chromosome and one plasmid (N50 5 MB); the A9 assembly was less complete (N50 367 KB) but reduced from 188 to 28 contigs. Raw reads and assemblies are uploaded to NCBI BioProject accession PRJNA1335816.

Pangenome and Phylogeny

We used the open-source pipeline anvi’o v8 [101] to examine pangenomes of our five focal pathogens alone (Table 1) and with 32 additional strains from various hosts (Table S2). The pipeline entailed: (1) identifying open reading frames using Prodigal [102] v2.60X, (2) identifying single copy core genes using HMMER [103] v3.2.1X and built-in HMM profiles, and (3) pangenome construction with the flag –mcl-inflation 10 to account for high relatedness. We calculated Average Nucleotide Identity using fastANI [104]. For the phylogeny, we extracted single copy core genes, removed alignment gaps present in > 50% of sequences using trimAl [105], and inferred a maximum likelihood tree with IQ-TREE [106] using the ‘WAG’ model and 1,000 bootstraps. The tree was annotated in iTOL [107].

Gene Content Analyses

The laboratory strains Dc3000, ES4326, and 1448 A have functionally verified Type III Secreted Effector (T3E) repertoires (e.g., DC3000 and ES4326 [108]; 1448 A [109]), but T3E are less well-studied in A9 and NP29. To identify T3Es in these strains, we used the PsyTec Compendium [110] with a custom nucleotide BLAST + pipeline [111], filtering matches for > 90% identity and an E-score < 0.01. Note that this method includes effectors that have not yet been functionally verified and may not be expressed or secreted. For multiple calls at one locus, we kept the largest allele. For all strains, T3E families are reported, rather than individual alleles, in accordance with convention [110]. To identify general virulence factors, we used ABRicate version 0.8.13 [112] with the database VFDB 2.0 [113], keeping all matches with > 80% identity. We manually added virulence factors that were missed with the computational methods but present in the Bakta annotations. To analyze singletons (Table S3), we extracted their sequences from the expanded pangenome (Table S2) and re-annotated them with Bakta.

Results

Growth in vitro

To observe growth in the absence of a host, we grew our pathogens in three non-host (liquid media) environments (Fig. 2). A two-way ANCOVA on r indicated significant effects of environment (F = 11.54, p = 0.0002), pathogen (F = 26.77, p < 3.263e-09), their interaction (F = 2.70, p = 0.0244), and the covariate plate (F = 3.97, p = 0.0303). The significant interaction indicates that pathogen growth rate varied by environment. We expected higher growth in KB, as it was designed for the cultivation of Pseudomonads [114]. Instead, a Tukey post-hoc test showed that pathogens had similar r in AB and KB, which were both significantly higher than in LB (AB-LB, p = 0.0283; KB-LB, p = 0.0001). This post-hoc test further revealed that the tomato pathovars bracketed the extremes over all environments (Fig. 2a): A9 had significantly lower r than all others (p < 0.0005), and DC3000 had significantly higher r than all others (p < 0.003). These patterns are not explained by metabolic costs being proportional to genome size, as the two strains have similarly, intermediately sized genomes (Table 1), and more likely reflect differences in gene content. For K (Fig. 2b), NP29, the Arabidopsis isolate, significantly outperformed all other strains across environments (p < 0.0001). Interestingly, this was not driven by a maximum K in AB, despite its nutritional similarity to NP29’s host-of-isolation.

Fig. 2.

a, b Summary of in vitro growth after 18 h. The x-axis at bottom indicates environment and at top indicates pathogen. The y-axes show (a) growth rate per hour and (b) carrying capacity. Points and error bars show the mean and standard deviation over three biological replicates, with smaller points showing the individual estimates. Pathogens are distinguished by color

Growth in planta

We next performed reciprocal infections of all host–pathogen combinations. To aide interpretation of our comparisons, we summarized the results of our planned contrasts in Table 2. First, we assessed local adaptation by comparing pathogen r in the host-of-isolation vs. all other hosts (e.g., Fig. 1a). A two-way ANCOVA revealed that pathogen r varied significantly by host (Fig. 3a), generating significant main effects of pathogen (F = 18.65, p = 1.280e-09), host (F = 28.23, p = 1.397e-12), and their interaction (F = 33.95, p < 2.2e-16), with no effect of plate (F = 2.68, p = 0.1074). Thus, pathogen r depended on host identity. Two pathogens (A9 and NP29) achieved their highest r in their host-of-isolation (Fig. 3a). Planned contrasts between r in local vs. all other hosts were significant in these cases (A9: p = 0.0002; NP29: p < 0.0001), supporting local adaptation. Repeating the ANCOVA with K yielded only a significant main effect of pathogen (F = 12.559, p = 1.813e-09), indicating that r was more sensitive to host identity, and that despite host-specific differences in r, most pathogens ultimately reached similar K across hosts. This was supported by a significant contrast for NP29 (p = 0.0425), but not A9 (p = 0.1114). 1448a also reached significantly higher K on its host-of-isolation (p = 0.0447), consistent with local adaptation for K but not r. Notably, NP29 was the only strain to demonstrate local adaptation across both metrics, with significant contrasts for both r and K on A. thaliana vs. all other hosts.

Table 2.

Summary of comparisons of in planta growth

*^{ES4326 did not grow best in its host−of−isolation and was thus not locally adapted, despite the significant p−value}

**^{In these infections, significance does not indicate local dominance as the host supported a common growth maximum rather than a single dominant pathogen}

p-values are from planned contrasts to compare growth metrics across strains and hosts, with significant values bolded, and values that were significant but not indicative of the larger trend italicized and explained in footnotes. Comparisons: Local Adaptation, contrasts between each strain in its host-of-isolation vs. all other hosts; Local Dominance, contrasts between each strain vs. all other strains in each host; and Overall Performance, each strain vs. all others across all hosts

Fig. 3.

a, b Reciprocal infections reveal local adaptation. The x-axis at bottom indicates host and at top indicates pathogen. The y-axes show (a) growth rate per hour and (b) K as bacterial quantity after logarithmic transformation to base 10. Points and error bars show the mean and standard deviation over four biological replicates, with smaller points showing the individual estimates. Pathogens are distinguished by color, with grey rectangles indicating the host-of-isolation, and asterisks indicating significant contrasts in performance on the host-of-isolation vs. all other hosts where the highest estimate was also on that host

Having detected evidence of local adaptation in our system, we asked whether strains were also locally dominant (e.g., Fig. 1b). To do this, we performed additional planned contrasts of r and K for each strain vs. all other strains on each host, interpreting local dominance where the local strain reached significantly higher growth. We note that this analysis is an indirect estimate limited by preclusion of ecological interactions that may favor less virulent strains [91]. For A9’s host-of-isolation, tomato cv. Brandywine, A9 had significantly higher r (p < 0.0001) and K (p = 0.0368) than all other pathogens, suggesting a meaningful competitive advantage for this pathovar in this cultivar. In DC3000’s host-of-isolation (tomato cv. Guernsey), A9 again had higher r (p < 0.0001), but DC3000 had higher K (p = 0.0023), reflecting strain-specific strategies and partial evidence of local dominance for DC3000. For the remaining hosts, K did not significantly differ, consistent with our previous ANCOVA showing no effect of host on K. For ES4326’s host-of-isolation (radish), DC3000, ES4326, and 1448 A had equivalent and significantly higher r than the remaining pathogens (ANOVA, p > 0.4; contrasts < 0.03), such that three strains reached one growth maximum rather than any strain dominating. For bean, 1448A’s host-of-isolation, ES4326 had significantly higher r than all other pathogens (p < 0.0001). Thus ES4326 demonstrated a competitive advantage over the locally adapted 1448 A, consistent with cost-free generalism. For NP29’s host-of-isolation A. thaliana, A9, ES4326, and NP29 all performed equally well and significantly better than the remaining strains (ANOVA, p > 0.8; contrasts < 0.01), again revealing a common growth maximum rather than dominance of a single strain. In all, pathogens tended to perform competitively in their host-of-isolation, but not exclusively, with most strains reaching growth maxima in multiple hosts (Table 2).

Finally, we performed a third series of comparisons to distinguish patterns of local adaptation from broadly high performance across hosts. This involved additional planned contrasts of r and K for each strain vs. all others across all hosts. For r, this analysis revealed a clear standout, with A9 reaching significantly higher values over all hosts (estimate = 0.0293, p = 0.0413), and no other significant differences. For K, DC3000 reached significantly higher overall values (estimate = 0.3899, p = 0.0152). Thus, both tomato pathovars appear to be strong generalists, with A9 also demonstrating a marked preference for its host-of-isolation. In contrast, 1448 A had significantly higher K in its host-of-isolation, but significantly lower Ks overall (estimate = −0.2970, p = 0.038), indicating a cost associated with specialization to bean that was not correlated with local dominance. There were no other significant contrasts.

Pangenome and Phylogeny

To contextualize phenotypes and identify the potential molecular bases of adaptation, we analyzed pathogen genome characteristics and gene content. The five strains comprised a pangenome of 27,741 genes categorized into 8,106 homologous gene clusters by amino acid similarity (Fig. 4). Of these, 44% (3,569) were core genes (present in all strains), while pathogens contained 404 to 1,022 singleton genes (present in only one strain) (Table 1). To estimate sequence divergence among genomes, we calculated pairwise Average Nucleotide Identity (ANI) [104, 115–117]. This revealed striking intraspecific variation congruent with phylogroup assignments (Fig. 4). Only A9 and DC3000 shared an ANI typical of genomes within the same species (> 95% [104]) whereas all other pairs shared ANI well below this cut-off. To view our focal strains within the broader species complex, we constructed a phylogeny incorporating isolates from all primary phylogroups (Fig. 5). This highlighted the close relatedness of the tomato pathovars, and revealed that NP29 and 1448 A are more closely related to each other than to the other strains, despite their low pairwise ANI and classification into different phylogroups. This aligns with a prior report that P. syringae isolated from A. thaliana resemble bean pathovars at five housekeeping loci [118]. Interestingly, despite their genetic proximity, 1448 A and NP29 exhibited distinct host preferences. Together, these findings underscore the vast standing genetic variation in this species, suggesting that gene presence/absence polymorphism underlies the phenotypic diversity we observed.

Fig. 4.

Pangenome of the five pathogens. 8,106 gene clusters are grouped by co-occurrence, such that shared clusters are closer to each other. Genomes are organized by Euclidian distance and ward ordination with pathogens distinguished by color. The matrix at top right shows ANI, with darker shades corresponding to more similar genomes. The dendrogram at top right was constructed using hierarchical clustering by gene cluster occurrence

Fig. 5.

Pathogen phylogeny derived from single copy core gene genes and rooted at the outgroup (P. fluorescens strain SBW25). Colors differentiate the primary phylogroups (1: orange, 2: chartreuse, 3: aquamarine, 4: yellow, 5: pink, 7: dark green, and 10: purple)

Gene Content Comparisons

We next examined gene presence/absence among strains to highlight candidate genetic factors that may contribute to the observed phenotypic differences. We initiated our gene content comparisons by focusing on T3Es, which represent well-studied host interactions [110]. Each strain harbored intact Type III Secretion System structural genes, and T3E family counts ranging from 5 (NP29) to 36 (DC3000) (Fig. 6a). Two families (avrE and hopB) were common to all strains, and are furthermore present in > 95% of sequenced P. syringae isolates [119]. All strains except NP29 also carried at least one unique T3E family (Fig. 6a, b). The tomato pathovars encoded four effectors (hopA, hopL, hopS and hopY) absent from all other strains. Each of these failed to elicit an immune response in five tomato cultivars [120], suggesting they contribute specifically to growth on tomato. A9, which was locally adapted to tomato and demonstrated the strongest overall performance, encoded 25 T3E families, including two that were unique: hopBV and hopBX. While hopBV is a small family of likely pseudogenes [119], hopBX is believed to target chloroplasts and disrupt photosynthesis [121]. DC3000 possessed the largest T3E repertoire and encoded eight unique families (Fig. 6a, b), most of which are known to manipulate plant immunity (e.g., the canonical avrPto suppresses immune responses in tomato [122] and A. thaliana [123]). The remaining strains had smaller T3E repertoires. ES4326, which grew competitively on many hosts, carried three unique T3Es: hopBG, hopBO, and hopZ (Fig. 6a, b). While hopBG is rare [88] and not functionally characterized, hopBO, a cysteine protease, elicits an immune response in only some bean cultivars [124] and not in A. thaliana [108, 110], and thus may have enhanced the performance of ES4326 on these hosts. hopZ alleles are found in P. syringae isolated from diverse hosts [125] with which they engage in arms-race dynamics [126]. 1448 A encoded two unique T3E families including avrB (Fig. 6a, b), which is recognized by cognate resistance proteins in A. thaliana [127] and soybean [128], but not in common bean [128], making it a likely factor in local adaptation of 1448A. hopAW is associated with disease on hosts absent from our system, suggesting limited relevance to our phenotypes (coffee [129]; Prunus species [130]; and cherry [131]. NP29, isolated from A. thaliana, had the smallest T3E repertoire, yet showed the strongest evidence of local adaptation (by both r and K metrics), suggesting that adaptation to A. thaliana may favor effector loss.

Fig. 6.

a, b Pathogen effector families. (a) Venn diagram showing overlapping effector families, and (b) table of unique effector families per pathogen

Beyond effectors, pathogens use a range of other virulence factors to establish and promote infections. All strains shared genes related to biofilm formation, flagellar biosynthesis, coronatine production, and pyoverdine production (Fig. 7). Notably, 1448 A and NP29 carried unique pathogenicity islands encoding phytotoxins: 1448 A encoded phaseolotoxin with amtA, argK, and an 18-gene pht cluster, while NP29 encoded syringomycin and syringopeptin with syr and syp. Both strains exhibited growth patterns consistent with local adaptation, suggesting that these phytotoxins contribute to host specificity. In contrast, A9 lacked a unique phytotoxin, indicating that local adaptation can occur through alternative mechanisms.

Fig. 7.

Presence/absence of general virulence factors and phytotoxins. The y-axis indicates pathogen, with colored boxes indicating gene presence for a particular pathogen. The x-axis at bottom indicates gene names, and at top lists functional categories. Abbreviated functional categories are corona: coronatine; phaseo: phaseolotoxin, pyover: pyoverdine, s-myc: syringomycin, and s-pep: syringopeptin

Finally, we examined singletons (strain-specific genes), which were historically thought to encode rapidly evolving, nonessential proteins that were tending toward extinction [132–134], but have more recently been recognized as potential drivers of ecological adaptation [135–137]. Excluding hypothetical and “domain-of-unknown-function” annotations, we highlight a curated subset of singletons most likely involved in host–pathogen interactions (Table S3).

In the tomato pathovars, three otherwise unique genes are shared: a TIR domain protein that induces host cell death [138], a TniQ protein involved in DNA transposition [139], and a KAP NTPase linked to phage exclusion [140]. A9 contained 51 singletons, including numerous histidine kinases which are often linked to virulence factor expression [141], and a secreted RING type E3 ubiquitin transferase that suppresses chloroplast function and enhances susceptibility in A. thaliana [142]. DC3000 contained 87 singletons, including many transcriptional regulators like marR, which regulates stress response and virulence genes in Escherichia coli [143], and copG, which modulates expression of Type III and IV Secretion Systems in Bradyrhizobium sp. SUTN9-2 [144]. DC3000 also encoded a ryanodine receptor (RyR domain protein), which binds plant alkaloids [145]. ES4326 had both the largest genome and number of singletons (487). These include Inducer of phenazine A/B, associated with the production of a broad-spectrum antibiotic that acts as a virulence factor and biocontrol agent [146], and a Big-1 domain adhesin that mediates host interactions with the cell membrane [147]. 1448 A had the second-highest number of singletons (130), including multiple transcriptional regulators, transmembrane proteins, and unique adhesins such as afaD, which facilitates internalization of E. coli into host cells [148]. 1448 A also harbors a gene from the pectate lyase superfamily, involved in plant cell wall degradation [149]. NP29 had the fewest singletons (39), many of which had stress response functions, including a CsbD family protein [150], two PIN domain proteins associated with toxin-antitoxin systems [151, 152], and two response regulators [153]. These strain-specific genes, especially those related to host manipulation and secondary metabolite production, underscore the diverse molecular strategies that drive pathogen fitness and host adaptation in this system.

Discussion

Understanding how pathogens adapt to their hosts is a central question in evolutionary biology with important applications to plant disease epidemiology. Herein, we sought to describe patterns of in vitro and in planta growth, as well as identify potential genetic mechanisms, across five P. syringae pathovars. We found evidence of local adaptation in three pathovars (A9, 1448 A, and NP29), indicating that, while common, local adaptation is not universal. Surprisingly, local adaptation was rarely (only for A9) associated with local dominance, with most hosts supporting a common growth maximum rather than favoring a single dominant strain. Furthermore, our strongest (A9) and weakest (1448A) overall performers were both locally adapted, indicating that local adaptation does not necessarily constrain performance on alternate hosts. We also found several examples of cost-free generalism, with multiple strains matching or exceeding performance of the locally adapted strain in that strain’s host-of-isolation. Together, our results demonstrate that cost-free generalism is relatively common, and that local adaptation does not guarantee local dominance or host exclusivity. This indicates a more complicated reality than the strict co-occurrence of local adaptation and local dominance, and a fundamental disconnect between a strain’s host-of-isolation and its in planta performance.

Our findings support two key conclusions. First, they reinforce the growing consensus that taxonomic systems based on host-of-isolation are fundamentally flawed. The mismatch between P. syringae pathovar designations and pathogenic phenotypes is well-documented [51, 53, 62], and unsurprising given the organism’s ecology. A true opportunist, P. syringae does not occupy a single, well-defined niche [153], but thrives across diverse host and non-host environments. This ecological flexibility contributes to its high intraspecific phenotypic and genetic diversity [56, 59, 154–156] and likely to its long-term success as a species. We found, as have others, that in vitro phenotyping is also insufficient; it neither reliably distinguishes strains nor predicts in planta performance. While phenotyping will remain valuable for high-priority strains [157], it lacks the consistency and specificity needed for taxonomy, and is furthermore impractical given the increasing number of new bacterial taxa. For these reasons, we agree with calls to exclude phenotype from bacterial classification altogether [41, 158, 159]. Second, there is a pressing need to better understand the spatial scales at which microbial populations diverge. Although local adaptation is a well-documented and important feature of microbial evolution, applying this knowledge to predict pathogen adaptation or to model epidemic outbreaks remains challenging [61, 85, 160], largely because population boundaries remain unclear [35, 153]. Local adaptation varies based on spatial structure, and populations can differentiate at a range of scales [161, 162]. Better estimates of these scales are critical for improving our ability to predict and manage disease dynamics.

A9 emerged as a particularly important pathogen in our study, showing both local adaptation and local dominance while maintaining high growth rates across diverse hosts. DC3000 also performed well, but is arguably less ecologically relevant as it is more distantly related than is A9 to contemporary field isolates. A9’s higher overall r and DC3000’s higher K suggest that these close relatives have evolved distinct in planta strategies (rapid colonization vs. long-term persistence). Notably, in vitro phenotypes were not predictive of in planta phenotypes, and in fact, the patterns were largely reversed, with A9 growing at the slowest r and DC3000 at the fastest (Fig. 2). The tomato pathovars had the larger T3E repertoires in our system, suggesting that the potential risk of effector recognition is outweighed by the benefits of maintaining generalist capacity. Race 1 pathovars including A9, which demonstrate atypically strong virulence in wild tomato [87], lack the T3E avrPto, instead using alternative infection mechanisms that may include the unique T3Es, singleton histidine kinases, and ubiquitin protein we identified. Recently a gene, Ptr1, conferring resistance to race 1 pathovars was discovered in wild tomato [163, 164].

NP29 showed the strongest evidence of local adaptation in our system, with significant effects for both r and K metrics. It also outperformed the other strains in non-host conditions, reaching significantly higher K (Fig. 2), possibly reflecting a metabolic advantage associated with maintaining the smallest genome. Previously, Kniskern et al. [80] found that P. syringae isolated from A. thaliana more frequently elicited resistance responses than isolates from agricultural hosts. They suggested that these strains were either generally maladapted to all hosts, or specifically to A. thaliana (which would indicate that A. thaliana is winning the arms race against P. syringae). In contrast, our results demonstrate that NP29 is well-adapted to A. thaliana, where it performed significantly better than in any other host. However, since NP29 did not perform significantly worse across hosts than other strains, we cannot conclude that it is broadly maladapted.

Strains infecting natural hosts likely encounter diffuse selection in multispecies communities and persist in non-host niches. Supporting this, phylogroup 2 strains (like NP29) are strong epiphytes [165], and have been hypothesized to carry genes conferring stress tolerance [166]. Our findings are consistent with this hypothesis, as NP29’s singleton genes were enriched for stress response functions. Phylogroup 2 strains also show a negative correlation between phytotoxin genes and T3E count [58, 119, 167, 168], suggesting a distinct adaptive trajectory compared to other phylogroups. Many A. thaliana isolates fall within phylogroup 2 [80, 92], and the presence of phytotoxins alongside the absence of T3Es may reflect specific adaptation to this host. As a ruderal plant commonly found near agricultural fields, A. thaliana is likely exposed to crop-adapted P. syringae and their T3Es over evolutionary timescales, potentially selecting against many circulating T3Es. Supporting this, the syringomycin/syringopeptin pathogenicity island carried by NP29 is conserved in many A. thaliana isolates [92]. It is possible that the adaptive strategy is not specific to A. thaliana, but broadly effective in relatively underexplored hosts. Grenz et al. [169], for instance, found that phylogroup 2 isolates were more virulent than others in the natural hosts fern and liverwort. Taken together, the diversity of adaptive strategies with distinct underlying molecular mechanisms underscores the ecological breadth and adaptability of this species.

One limitation of our study is that we included only a single cultivar per host (except for tomato, which was represented twice). This decision was necessary for practical reasons, and we selected cultivars we believed to be representative. However, cultivar-specific differences may influence the outcomes of in planta assays, potentially affecting the repeatability of our results. Another important caveat to our approach, in which we compared relative performance across hosts to detect potential costs of adaptation, is that such tradeoffs do not necessarily reflect ‘true’ pleiotropic costs. Our method was limited to surveying broad performance and suggesting underlying mechanisms, rather than directly linking genotype to phenotype. This approach may also underestimate tradeoffs that have been obscured by subsequent fixation of compensatory mutations. For example, while our design does not preclude bacterial adaptation to laboratory conditions, such adaptations would be more pronounced in older isolates. Since we detected less local adaptation in older isolates, laboratory adaptations must have acted to reduce, rather than exaggerate, the magnitude of tradeoffs. Consequently, the effects we report represent conservative estimates. Selection is strong when bacterial populations are large, allowing rapid compensation for accrued fitness costs.

Another caveat is that our proxies for fitness, r and K, did not always yield concordant results. This is consistent with previous findings that these metrics are not necessarily correlated [170–172]. We included both to capture a more biologically meaningful picture of pathogen fitness. r reflects how fast a pathogen grows, which can be crucial to outpacing a host immune response when establishing an infection, whereas K estimates the maximum population size that a host can support, which is more reflective of long-term immune evasion. In theory, r is favored under low-density, stochastic conditions, and K is favored under high-density, resource-limited conditions, creating a tradeoff between speed and size [173]. We did find that r and K tended to be inversely correlated in our in vitro experiments (Fig. 2), but this was not the case for our in planta experiments (Fig. 3), likely due to the more complex and heterogeneous host environments.

The prevailing view that local adaptation should be common in nature is generalized by the Baas-Becking hypothesis, i.e., “Everything is everywhere but the environment selects” [174]. For microbes, which adapt quickly, disperse frequently, and can have global distributions, this seems especially plausible. Yet, studies in both natural and laboratory-evolved microbial populations reveal a more nuanced reality than this hypothesis predicts. Rather than a universal pattern of local adaptation and restrictive performance tradeoffs, microbial species often exhibit a mix of local adaptation and broad, cost-free generalism. Our findings reinforce this pattern in P. syringae, highlighting the species’ remarkable adaptability.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

RS and JB conceived of the project, RS completed the experiments and wrote the manuscript, and JB reviewed the manuscript.

Funding

This work was funded by the University of Chicago Department of Ecology and Evolution Hutchinson Botany Fund to RS and a National Science Foundation award to JB (NSF MCB 0603515).

Data Availability

Sequencing reads are deposited at NCBI under BioProject accession PRJNA1335816. Scripts and growth data are currently available on the lead author’s GitHub at the following URL: [https://github.com/rebeccasatterwhite/Psyringae-pathovar-adaptation](https:/github.com/rebeccasatterwhite/Psyringae-pathovar-adaptation) and will be archived on Dryad prior to publication.

Declarations

Competing interests

The authors declare no competing interests.