Twenty Rhodococcus isolates were selected for genome sequencing

To quantify the virulence of the 20 selected isolates, we measured their effects on root growth of Nicotiana benthamiana seedlings [13]. First, we used wild type D188 and key mutants previously shown to be non-pathogenic or compromised in virulence to standardize the root inhibition assay (Figs. 2A and 2B). Of the 20 isolates, 15 isolates significantly inhibited root growth of N. benthamiana (Fig. 2C). The isolates we selected represent multiple clades of Rhodococcus (Creason and Chang, data not shown). Therefore, isolates, representing the genetic diversity of the sequenced samples, were tested and demonstrated to cause leafy galls to N. benthamiana (Fig. 2D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Sequenced isolates of Rhodococcus vary in phytopathogenicity.

(A) Inhibition of N. benthamiana root growth by phytopathogenic Rhodococcus. Seedlings were independently inoculated with D188, its genetic variants, or 10 mM MgCl₂ buffer (mock). Photographs were taken at 7 days after inoculation; scale bar = 0.25 cm. (B and C) Average root growth of N. benthamiana seedlings infected with D188 and its genetic variants (B) or with Rhodococcus isolates (C). Root lengths were measured (cm) and averaged. Error bars indicate standard error of the mean (SEM); *significant (p-value threshold ≤0.01). All treatments were repeated at least three times with similar results. (D) Isolates of Rhodococcus cause leafy gall disease. N. benthamiana was infected with the ΔpFiD188 strain of D188, mock, or members that represent the diversity of the clade. The black scale bar = 1 cm.

https://doi.org/10.1371/journal.pone.0101996.g002

A minimum of 17.5 million reads was generated for each genome sequence (Table 1). Isolates A44a and D188 were selected as references and were more deeply sequenced using hybrid approaches. The reads were processed and independently de novo assembled for each genome. The number of scaffolds ranged from 9–50. Isolate A44a had the fewest scaffolds, which we attribute to the use of mate pair sequencing. The A44a assembly had two scaffolds that corresponded to its linear and circular plasmids and seven for the chromosome. Based on the finished genomes of R. jostii RHA1 and R. equi, we infer that A44a has four rRNA-encoding loci that could account for three of the physical gaps in the sequence. The use of 454 reads helped reduce the number of scaffolds for D188 when assembled using Illumina reads alone, but relative to other assemblies, was not effective in dramatically improving the quality of the assemblies. All 20 sequenced isolates have a high GC% of around 64%. The average genome size and number of coding sequences (CDSs) for all 20 isolates was estimated at 5.8 Mb and 5,475, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary statistics of Rhodococcus draft genome assemblies.

https://doi.org/10.1371/journal.pone.0101996.t001

A pFiD188-like plasmid is present in most, but not all, phytopathogenic isolates of Rhodococcus

The linear plasmid, pFiD188 is necessary for the virulence of R. fascians D188 [14]. We therefore used the sequence of pFiD188 as a query to search the genome assemblies to examine its necessity for Rhodococcus virulence [19]. CDSs homologous to those along the entire length of pFiD188 are present in 13 of 15 genome sequences of phytopathogenic isolates (Fig. 3A). The sequences of att, fasR, and fas are conserved (85% of the 204 CDSs in this cluster of virulence CDSs are identical and the remaining 15% are >99% identical to corresponding loci of the pFiD188 sequence generated in this study; Fig. 3B). However, within these loci, we identified 24 sequence discrepancies that differed in comparison to the previously reported pFiD188 sequence [19]. Because all linear plasmid sequences from this study were identical at these positions, we concluded the published sequence had errors. Of the 18 affecting the translated sequences of CDSs, 17 resulted in single amino acid changes. One was an insertion of a guanine residue in fasR, relative to the published sequence, that leads to substantially longer translated sequences because of an upstream, in-frame ATG start codon.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. The att, fasR, and fas virulence loci are variable in organization among phytopathogenic isolates of Rhodococcus.

(A) GenomeRing of the 13 pFiD188-like sequences. Colored lines, corresponding to each isolate, indicate presence of a block. Transparent lines skip over absent blocks and connect co-linear blocks. A minimum block size of 1 kb was used. The inner portion defines conserved (R) and unique (U) regions of pFiD188 from isolate D188, as previously reported. The att-fas virulence loci and the NRPS-encoding region are also indicated by dotted regions. (B) Homology and co-linearity of the pFi_55-pFi_84 regions in the 15 phytopathogenic isolates. Top: structure of the region of pFiD188 from pFi_55 to pFi_84, with arrows depicting CDSs and the direction indicating the strand in which they are encoded. For isolates with a linear plasmid (Other 12), A25f, and A21d2, bars indicate the presence and range of sequence similarity. Relevant locus ID numbers are shown without RF25f_ and RFA21d2_ tags. The two cyan bars connected by a single line represent a gene fusion. (C) PCR-based detection of virulence loci (fasD) and pFiD188-like (pFi-013) sequences. The 16s rDNA gene was used as a positive control. Products were resolved on a 1%, 1X TAE agarose gel. Estimated product sizes (kb) are listed along the side.

https://doi.org/10.1371/journal.pone.0101996.g003

A linear plasmid sequence is absent in two other pathogenic isolates, A25f and A21d2, and the five non-pathogenic isolates. PCR for a CDS located in the R1 regions of the pFiD188-like plasmids and implicated in their maintenance confirmed in silico predictions (Fig. 3C). Therefore, the linear plasmid is correlated with phytopathogenicity, but is not strictly necessary.

Between ∼40%–80% of the CDSs associated with the 13 linear plasmid sequences were identified as being acquired by HGT. Most are located in or at the border of the three “U” regions that were previously classified based on regions within pFiD188 of D188 that are unique in sequence relative to linear plasmids of other members of Rhodococcus [19]. Additionally, the gene gain/loss polymorphisms largely corresponded to the U regions and also correlated with evidence for HGT (Fig. 3A; Table S1). The pattern of insertions and deletions in the NRPS-encoding CDS of the U2 region is likely a reflection of limitations in assembling short reads exacerbated by the modularity of NRPS genes. The predicted functions for the polymorphic genes in the U regions did not provide strong evidence for a role in pathogen virulence (Table S1). Of the four conserved R1, R2, R3, and R6 regions, previously defined based on sequence similarities to other linear plasmids of Rhodococcus, there were fewer gene gain/loss polymorphisms but R3 and to some degree, R1 had evidence for HGT. In summary, results from the analysis of 13 linear plasmid sequences were consistent with previous reports in regards to the presence of conserved and unique regions, with higher incidences of recombination in the unique regions [19].

The contribution of horizontal gene transfer to virulence evolution is limited in its scale

Virulence genes of plant pathogens can be located in islands within the chromosome. We therefore searched the genomes for regions with signatures of HGT [26]. The average size for the identified regions was 11.5 kb (Fig. 4A). The scale of HGT was variable between isolates, with no clear correlation to phytopathogenicity (Fig. 4B). In phytopathogenic isolate LMG3605 for example, only 240 CDSs were identified within regions acquired by HGT and their total size represented just 5.5% of the genome. In contrast, non-pathogenic isolate GIC36 has 78 regions with ∼800 CDSs representing 16.2% of the total size of the genome as potentially acquired by HGT. Of all identified regions, some of the outliers were substantial in size, highlighted by the ∼107.5 kb region, encoding 103 CDSs, in 05-339-1 (Fig. 4A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. The scale of horizontal gene transfer varies among isolates.

(A) Box-and-Whisker plots of regions acquired by HGT as a factor of size. Regions putatively acquired via HGT were identified using Alien Hunter, with minimum criteria of size ≥5 kb and minimum score ranging from 10.297 to 16.047. Bottom and top of the boxes indicate the first and third quartiles, respectively, with the yellow bar indicating the median. Whiskers delimit the lowest and highest data within 1.5 interquartile ranges of the lowest and highest quartiles, respectively. Red circles represent outliers. Pathogenic isolates are underlined. (B) Summary statistics of HGT for the sequenced isolates. Pathogenic isolates are underlined. (C) Histogram presenting the number of groups of CDSs as a factor of group size. All CDSs identified in regions putatively acquired by HGT were compared using BLAST analysis to determine groups of homologs. The numbers of groups of CDSs (y-axis) were enumerated and plotted according to the size of the groups (x-axis). Analysis was done for only the 15 phytopathogenic isolates.

https://doi.org/10.1371/journal.pone.0101996.g004

We compared the non-redundant set of 6,867 CDSs present in the horizontally acquired regions from all 20 isolates to a custom database consisting of CDSs gathered from non-phytopathogenic Actinomycetes. Approximately 33% of the translated Rhodococcus CDSs did not have a BLASTP hit and were thus considered specific to this group of Rhodococcus isolates. The only two significantly enriched clusters of orthologous genes (COGs) categories were “General functional prediction only” and “Function unknown”, as would be expected for a group of relatively poorly studied bacteria. Of the CDSs horizontally acquired by only the phytopathogenic isolates, the significantly enriched functional categories related to energy, metabolism, and general functions (Fig. S1A). A parallel analysis using Gene Ontology (GO) terms yielded similar results, with general functions in metabolism, transport, as well as transcription and translation being the more prominently identified terms (Fig. S1B). The few clusters that are likely associated with virulence are cytokinin and isoprenoid biosynthesis and metabolic processes.

Next, we reasoned that horizontally acquired candidate virulence genes should be conserved across the majority of the phytopathogenic isolates. An overwhelming majority of the CDSs associated with regions putatively acquired by HGT were present in only one or two of the phytopathogenic isolates (Fig. 4C). Even when we lowered our criteria and considered CDS that had homologs in approximately one-half of the phytopathogenic isolates, fewer than 100 homologous families were identified. Each of the families had homologs present on pFiD188. Overall, data suggest that other than the linear plasmid, HGT does not play a large-scale role in virulence evolution of phytopathogenic Rhodococcus and likely contributes more to the outside-host lifestyle of the bacteria.

Few genes are conserved and unique to phytopathogenic isolates

We hypothesized that, regardless of phylogenetic structure and HGT, the phytopathogenic isolates can be expected to have a core and exclusive set of CDS that distinguish them as pathogenic. We compared CDSs from 24 genome sequences, including the 20 from this study as well as Rhodococcus spp. AW25M09, JG-3, 29MFTsu3.1, and 114MTsu3.1 ([27]; GenBank BioProjects PRJNA200424, PRJNA195882, and PRJNA201196). The four additional isolates were identified from various environments, clustered with the phytopathogenic Rhodococcus, and lack loci known to be necessary for Rhodococcus phytopathogenicity (Creason and Chang, data not shown). Overall, we identified a “core” of 2,870 and a “pan” of 16,733 CDSs (Fig. 5). The COG categories in the “flexible” genome (13,863 CDSs) were enriched in categories related to energy and responding to the environment and its co-inhabitants (Fig. S2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. The pathogenic isolates have a small set of core coding sequences.

A total of 16,733-redundant CDSs are present in the “pan” genome of the 24 isolates of Rhodococcus that were examined. Based on reciprocal best-hit BLASTP analysis of the translated sequences, 2,870 are core. Coding sequences core to pathogenic (red) and non-pathogenic (blue) genomes were similarly identified for the 15 and 9 isolates, respectively.

https://doi.org/10.1371/journal.pone.0101996.g005

Because we were interested in identifying CDSs that are exclusive to pathogens, we compared pan and core genomes of the 15 phytopathogenic isolates to the pan and core genomes of the 9 non-pathogenic isolates (Fig. 5). The 15 pathogenic isolates have a core of 234 CDSs that were not also identified as core to the nine non-pathogenic isolates. However, only 12 of the 234 were classified as exclusive based on their complete absence from the nine non-pathogenic isolates; the other 222 CDSs are part of the flexible genome of the non-pathogenic isolates. The 12 exclusive CDSs correspond to the att locus (10 CDSs), fasR, and a putative FAD-binding monooxygenase-encoding gene (pFi_057 of pFiD188), all of which have members on pFiD188. The reciprocal best hit for the FAD-binding monooxygenase-encoding gene of A21d2 barely exceeded threshold, with an identity of only 35.2% suggesting it may not be a bona fide member of the gene family, as each isolate has large numbers of monooxygenase-encoding genes (Davis and Chang, data not shown). Most notably, the six fas CDSs were not identified as core to the pathogenic isolates because of the absence of fas from isolate A21d2. PCRs using oligonucleotides specific to each CDSs of the fas operon were negative (Figs. 3B and 3C; Table S1).

The virulence loci of A25f and A21d2 are novel in structure and sequence

We used the pFiD188 sequence and reciprocal best-hit analysis to identify scaffolds of ∼190 kb and 18 kb in the A25f and A21d2 assemblies, respectively (Table S2). In A25f, 30 of the 33 best hit CDSs (RFA25f_04806-RFA25f_04843) are co-linear to their homologs on pFiD188, starting from nine CDSs upstream of the att locus and extending to two CDSs past the fas locus (Fig. 3B). The highest levels of sequence similarities were observed for att through fas (>95% identity; AttR = 94.5%). Despite that, the degree of sequence similarity was lower relative to the degree of similarity found among the 13 isolates that carry a linear plasmid like pFiD188. Analysis of flanking sequences did not provide any clues to the location of this scaffold in the genome because there were no regions of co-linearity with the D188 assembly. However, the scaffold lacked features typically associated with plasmids and the most parsimonious explanation is that a recombination event occurred between a portion of a pFiD188-like plasmid and the chromosome of A25f. Consistent with previously reported results, A25f delineates a small portion of the linear plasmid as necessary for phytopathogenicity [19].

In A21d2, only 11 of the 28 best hit CDSs in the 18 kb scaffold are co-linear to the U1 region, starting at RFA21d2_02316 (attR) and unexpectedly and abruptly ending after RFA21d2_02306, an AraC-like encoding CDS (fasR; Fig. 3B). We used thermal asymmetric interlaced (TAIL) PCR to determine flanking sequences. The resultant sequences were nearly 1 kb long, partially similar in sequence to each other, and mapped to separate single locations that coincided to a region between RFA21d2_03323 and RFA21d2_03324 within a 169 kb long scaffold, node_3. Consistent with previous searches, no fas operon was present in node_3 and no functional replacement of the Fas proteins could be predicted from the translated sequences of node_3. The A21d2 homolog of pFi_057 was not found in this block of co-linear CDSs and is instead located over 1 Mb away.

The fas operon of A21d2 is replaced by a gene fusion encoding IPT and LOG domains

The absence of the fas operon from A21d2 was not expected. The locus is replaced by a novel cluster of three overlapping CDSs (Fig. 6A). The outer two CDSs, RFA21d2_02305 and _02303, are annotated as containing a single Radical SAM domain and an Abhydrolase_6 domain, respectively. Neither of the domains are present in the Fas proteins of pFiD188 and the single Radical SAM domain is different from those predicted for the translated sequences of mtr1 and the homologous mtr2, both of which are located between fasR and the fas operon in pFiD188 [19]. RFA21d2_02305 and _02303 were not expressed in cells grown in any of the three media tested and not predicted to be involved in cytokinin metabolism (Fig. 6A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. RFA21d2_02304 is a gene fusion unique to A21d2.

(A) Structure and predicted functional domains for RFA21d2_02303-02305. Functional domains were identified using PFAM (PFxxxx) and mapped to each of the sequences. Sequence logos were generated and shown for the isopentyltransferase and phosphoribohydrolase domains. The best matching residues of RFA21d2_02304 are shown and mapped to their corresponding locations in the diagram. Amino acids are colored according to class. The “*” highlights conserved residues; the single “R” at positions 166 and 378 are not part of the sequence logos but are included to show their conservation. Small arrows show the locations of the binding sequences for oligonucleotides used in RT-PCR (see inset: L = LB; I = minimal media+leafy gall extracts; M = minimal media). RT-PCRs with oligonucleotides that span both overlapping regions were negative and are not shown. (B) Unrooted phylogenetic tree for IPTs. Scale bar = number of amino acid substitutions per site. (C) Unrooted phylogenetic tree for phosphoribohydrolases. Scale bar = number of amino acid substitutions per site. One branch was split for sizing purposes.

https://doi.org/10.1371/journal.pone.0101996.g006

Remarkably, RFA21d2_02304 is a chimera between fasD- and fasF-like sequences and the possibility of an assembly mistake was dismissed based on PCR and Sanger sequencing that confirmed the sequence of the locus (Fig. 6A; Table S2). Both IPT and cytokinin phosphoribohydrolase domains are present in RFA21d2_02304. Three residues necessary for IPT function and conserved in all IPT sequences previously examined by others are present in the N-terminal portion of RFA21d2_02304 [28]. There are also three conserved and predicted catalytic residues in the C-terminal portion of RFA21d2_02304 that are found in similar positions relative to other bona fide LOG proteins [29]. Unlike fasD of D188 and the CDSs flanking RFA21d2_02304, the chimera is expressed under what is considered non-inducing conditions (Fig. 6A; Fig. S3). These results indicated the three CDSs are expressed as independent monocistronic messages.

The two domains of RFA21d2_02304 formed branches distinct from the 14 other FasD and FasF homologs (Figs. 6B and 6C). The IPT region is more similar to FasD than the LOG region is to FasF, and as previously shown, IPTs from phytopathogenic Rhodococcus cluster separately from those of plants and other bacteria [21], [30]. The exceptions are IPTs of Streptomyces spp., which also encode the fas operon [31]. The LOG region of RFA21d2_02304 and FasF are more similar to LOG of plants than to either of the two homologs present in Rhodococcus. In all, the data suggested that RFA21d2_02304 was acquired horizontally from another source, as opposed to being derived from a series of recombination events within a pFiD188-like plasmid.

Isopentenyladenine is the only cytokinin that specifically accumulates in extracts of D188 grown in culture and under inducing conditions

The absence of fasA from A21d2 is not consistent with the model that predicts the necessity for a mixture of cytokinin types (isopentenyladenine (iP), tZ, ciz-zeatin (cZ), and the methylthio derivatives of the latter two) for virulence [18], [20]. The 20 Rhodococcus isolates encode homologs of all the enzymes necessary for the methylerythritol phosphate (MEP) pathway, including a homolog of isopentenyl diphosphate (IPP) isomerase (IDI1) that catalyzes the isomerization of IPP to dimethylallyl diphosphate (DMADP; Fig. 1; Table S3). The MEP pathway synthesizes DMADP and the intermediate, (E)-4-hydroxy-3-methyl-but-2-enyl diphosphate (HMBDP), both of which are isoprenoid side chain donors that can be used by IPTs to synthesize the isopentenyladenine ribotide (iPAMP) and tZ ribotide (tZRMP; [32]–[35]). In contrast, homologs for most of the enzymes of the mevalonate (MVA) pathway that synthesizes DMADP were not identified (Table S3).

Because DMADP and HMBDP are potentially available, it is conceivable that the activities of FasD and FasF could enable the synthesis of both iP and tZ by Rhodococcus in culture. To test this, we profiled cytokinins from genetic variants of D188 grown in minimal media supplemented with extracts of leafy galls, uninfected plants, or H₂O as a mock treatment. Of the 32 cytokinins that were profiled, only eight could be detected: iP, cZ, 2-methylthio-cis-zeatin (2MeScZ), their ribosides, as well as iPAMP and tZRMP; tZ was not detected (Table S4). iP was the only active cytokinin type that significantly accumulated specifically in preparations from bacteria grown in leafy gall extracts and in a fasD/pFiD188-dependent manner (Fig. 7). In contrast, none of the detectable Z-type cytokinins accumulated to significant levels in a treatment-dependent manner. All detectable cytokinin types started to accumulate at later time points in the ΔfasD mutant, independent of treatment (Fig. S4). Expression profiles of attE and fasD showed induced expression specific to bacterial cells grown in extracts of leafy galls, as previously reported (Fig. S2; [16]. Maximum expression was observed at 6 hours (10,000X and 100X more expression for attE and fasD, respectively, relative to expression in cells grown in extracts of uninfected plants) and rapidly declined to basal levels by 12 hours after growth. Data indicated that phytopathogenic Rhodococcus synthesizes only one active type of cytokinin in a fasD-and leafy gall-dependent manner.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. Isopentenyladenine is the only active cytokinin that accumulates in a fasD and leafy gall extract-dependent manner.

Wild type D188 (blue), ΔfasD (green) and ΔpFiD188 (red), were grown in media augmented with extracts from leafy galls, uninfected plants, and water as a mock. Of 32 cytokinin types that were profiled, only eight were detected and the six most abundant types are shown. Change in concentration (y-axis) is presented for only the first four time points (x-axis). Colored “*” indicates significant difference in cytokinin accumulation in corresponding mutant genotype relative to wild type and corresponding time point (p-value threshold = 0.05). Experiments were repeated twice with similar results.

https://doi.org/10.1371/journal.pone.0101996.g007

The fasR CDSs of A25f and A21d2 are substantially more polymorphic than other family members

Putative AraC-like encoding CDSs are located between the att operon and fas operon or fas-like CDSs in both A25f and A21d2 (RFA25f_04127 and RFA21d2_02306, respectively; Fig. 3). RFA25f_04127 and RFA21d2_02306 are predicted to be functional and clustered with fasR in a phylogenetic tree, supporting their membership to this family (bootstrap percentage of 100%; Fig. S5). However, the sequences of fasR are highly polymorphic (Fig. 8A). The translated sequence of the A25f CDS shares 97% identity with FasR of D188. In RFA21d2_02306, there are 343 SNPs (32% of all possible sites), of which 89, or 26% of the SNPs, are non-synonymous substitutions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 8. RFA25f_04833 and RFA21d2_02306 are polymorphic in sequence.

(A) Predicted structure of the 348 amino acid FasR sequence. Functional domains are indicated along the top with arrows and barrels representing predicted β-sheets and α-helices, respectively. Boxed region = a region of fasR from D188 that is absent from the reported sequence. Synonymous (red) and non-synonymous substitutions (blue) and INDELs (green arrow heads with number of nucleotide differences indicated) are plotted according to their location in the sequences. Clusters of non-synonymous substitutions in fasR of A21d2 are denoted with, “*”. Substitutions for the other 12 FasR sequences (Other 12) represent the sum total of all sites with substitutions. (B) Sliding-window analysis of synonymous and non-synonymous substitutions in fasR of A21d2 paired with fasR of D188. (C) Sliding-window analysis of codon adaptation index in fasR of A21d2 and D188. The x-axis is based on coordinates of the translated sequence that are aligned to predicted structure shown in panel A.

https://doi.org/10.1371/journal.pone.0101996.g008

RFA21d2_02306 is nevertheless predicted to be under purifying selection (K_a/K_s = 0.327; p-value = 9.3×10⁻²⁰; paired with allele from D188). RFA21d2_02306 is also not optimized for translation (CAI = 0.221; 59 ribosomal protein encoding genes = 0.667; genome average = 0.530). This was not surprising since fasR includes an unusually high number of rare TTA leucine codons [16]. In attempts to model the patterns of substitutions, we mapped them onto the predicted secondary structure of FasR (Fig. 8A). Both domains of the predicted structure are similar to other AraC-like regulators [36]–[39]. One region within the putative ligand binding domain was identified as having a high K_a/K_s and was significant in a permutation test (p-value = 0.033; Fig. 8B; [40]). The K_s values started trending upward thereafter within the putative DNA binding domain and culminated in a region with very high values (p-value = 0.04). This region overlaps two helix-turn-helix motifs predicted to be involved in binding DNA, based on alignments to AraC. In addition, a larger portion of this region correlated with above average CAI values in RFA21d2_02306 relative to FasR of D188 (Fig. 8C). Therefore, despite evidence for purifying selection and a high density of synonymous substitutions in the putative DNA binding domain, diversifying selection was identified in a putative ligand binding region.

Rhodococcus requires four horizontally acquired functions to be phytopathogenic

Virulence evolution of phytopathogenic Rhodococcus species can be modeled by the co-option mechanism proposed for R. equi [10], [24]. It has been previously reported that the chromosomal locus vicA, which encodes malate synthase that functions in the glyoxylate shunt pathway of the Krebs cycle, is necessary for full symptom development [41]. Additionally, the predicted dependency of the Fas proteins on the MEP pathway for the necessary substrate for cytokinin biosynthesis is another example of co-option (Fig. 1). Few other loci could be identified that had evidence for HGT and implicated in virulence. Only a miniscule fraction of the CDSs associated with regions with signatures of HGT were found conserved in pathogenic isolates and all of these CDSs have homologs encoded on a pFiD188-like plasmid (Fig. 4). Analysis of functional categories of CDSs associated with regions acquired via HGT also failed to support HGT as having a large-scale contribution in virulence evolution (Fig. S1). Similar findings were described for the COG categories enriched in the “flexible” genome (Fig. S2). The results are consistent with the linear plasmid being the key acquisition for virulence and the majority of horizontally acquired functions contributing to the adaptation of Rhodococcus to their outside-host environments. Indeed, R. fascians is considered a soil bacterium and closely related isolates have been isolated from extreme environments, suggesting an ability of these organisms to use a wide range of substrates [42]–[44].

The structures and sequences of virulence loci of A25f and A21d2 are novel

We suggest that the horizontal acquisition of just four functions involved in secondary metabolism (att), gene transcription (fasR), cytokinin biosynthesis (fasD), and cytokinin activation (fasF) is sufficient for Rhodococcus to infect plants. Isolates A25f and A21d2 are unique among the sequenced samples. In A25f a recombination event likely occurred that introduced a block of CDSs from a linear plasmid into the chromosome. Gene fusions are generally predicted to occur by insertions and deletions between juxtaposed genes, such as those within an operon, and disseminated via HGT [45], [46]. However, for A21d2 a fusion would have had to occur between two CDSs separated by fasE. Stronger evidence against a direct fusion event comes from the observation that the IPT and LOG domains of RFA21d2_02304 formed branches clearly distinct from their homologs and that RFA21d2_02304 overlaps two CDSs absent from other sequenced isolates of Rhodococcus (Fig. 6). We therefore conclude that RFA21d2_02304 was acquired from another organism in what could be a non-orthologous displacement of the fas operon.

Integration of virulence loci into the chromosome is expected to confer a greater state of stability. However, the rarity of the virulence loci of A25f and A21d2 among the sequenced isolates is not consistent with this expectation. A potential explanation is that because the pFiD188-like linear plasmids are conjugative, the mobility of these vehicles may compensate for any lowered stability by enabling rapid dissemination throughout the population. This is highly plausible especially considering that acquisition of the four functions, most frequently vectored by a plasmid, is sufficient for genetically divergent Rhodococcus isolates to be phytopathogenic (Fig. 2). Alternatively, the events that led to the changes in A25f and A21d2 could have occurred relatively recently or the 20 isolates that were sequenced do not adequately reflect the diversity of the population. We also cannot ignore the likelihood for trade-offs that occur in non-obligatory pathogens that persist in both within and outside-host environments.

The att locus is conserved in the phytopathogenic Rhodococcus isolates

The att locus is the most conserved in sequence and structure across all phytopathogenic isolates examined and consistently clustered with the fas locus (Fig. 3). Its product is also the only known target of host resistance [47]. Together, these data suggest the product of att is an important virulence determinant in Rhodococcus despite the attenuated phenotype of the Δatt mutant [17]. The infection methods used in the laboratory may possibly circumvent att dependence, since the Δatt mutant can cause disease if the host is wounded prior to infection. It has been suggested that att is important for the early transition of R. fascians from epiphyte to endophyte and its ability to ingress into host tissue [17]. Indeed, attE expression peaks to extremely high levels within six hours of growth in inducing media and rapidly decays thereafter (Fig. S3). Seemingly counter to the importance of att is its absence from Streptomyces turgidiscabies, which does have the fas operon and can form galls on plants ([31]; PRJNA185785). However, att may not be necessary because S. turgidiscabies does not ingress but rather, it primarily enters its host through wounds and natural openings.

A new model for Rhodococcus virulence

The data generally supported the model identifying pFiD188 as necessary for R. fascians and the role of cytokinins in the pathology of plant-associated Rhodococcus. However, several lines of evidence presented in this study are consistent with a mechanism of virulence where only the iP cytokinin type, not a mixture of cytokinins, is synthesized and necessary for Rhodococcus phytopathogenicity (Figs. 1, 3, 6, and 7). We and others could not detect tZ and/or demonstrate its FasD-dependent synthesis in culture (Fig. 7; [18], [48]). The other two free base types that could be detected, cZ and 2MeScZ, accumulated independently of fas gene induction, similar to previous reports (Fig. 7; Fig. S4; [20], [21], [49]–[53]). The generation of tZ from cZ by the action of a cis-trans isomerase is not well substantiated, so it is unlikely that this could be a source of tZ [21].

Evidence also argues against FasD being able to directly produce significant levels of tZ even though bacterial IPTs can use HMBDP as a prenyl donor without the activity of a P450 monooxygenase [28], [32]. The Tzs IPT of A. tumefaciens has a hydrophilic region in the substrate-binding cavity that contributes to its ability to use both DMADP and HMBDP as substrates [28]. A substitution mutation that affected the hydrophilic region increased the specificity of A. tumefaciens Tzs for DMADP by ∼100-fold [28]. FasD of D188 has an alanine at a position that corresponds to one of the key hydrophilic residues of Tzs and the K_cat/K_m value for FasD is 10X higher when DMADP is provided as a substrate, relative to HMBDP [18]. Therefore, while it is possible that FasD could use HMBDP to generate Tz, data suggests otherwise. Additionally, the novel gene chimera of A21d2 indicates that only the IPT and LOG functions are necessary for virulence (Figs. 3 and 6). The IPT domain of RFA21d2_02304 also has amino acid substitutions that affect the predicted hydrophilic region that contribute to the increased specificity of Tzs for HMBDP [28]. RFA21d2_02304 is thus not predicted to efficiently synthesize tZ.

The original model for R. fascians phytopathogenicity is predicated on the necessity of fasA for virulence [15]. However, the original fasA mutant was generated via insertion of a plasmid and the sufficiency of fasA in complementing the mutant was not tested, opening the possibility for polar effects. In addition, FasA has not been confirmed as having P450 monooxygenase activity and the absence of detectable levels of tZ indicate fasA may be dispensable. FasB and FasC, like deoxyxylulose 5-phosphate (DXP) synthase, the first enzyme of the MEP pathway, are members of thiamine pyrophosphate enzyme superfamilies [54]. We speculate that the possible redundancy in function could explain the dispensability of fasB and fasC. The superfluousness of fasE (cytokinin oxidase/dehydrogenase) for virulence solves the conundrum of why a pathogen has an operon that encodes an enzyme with a function that diametrically opposes those that synthesize and activate cytokinins as virulence factors. There are two explanations as to why functions predicted to be dispensable are still in the population of phytopathogenic Rhodococcus clades. The enzymatic functions of the FasA-C, and -E may be unnecessary but the proteins may be required for scaffolding and forming a protein complex for efficient cytokinin biosynthesis and activation. Alternatively or additionally, the CDSs may have not been purged because membership in the fas operon could confer a level of immunity to the effects of mutation or are important in contributing to the regulation of gene expression [55].

New insights into cytokinin metabolism

The characterization of pathogen-encoded virulence proteins and molecules has provided many new insights into host growth, development, and signaling. Specifically, bacterial-encoded IPTs have contributed much to our understanding in cytokinin metabolism in plants [21]. The association of fasF with fasD in the fas operon was key to recognizing that LOG of plants is involved in cytokinin metabolism and provided evidence for the existence of the direct activation pathway of cytokinins [22]. The discovery of RFA21d2_02304 in this study raises the possibility that the phosphoribohydrolase complexes with IPT in other phytopathogenic Rhodococcus and also in plants. Gene fusions are important for increasing functional complexity through the generation of multi-domain proteins, often by uniting domains from those that physically interact and function in a common pathway. The low in vitro turnover rate of FasD led to the speculation of product inhibition [18]. The possibility of a complex with FasF indicates the potential for a direct activation of cytokinins and alleviation of inhibition. In A21d2, the fusion could be more efficient at coupling these two reactions, a possibility that will be addressed in future studies. It will also be interesting to determine if, and which, members of the IPT and LOG families form complexes in plants [56], [57].

Nucleic acid preparations, Sanger sequencing, and PCRs

Genomic DNA was extracted from cells grown directly from stocks. The Wizard Genomic DNA Purification Kit was used, according to the instructions of the manufacturer, to extract genomic DNA (Promega Corporation, Madison, WI, USA). Except for D188, isolates had undergone a limited number of passages.

Genomic DNA isolated from appropriate isolates was used as templates in PCRs with primers specific to the CDS (Table S5). Thermal Asymmetric Interlaced PCRs (TAIL) were done using nested oligonucleotides designed to Contig_21 and genomic DNA from A21d2, according to methods previously described (Table S5; [58]). Eight arbitrary degenerative primer pools were used in the amplification steps and successfully amplified products were sequenced and aligned to the A21d2 genome sequence.

For Sanger sequencing, products were treated with 2.5 U of Exonuclease I and 0.25 U of Shrimp Alkaline Phosphatase (SAP) for 40 minutes at 37°C and heat killed at 80°C for 20 minutes. Products were sequenced on an ABI 3730 DNA Analyzer in the Center for Genome Research and Biocomputing (CGRB) at Oregon State University.

Trizol Reagent (Life Technologies, Carlsbad, CA, USA) and ∼200 µl of 500 nm glass beads were added to Rhodococcus cells and disrupted using a FastPrep-24 homogenizer (MP Biomedical, Santa Ana, CA, USA) at 6.0 m/s for 1 minute. Samples were treated with DNase I (NEB, Ipswich, MA, USA), quantified and analyzed using a BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA). First-strand cDNA was synthesized using 100 ng of total RNA and the Reverse Transcription System (Promega Corporation). One µl of the reverse transcription reaction was mixed with 10 µl LightCycler 480 SYBR Green I Master, 2.0 µl primer mix (250 nM each) and water for a final volume of 20 µl. Quantitative PCR was done on a LightCycler 480 System (Roche Applied Science, Germany). Gene expression was calculated using the deltadeltaCt method relative to the expression of the 16S gene and corresponding genes in the mock treatment [59]. For RT-PCR, first-strand cDNA was synthesized from 1.0 µg total RNA using Quantas (Life Technologies, Carlsbad, CA, USA). cDNA equivalent to 50 ng of starting RNA template was used in PCRs. Samples lacking reverse transcriptase were included as controls. Oligonucleotide sequences are available in Table S5.

Next-generation sequencing, assembly, and annotation

Library construction and sequencing were done in the CGRB (Illumina GAIIx, HiSeq, and MiSeq, 454 Jr.), DNAVision SA (Illumina HiSeq of D188), or HTSF@UNC (Mate-pair sequencing of A44a).

Genomes were assembled using Velvet, with hash lengths of 75 [60]. Insert sizes were independently determined based on estimated fragment sizes from each library preparation. A44a was assembled using Velvet 0.7.55, D188 was assembled using Velvet 1.1.05, and the other 18 isolates were assembled using Velvet_1.2.08. For each genome, multiple assemblies were done, in which coverage cutoff, expected coverage, and hash length parameters were changed. The highest quality assembly was identified based on number of contigs and having a sum total size between 5–6 Mb. Contigs from isolate A44a were reordered with Mauve, using the genome sequence of Rhodococcus jostii RHA1 as a reference [61], [62]. All other assemblies were directly or indirectly re-ordered based on the genome sequence of A44a. Prokka was used to annotate genomes. Prokka identifies tRNAs, rRNAs, and CDSs using Aragaron, rnammer, and prodigal, respectively (Prokka: Prokaryotic Genome Annotation System - http://vicbioinformatics.com/). CDSs were annotated based on BLAST analysis to a database of genomes core to the Rhodococcus genus, including all whole-genome assemblies, followed by BLAST analysis to the UniProt database [63], [64]. The remaining CDSs were then used in searches against the HMM databases using HMMER3 [65].

Phylogenetic analyses

Publically available sequences were gathered from the NCBI nr, nt, or wgs databases (http://www.ncbi.nlm.nih.gov/). Lists of sequences were manually curated. Sequences were aligned using L-INS-i algorithm in MAFFT and the most appropriate models of substitution were selected using the BIC selection method implemented in ProtTest 3 [66], [67]. Trees were generated using RAxML, with the -f a setting, and 1000 bootstrap replicates, unless otherwise noted [68]. Alignments were visualized and trimmed using Belvu [69]. Gblocks was used to trim the MLSA alignments with half gapped positions allowed [70]. Images were generated using the iTOL [71]. Only bootstrap values equal to or greater than 50 were shown.

For IPT-encoding genes, sequences were gathered by using RFD188_04926 (fasD) from D188 as the query in tBLASTn searches against both the nt and wgs NCBI databases. A total of 50 and 25 translated sequences, from the nt and wgs databases respectively, all IPT-encoding genes from Arabidopsis thaliana and Oryza sativa, and fasD from the Rhodococcus strains sequenced in this study were used to construct the tree. A similar approach was used for LOG-encoding genes with the exception that RFD188_04928 (fasF), RFD188_01289 (LOG-1), and RFD188_01290 (LOG-2) were used as query sequences. The top 50 hits for each query, 28 LOG family genes in the UniProtKB database and those found in the Rhodococcus strains sequenced in this study were used to construct the tree1. A total of 308 AraC-type sequences were identified, using BLASTP analysis, from the translated CDSs of all 20 Rhodococcus isolates (exceeded an e-value threshold of 1×10⁻⁵). Seventy of the sequences were used to construct the tree.

Bioinformatic analyses

InterProScan was used to analyze the translated sequences of RFA21d2_02303-_2305 for domain analysis and generation of sequence logos [72]. Sequences from IPR005269 (LOG family, 9681 sequences) and IPR002648 (IPT domain, 144 sequences) were downloaded from the InterPro database and aligned using MAFFT–auto and L-INS-i, respectively [67], [73]. Sequences without the conserved domains were removed, leaving 9648 and 75 sequences, respectively. Logos were generated using WebLogo 3 (http://weblogo.threeplusone.com/; [74]).

For reciprocal best-hit BLAST analysis, the translated sequences of all CDSs of pFiD188 were used as queries in BLASTP searches of the A25f and A21d2 assemblies. We used the pFiD188 sequence from D188 generated in this study because of the discrepancies between it and the deposited sequence (JN093097; [19]). Top hitting homologs (p-value threshold = 1×10⁻⁵) were used as queries in reciprocal BLASTP analysis of the D188 assembly. Results were confirmed with tBLASTn searches.

The core and flexible genomes were identified using reciprocal best-hit BLASTP analysis of translated CDSs. Translated sequences were iteratively added into groups of orthologs, starting with strain D188. Threshold cut-offs of 35% identity and at least 50% gene coverage of the subject by the query were used.

Secondary structure predictions were made using jpred and visualized using jalview [75], [76]. ParaAT was used to generate alignments as input for Ka/Ks calculator [77], [78]. A 120-nucleotide window was slid along the sequence in 12 nucleotide increments for analysis of Ka/Ks and codon usage. For the permutation test, the ParaAT-aligned sequences were split based on aligned codons and the order of the codons was randomly permuted 1000 times. Values were calculated for each of the permuted sequences, using the same sliding window approach. We enumerated the number of times four (Ka and Ka/Ks) and six (Ks) consecutive windows were found in the permutations. Threshold values for window peaks were set at 1.5 * (standard deviation) from the mean for Ka and 2 * (standard deviation) from the mean for Ks and Ka/Ks, in relation to the initial aligned fasR sequences.

The SuperGenome based on the 13 pFiD188-like sequences was aligned with Mauve then generated and visualized using GenomeRing [62], [79].

Alien hunter (default settings) was used to predict regions acquired by horizontal gene transfer [26]. A custom BLAST database was generated using the translated CDSs from Streptomyces coelicolor A3, Nocardia farcinica, Corynebacterium diphtheriae NCTC 13129, Gordonia KTR9, Mycobacterium tuberculosis H37Rv, Rhodococcus jostii RHA1, Rhodococcus opacus B4, Rhodococcus equi 103S, and Rhodococcus erythropolis PR4 [24], [61], [80]–[86]. Homology was determined based on BLASTP top-hit analysis (threshold cut-offs = >50% in length and ≥35% identity).

WebMGA (default settings) and Revigo (Allowed similarity = 0.9) were used to assign genes to Cluster of Orthologous Groups and Gene Ontology (GO) categories, respectively [87], [88]. Revigo was also used to analyze GO categories. Fisher’s exact test was used to determine statistically significant differences in COG assignments (p-value≤0.01).

Graphs were generated in R [89].

Inkscape was used to design line drawings and figures (inkscape.org/).

Bacterial strains and growth conditions

The 20 isolates of Rhodococcus (Table 1) were grown in Luria-Bertani (LB) media at 28°C. When appropriate, media were amended with 30 µg/ml kanamycin.

Plant growth conditions and plant infection

For root inhibition assays, N. benthamiana was germinated on MS agar plates (half-strength MS, 0.5 M MES). Three-day-old germinated seedlings were inoculated with suspensions of Rhodococcus (OD₆₀₀ = 0.5; 10 mM MgCl₂ buffer) or mock inoculated (buffer only) and grown vertically for 1 week at room temperature with 16 hours of light. Plantlets were photographed and ImageJ was used to measure root lengths [90]. Comparisons for each Rhodococcus isolate were made relative to average root length of corresponding seedlings inoculated with ΔpFiD188 in the same experimental replicate. Tukey’s HSD test was used to determine significance.

Leafy galls were induced in 3–4-week-old N. benthamiana plants, as previously described [14]. The apical meristems of plants were pinched using forceps and inoculated with 10 µl of Rhodococcus (OD₆₀₀ = 0.5; 10 mM MgCl₂ buffer). Plants were grown under conditions described above and assessed weekly. Leafy galls were photographed at 4 weeks post-inoculation. For leafy gall extracts, 4 week-old leafy galls were excised from plants, ground in sterile water using a mortar and pestle, and filter sterilized (22 µm filter). Extracts were frozen at −80°C until use. Induction was done as previously described [91].

Cytokinin profiling

Cytokinins were extracted from R. fascians D188 and its variants and quantified, using previously published methods [91], [92]. All supernatants were spiked with deutered cytokinin standards, including [²H₆]iP, [²H₆][9R]iP, [²H₆](9G)iP, [²H₆](7G)iP, [²H₅]DHZ, [²H₅][9R]DHZ, [²H₅](9G)DHZ, [²H₅](7G)DHZ, [²H₅](OG)DHZ, [²H₅](OG)[9R]DHZ, [²H₅]2MeScZ, [²H₅]2MeScZR, [²H₆]MS-iP and [²H₆]MS-iPA (Table S5; Olchemlm Ltd., Olomouc, Czech Republic). The pH of the samples was adjusted to pH 7 and samples were purified on a DEAE-Sephadex column (2 ml, HCO₃⁻) with a RP-C18 column coupled underneath. After washing with ddH₂O, the fraction containing the cytokinin free bases, ribosides, and glucosides was eluted with 80% methanol. Eluates were dried under vacuum, diluted with PBS and immunoaffinity purified using isoprenoid cytokinin IAC columns (Olchemlm Ltd.). After elution with ice-cold 100% methanol, eluates were dried in vacuum and stored at −20°C until further analysis. Cytokinin-phosphates (retained in the DEAE column) were eluted with NH₄HCO₃ and concentrated on a RP-C18 column before elution with 80% methanol. After vacuum concentration and dissolution with 0.01M Tris-HCl (pH 9.6), the cytokinin-phosphates fraction was treated with alkaline phosphatase and further immunoaffinity purified as described above.

Cytokinins were quantified using an electrospray ACQUITY TQP UPLC-MS/MS device (Waters, Micromass Ltd., United Kingdom). Samples (10 µl) were injected into an ACQUITY TQP UPLC BEH C18 column (1.7 µm×2.1 mm×50 mm, Waters) and eluted with ammonium acetate (1.0 mM) in 10% methanol (A) and 100% methanol (B). The UPLC gradient was as follows: linear gradient of 100% A to 55.6% A and 44.4% B in 8 minutes, followed by a wash with 100% B and an equilibration to initial conditions with 100% A at a flow rate of 0.3 ml/min. The effluent was introduced into the electrospray source at a temperature of 150°C. Quantitative analysis was carried out using the internal standard ratio methods and the deuterated isotopes. The ESI(+)-MRM (multiple reactant monitoring) mode was used for quantification based on specific diagnostic transitions for the different compounds analyzed (Table S6). Chromatograms were processed using the Masslynx software (Waters) and the concentrations were calculated following the principles of isotope dilution.

Two-way analysis of variance followed by Dunnett’s multiple comparison test were used to determine statistical significance. The two-way analysis of variance examined the influence of both time and bacterial strain on cytokinin levels assuming equal variance across both groups. Dunnett's test was used to compare cytokinin levels in ΔfasD and ΔpFiD188 to D188, within each time point, assuming equal variance among the treatment strains.