Escherichia coli strain Top10 (New England Biolabs, Frankfurt am Main, Germany) was grown in LB medium at 37°C, Agrobacterium tumefaciens strain GV3101 was grown in YEB medium at 28°C and Xanthomonas oryzae pv. oryzae (Xoo) strains PXO83, PXO142, ICMP 3125^T and Roth X1-8 were cultivated in PSA medium at 28°C.

Oryza sativa ssp. japonica cv. Nipponbare was grown under glasshouse conditions at 28°C (day) and 25°C (night) at 70% relative humidity (RH). Leaves of 4-week-old plants were infiltrated with a needleless syringe and a bacterial suspension as previously described (Reimers and Leach, 1991). Nicotiana benthamiana plants were grown under 16 h of light, 40–60% RH, at 23°C (day) 19°C (night) in a growth chamber. Leaves of 4- to 6-week-old plants were inoculated with A. tumefaciens strains using a needleless syringe.

Libraries of genomic DNA of Xoo strains ICMP 3125^T and PXO142 were sequenced on a Pacific Biosciences RS II instrument. The library for Xoo strain ICMP 3125^T was sequenced using two SMRT cells yielding a total of 68,513 reads with an N50 read length of 22,914 bp. Reads were assembled using the HGAP_Assembly.2 pipeline from the Pacific Biosciences SMRT Portal with default parameters. This resulted in two contigs, one of 12,940 bp with only 35× coverage and one large contig of approximately 5 Mbp. Due to spurious coverage, the first contig was removed. The latter contig could be circularized manually and was shifted such that dnaA was located at position 45 in forward orientation relative to the origin. After circularization and shifting, this contig was further refined in an additional polishing step using the Resequencing.1 pipeline from the Pacific Biosciences SMRT Portal, yielding a chromosome of 4,990,672 bp in total. In resequencing, coverage across the chromosome was largely uniform (Supplementary Figure S1) with an average coverage of 170×.

The library for Xoo strain PXO142 was also sequenced using two SMRT cells with a total of 166,038 reads and an N50 read length of 23,191 bp. For PXO142, the HGAP_Assembly.2 pipeline was executed using default parameters except for “Minimum Subread Length = 3000” and “Minimum Seed Read Length = 5000,” yielding a single contig of approximately 5 Mbp. This contig was also further polished by the Resequencing.1 pipeline and could be circularized using the Circlator pipeline (Hunt et al., 2015) with default parameters, resulting in a chromosome of 4,982,118 bp in total. Finally, this chromosome was also shifted such that dnaA was located at position 45 in forward orientation relative to the origin. Again, coverage was largely uniform with an average coverage of 376×, except for a coverage peak at approximately 45 kbp that did not overlap any TALE genes (Supplementary Figure S2). Genome sequences of Xanthomonas strains have been deposited in NCBI GenBank¹ under accessions CP031697 (ICMP 3125^T) and CP031698 (PXO142).

From the genomes of Xoo ICMP 3125^T and Xoo PXO142, respectively, TALE genes were predicted by the “TALE Prediction” tool of AnnoTALE (Grau et al., 2016) (version 1.3) and subsequently assigned to TALE classes by the “TALE class assignment” tool. For Xoo ICMP 3125^T, 17 TALE genes (including two pseudo genes) were predicted, two of which belong to novel TALE classes TalES and TalET. For Xoo PXO142, 19 TALE genes (including three pseudo genes) were predicted, which were all assigned to classes already present in the AnnoTALE class builder.

TALE target genes and corresponding target boxes were predicted by the “Predict and Intersect Targets” tools of AnnoTALE (Grau et al., 2016). To this end, putative promoter sequences were extracted from the rice genome (MSU7 genome and annotation^2,3 ) as sequences spanning from 300 bp upstream of the transcription start site (TSS) until 200 bp downstream of the TSS or the start codon, whichever comes first, as proposed previously (Grau et al., 2013).

Rice cultivar Nipponbare leaves were inoculated with Xoo strains PXO142, ICMP 3125^T, or 10 mM MgCl₂ as a mock control in five spots in an area of approximately 5 cm using a needleless syringe. Two leaves of three rice plants each were inoculated for each strain and control, respectively. 24 h later, samples were taken, frozen in liquid nitrogen, and RNA prepared. Three replicates of this experiment were done on separate days and subjected to RNAseq analysis, separately. Stranded libraries were sequenced on an Illumina HiSeq 2500 instrument (Eurofins Genomics) as 100 bp single-end reads. General statistics of the sequencing data are provided in Supplementary Table S1.

RNA-seq data after inoculation with Xoo ICMP 3125^T and Xoo PXO142 were adapter clipped using cutadapt (v1.15) (Martin, 2011) and quality trimmed using trimmomatic (v0.33) (Bolger et al., 2014) with parameters “SLIDINGWINDOW:4:28 MINLEN:50.” Transcript abundances were computed by kallisto (Bray et al., 2016) using parameters “–single -b 10 -l 200 -s 40” using the cDNA sequences⁴ as reference transcripts. Differentially expressed genes relative to the control were determined by the R-package sleuth (Pimentel et al., 2017). Since replicates have been paired during library preparation and sequencing, the replicate was considered as an additional factor when computing p-values of differential expression. Differential expression was aggregated on the level of genes using the parameter target_mapping of the sleuth function sleuth_prep(), and log₂-fold change and Benjamini–Hochberg-corrected P-value were recorded. Gene abundances and sleuth outputs with regard to differential expression are provided as Supplementary Data S1.

RNA-seq reads were also mapped to the rice genome (MSU7) to obtain detailed information about transcript coverage and transcription starts. To this end, adapter clipped and quality trimmed reads were mapped using TopHat2 v2.1.0 (Kim et al., 2013). The BAM output of TopHat2 was then pooled across replicates and mapped reads were visualized using IGV v2.3.90 (Robinson et al., 2011; Thorvaldsdottir et al., 2013). RNA-seq data have been deposited in the European Nucleotide Archive⁵ under study accession PRJEB28127.

Phylogenetic trees of Xoo ICMP 3125^T, Xoo PXO142, and 15 further Xoo strains including Xoo AXO1947 as an outgroup were determined (i) based on their TALEs using the “TALE Class Presence” tool of AnnoTALE and (ii) using bcgTree (Ankenbrand and Keller, 2016) based on the protein alignments of its default set of 107 conserved genes.

For the TALE-based phylogenetic tree, distances between Xoo strains were determined as the sum of (a) the divergence score of the RVD-based alignment of TALEs in the same AnnoTALE class (Grau et al., 2016), (b) the divergence score of the most similar TALE of the respective other strain if two strains do not comprise TALEs from the same AnnoTALE class, or (c) a divergence score of 6 if no matching TALE exists (chosen to be larger than the cut height of 5 leading to AnnoTALE classes). The matrix of pairwise distances between Xoo strains served as input of agglomerative hierarchical clustering using single linkage as implemented in the Jstacs library (Grau et al., 2012) to yield the phylogenetic tree. This procedure is available as part of the “TALE Class Presence” tool since AnnoTALE version 1.3.

The second phylogenetic tree was computed using bcgTree (Ankenbrand and Keller, 2016) based on a set of 107 essential genes as described in Dupont et al. (2012) represented as HMMs. Internally, bcgTree uses multiple other tools, namely hmmsearch (Eddy, 2009) for searching matches to the HMM models, MUSCLE (Edgar, 2004) and Gblocks (Castresana, 2000) for aligning sequences as identifying conserved blocks, and RaxML (Stamatakis, 2014) for computing phylogenetic trees. The bcgTree wrapper was run with default parameters, specifying 200 bootstraps for RaxML. The final phylogenetic tree was then extracted from the “RaxML_bestTree.final” output of bcgTree.

Both trees were visualized comparatively using phylo.io (Robinson et al., 2016).

Genomic DNA from Xoo was isolated using phenol/chloroform extraction, digested with BamHI and separated on a 0.8% agarose gel at 90 V for 24 h. The genomic TALE sequences were detected with chemiluminescence on Southern blot using a digoxigenin-labeled (Roche Applied Science, Mannheim, Germany) probe derived from 500 bp of the 3′ part of talC from Xoo BAI3 (Yu et al., 2011) that hybridizes to TALE genes.

TALEs were constructed using the Golden TAL technology kit (Geißler et al., 2011). Individual repeats were subcloned into hexa-repeat modules and subsequently assembled to final TALE expression constructs. N- and C-terminal parts of Hax3 were employed for expression constructs (pSKA2) used in A. tumefaciens. Expression constructs for Xanthomonas harbored the N-terminal region from TalAG4 of strain PXO83 and the C-terminal region from TalAO3 of strain PXO83. The Xanthomonas expression vector (pSKX1) fuses a C-terminal FLAG Epitope to the TALE.

The Xoo strains carrying a plasmid encoding an artificial TAL effector, or empty vector, were grown in liquid PSA medium supplemented with 20 μg ml^-1 gentamicin at 30°C. Cells of 1 ml of a bacterial suspension at an OD₆₀₀ of 0.2 were harvested and the TAL effector expression was analyzed by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS–PAGE) and immunoblotting using an anti–FLAG antibody (Sigma–Aldrich).

The third leaves of 4-week-old Oryza sativa ssp. japonica cv. Nipponbare plants were clipped with Xoo bacterial solution adjusted to an OD₆₀₀ of 0.2 in 10 mM MgCl₂. 14 dpi leaves were harvested and lesion length was measured.

At two dpi, 5 cm inoculated segments were harvested and rice total RNA was isolated using the Qiagen RNeasy kit. cDNA was generated from 2 μg RNA using the Fermentas first-strand cDNA synthesis kit (Thermo Fisher Scientific Inc., Waltham, MA, United States) and real-time PCR was performed using the iCycler (Bio-Rad, München, Germany) as described before (Streubel et al., 2013 and Supplementary Table S6). The amplification efficiency for each primer pair was analyzed using a standard curve plot of a dilution series. cDNA amounts were normalized using actin as a reference gene. The fold change induction was calculated in comparison to leaves treated with 10 mM MgCl₂ by using the ΔΔCt method (Livak and Schmittgen, 2001). All experiments were repeated three times.

β-Glucuronidase assays from plant samples were performed essentially as described before (Boch et al., 2009). Briefly, PCR-amplified fragments of the promoters were cloned into a Golden Gate-compatible pGWB3 (Nakagawa et al., 2007) derivative containing a promoterless uidA reporter gene (Supplementary Table S6). To analyze reporter activity, A. tumefaciens strains delivering TALE constructs and GUS reporter constructs were resuspended in infiltration medium, resulting in an OD₆₀₀ of 0.8, mixed in equal amounts, and inoculated into N. benthamiana leaves. Two dpi, leaf disks were sampled and GUS activities were quantified using 4-methylumbelliferyl-β-D-glucuronide (MUG). Total protein concentrations were determined using Bradford assays. Leaf disks for qualitative staining were harvested in parallel. Histochemical staining was performed using 5-bromo-4-chloro-3-indolyl β-D-glucuronide (X-Gluc). Data were compiled from triplicate samples originating from different plants. All experiments were repeated three times.

To compare the TALomes of Xoo strains from different agricultural regions, we sequenced the strains PXO142 and ICMP 3125^T, originating from the Philippines and India, respectively. The complete genome sequences of both strains were obtained using PacBio sequencing, which produces long reads that are suitable to correctly assemble TALE genes in their genomic context. The genomes of Xoo PXO142 and ICMP 3125^T were assembled into single contigs of 4.99 Mbp length, each (Table 1 and Supplementary Figures S1, S2). These genomes were compared to the other available Asian Xoo genomes and phylogenetic trees were created based on conserved genes and TALE RVD sequences (Figure 1C and Supplementary Figure S3). Using the AnnoTALE prediction pipeline, TALE genes were identified in these genomes and assigned to individual TALE classes. At present, the total number of 278 Asian Xoo TALEs from the fully sequenced strains fall into 38 different classes⁶. PXO142 and ICMP 3125^T contain 19 and 17 TALE genes, respectively, which were assigned to 23 different TALE classes (Figure 1A and Supplementary Table S3). Eleven of these classes are shared by both strains whereas the remaining 12 are only present in one of the strains. Six identified TALE classes have a prominent member that has previously been described in the context of resistance reactions – AvrXa23 (TalAQ) and AvrXa27 (TalAO) – or as an important virulence factor – PthXo3 (TalBH), PthXo6 (TalAR), PthXo7 (TalBM), and PthXo8 (TalAP) (Figure 1). The location of TALE genes in both strains is confined to the previously described TALE clusters T-I to T-IX (Figure 1). The MAUVE alignment of the sequenced genomes (Figure 1B) revealed a high level of genomic rearrangements, which is typical for plant pathogenic bacteria (Darling et al., 2004). TALE clusters T-III, T-IV, T-VII, and T-VIII are in the direct vicinity of rearrangements, whereas the other clusters are harbored in more stable genomic regions (Figure 1B).

In order to evaluate the TALome diversity in all 16 sequenced Asian Xoo strains, the TALE classes were assigned to three abundance categories depending on how frequently they occur (Figure 2). Core TALE classes were defined as being present in more than 80% of strains, intermittent TALE classes in 20–80% of strains, and rare TALE classes in less than 20% of strains (Figure 2). Notably, TALE genes with truncated N- or C-terminal regions do not bind to DNA and do not activate gene expression, respectively. Some of these TALEs block resistance reactions and have been termed truncTALEs or iTALEs and were classified, accordingly. Other truncated TALEs were categorized as pseudogenes. PXO142 contains nine core TALEs, three intermittent TALEs, three rare TALEs, two pseudogenes, one truncTALE/iTALE type B and one truncTALE-related/iTALE type A. ICMP 3125^T carries ten core TALEs, two intermittent TALEs, three rare TALEs, one truncTALE/iTALE type B and one truncTALE-related/iTALE type A. Interestingly, TalES1 and TalET1 (both in cluster T-II) of ICMP 3125^T are unique and have no other class members, so far.

Our analysis revealed that the TALE clusters T-I to T-III contain the majority of rare and intermittent TALE classes and are highly diverse in their composition. Therefore, these clusters have the highest potential for new TALEs to be discovered in the future. On the contrary, T-IV to T-IX are highly conserved in their TALE class content and contain a high amount of core TALEs. The consistency of these clusters suggests that the TALEs in these clusters play an important role in Xoo infection.

A phylogenetic tree of all 38 different TALE classes from 16 Xoo strains was created by aligning the most likely common binding sites for each TALE class with Clustal Omega and visualization was done using iTol (Figure 2) (Sievers et al., 2011; Letunic and Bork, 2016). Apparently, some rare TALE classes are related to core TALE classes present in a different strain at the same genomic locus. Examples are the TALE classes TalAQ and TalDU, TalAC and TalDS, as well as TalAR and TalDT (Figure 2), which are mutually exclusively present in the same genomic location in different strains (Supplementary Figure S4). Evidently, the deletion of one to three repeats triggered a separate classification by the AnnoTALE software, because the insertion or deletion of repeats typically has a significant impact on TALE binding. As TalAQ (AvrXa23) and TalAC (AvrXa7) trigger a resistance reaction in rice cultivars carrying the Xa23 and Xa7 resistance loci, respectively (Yang and White, 2004; Wang et al., 2015), the deletion of repeats could circumvent recognition by the plant. Alternatively, these changes could be adaptations to indel mutations in target promoter sequences in different rice cultivars (Richter et al., 2014; Erkes et al., 2017). Using the daTALbase tool (Pérez-Quintero et al., 2018) we searched for variations in the TALE boxes of TalAQ, TalAC, and TalAR in the promoters of their target genes, OsFNS, OsSWEET14, and OsTFX1, respectively, but no variations were found.

All truncTALE/iTALEs are closely related in their RVD sequences, suggesting a common origin. TALE class TalDR is exclusively a truncTALE/iTALE type B and TALE class TalAW is exclusively a truncTALE-related/iTALE type A. TALE classes TalDQ and TalAI have TALE genes of either truncTALE/iTALE type.

In order to identify plant target genes of different TALEs, we combined in silico target predictions with in vivo gene expression data. The top 100 target sequences for each TALE from Xoo PXO142 and ICMP 3125^T were predicted using TALgetter (Grau et al., 2013). The promoterome of the cultivar Nipponbare was restricted to 300 bp upstream and 200 bp downstream of transcriptional start sites, which is the most promising region for TALE binding sites (Grau et al., 2013). Overall 2,430 unique potential target genes were predicted for both strains combined, of which 430 were identified for both strains (Figure 3A).

To identify real TALE-induced genes among the in silico predicted target candidates, the expression of rice genes was determined following infection with Xoo. The rice cultivar Nipponbare was inoculated with PXO142, ICMP 3125^T, or a mock control, and harvested after 24 h for subsequent RNA-seq analysis. This relatively early harvest time was chosen to favor primary TALE targets and minimize secondary effects at the risk of missing later changes. Comparing gene expression in infected and uninfected tissue enables the identification of Xoo-mediated gene induction. Differences in the TALE repertoire of these Xoo strains should enable the comparison of TALE class presence and absence. The RNA-seq analysis yielded 358 and 413 induced rice genes (≥1.5-fold) during infection with PXO142 and ICMP 3125^T, respectively. One hundred and fourteen of these rice genes were induced by both strains. Thirty-six promising target gene candidates for 15 TALE classes were identified by combining in silico prediction and RNA-seq analysis (Supplementary Table S2 and Figure 3B).

In order to assign induced plant genes to individual TALEs, Xoo strains expressing only one TALE gene were created and their impact on target gene expression was monitored (Figure 3C). To this end, the American Xoo strain Roth X1-8, which has no natural TALEs (Supplementary Figure S5), was transformed with expression plasmids carrying individual TALE genes. TALE expression constructs were created by using the Golden TAL Technology cloning Kit (Geißler et al., 2011). To facilitate the exact replication of the natural RVD composition in the artificial TALEs, we expanded the original cloning kit with additional RVD modules and assembly vectors (Supplementary Figure S6). Furthermore, to match the artificial TALEs as closely as possible to their natural counterparts, we used Xoo-specific N- and C-terminal regions (Supplementary Figures S7, S8). Twelve different Roth X1-8 derivatives were created carrying single TALEs ranging from 11.5 to 28.5 repeats. The production of the individual TALEs was verified via Western Blot (Supplementary Figure S9). The impact of single TALEs on the virulence of the TALE-less Xoo strain Roth X1-8 was tested by clipping inoculation of Nipponbare leaves with bacterial solution (Supplementary Figure S10). TalBH2 supported bacterial growth, which was expected, because it induces expression of the well-known OsSWEET14 virulence target. The other TALEs analyzed showed no clear contribution to virulence under the tested conditions.

To analyze expression of target genes, rice leaves were inoculated with Xoo strains and samples were taken for RNA isolation and qRT-PCR analysis. The rice variety Nipponbare was inoculated with Roth X1-8, Roth X1-8 carrying single TALE expression constructs, the wild type strains PXO142, ICMP 3125^T, PXO83, and a mock control. Xoo strain PXO83 was previously sequenced and the analyzed TALome is available (Grau et al., 2016). As PXO83 has members of the same TALE classes as PXO142 and ICMP 3125^T, we expected a similar induction in common predicted target genes. Samples were collected after 2 days to facilitate robust assessment of gene induction via qRT-PCR. For five TALE classes (TalAE, TalAF, TalAG, TalAN, TalET) no target candidates were identified in the combinatorial target prediction and RNA seq analysis. Furthermore, for TalAL, the only identified candidate (a retrotransposon) was not very promising. In these cases we chose other candidates from the target prediction list for qRT-PCR tests.

We were able to match twelve specific target genes to TALE classes and confirmed these as TALE-dependently induced genes (Figure 4 and Supplementary Table S4). All 12 target genes were induced significantly in a comparison between strain Roth X1-8 with or without the corresponding TALE with p-values below 0.05.

Three TALE targets were known previously: OsSWEET14, OsTFIIAγ1, and OsTFX1 are induced by TALE classes TalBH, TalBM and TalAR, respectively. Five TALE targets were only hypothesized to be TALE targets before and are now experimentally confirmed: OsLsi1, OsHEN1, OsPHO1;3, OsNPF6.3 and OsFNS are addressed by TALE classes TalAL, TalAP, TalAO, TalAE and TalAQ, respectively. Finally, four identified TALE targets have not been described before: OsPL, OsWAK51, OsHLS1 and OsFBX109 are manipulated by TALE class TalAB, TalES, TalBA, and TalAD. The target genes OsLsi1 of TalAL and OsNPF6.3 of TalAE were ambiguous in our RNAseq experiment at 24 hours post inoculation (hpi), but clearly induced in our qRT-PCR experiment at 48 hpi. This observation suggests that some TALEs might have a delayed effect within the host cell that is more pronounced at 48 hpi than at 24 hpi.

The predicted TALE boxes in the promoters of the target genes match well to the RVD-DNA base specificities (Figure 5). They are located on average 240 bp upstream of the start codon with outliers very close to the start codon (TalES1 box 21 bp upstream of OsWAK51 CDS) and quite far from it (TalBA8 box 635 bp upstream of OsHLS1 CDS) (Figure 5). Most TALE boxes are also very close to the annotated natural transcription start site with the average TALE box located 57 bp upstream (Figure 5). This distance is well suitable because TALEs typically determine the onset of transcription in a distance of 40–60 bp after their binding site. In contrast, the TALE boxes of TalAB16, TalAD23, and TalAO16 are located 122, 189, and 268 bp upstream of the natural transcription start sites, respectively (Figure 5). Accordingly, the apparent transcription start site of OsPHO1;3 in plants infected with Xoo strains carrying TALE TalAO appears to be shifted in comparison to the uninfected tissue in our RNA-seq data (Supplementary Figure S11). This supports the observation that TALEs can control the transcriptional start site of induced target genes. The TALE boxes had a mean mismatch rate of 11% with TalES1 fitting the promoter perfectly and TalBH2 tolerating six mismatches out of 29 bases.

The expression patterns of all twelve genes during infection coincide with the presence/absence of their corresponding TALE class in a given Xoo strain. This suggests that the TALE class affiliation is a reliable indicator of TALE function. The specific functions of 67% of all known TALEs of Asian Xoo strains can now be predicted because they belong to a TALE class containing a member with a known target gene.

We aimed to determine whether the TALEs directly or indirectly induced the identified target rice genes. Direct induction would involve binding of the TALE to the target promoter whereas indirect induction could be mediated via secondary effects or secondary regulatory rice genes. To distinguish this, transient GUS reporter assays were performed in the heterologous plant Nicotiana benthamiana using individual TALEs and subcloned rice promoters. A GUS activity would imply a direct recognition of the target promoter by the TALE because secondary targets would not be present in N. benthamiana (Figure 3D).

1,000 bp upstream of the start codon of the twelve identified rice target genes were amplified from rice cultivar Nipponbare DNA and cloned in front of a promoterless uidA coding sequence. Artificial TALEs were constructed with the same RVD composition as the natural TALEs (Figure 5) and cloned into Agrobacterium binary vectors. The GUS reporter constructs and the TALE expression constructs were introduced into Agrobacterium tumefaciens and the respective strains were co-inoculated into N. benthamiana. After 2 days, samples were harvested for quantitative and qualitative GUS assays (Figure 6).

Most reporter constructs show very little GUS activity without a TALE indicating that the basal expression rates of these rice promoters in N. benthamiana are low. Only OsNPF6.3 (Os08g05910) shows a relatively pronounced GUS activity without TalAE15. The GUS activity in samples with the corresponding TALEs compared to controls was significantly higher for all 12 tested target genes. The GUS activity of the different reporter constructs varied with OsNPF6.3 (Os08g05910), OsPHO1;3, OsTFX1, and OsHLS1 reporter constructs showing the highest absolute GUS activity when paired with their matching TALEs (Figure 6). This corresponds to induction rates of 10- to 100-fold for most combinations. The only exceptions are OsNPF6.3 and OsTFIIAγ1 with less than 10-fold increases in GUS activity and OsLsi1 and OsHLS1 with an increase of more than 100-fold (Figure 6). These results indicate that the examined TALEs are able to directly recognize the identified promoters in vivo and trigger gene expression.

Key virulence targets that are important for a successful infection are often addressed by multiple virulence factors across pathogen species and host plants. One of the best-studied examples of such a convergent evolution for TALE targets are SWEET genes. SWEET genes in rice, cassava, sweet orange, and cotton are induced by different TALEs from different Xanthomonas strains (Yang and White, 2004; Cohn et al., 2014; Zhou et al., 2015; Cox et al., 2017). Here, we propose two new cases in which TALEs from different origin converge on candidate virulence targets. The first target family consists of the 2-oxoglutarate dioxygenase (DOX) genes OsFNS (Os03g03034) and Os04g49194. These genes share 69.6% identity and 85.6% similarity (Figure 7C and Supplementary Figure S12). They are closely related homologs that are predicted to convert flavanones to flavones, but the experimental evidence is contradictory (Kim et al., 2008; Lam et al., 2014; Falcone Ferreyra et al., 2015; Zhang et al., 2017). Our target analysis revealed that Os03g03034 is induced by Xoo TalAQ (Figure 4, 6). In addition, induction of both genes by Xoc has been proposed by TALE target predictions and transcriptomic data (Cernadas et al., 2014), but experimental validation for a direct induction is lacking. The binding specificities of the Xoc TALE classes TalBR and TalBL match well to the promoter regions, respectively (Figure 7A). We re-constructed one Xoc TALE of each class and measured recognition of the respective target promoters in GUS reporter assays (Figure 7B). These experiments show that three distinct TALE classes of different X. oryzae pathovars are targeting these rice genes. Xoo strains address Os03g03034 with TALE class TalAQ, whereas Xoc strains target Os03g03034 and Os04g49194 with TALE classes TalBR and TalBL, respectively. Interestingly, TalAQ and TalBR have the exact same binding sequence in the Os03g03034 promoter even though 16 out of their 27 RVDs differ (Figure 7A). The functional relationship between both targets has so far been overlooked. To acknowledge this connection, we propose to rename Os03g03034 to OsDOX-1 and Os04g49194 to OsDOX-2.

The second new target of convergent TALE classes is the gene OsLsi1 (Os02g51110) which encodes a putative silicon transporter (Ma et al., 2006), and which has been identified by us as a Xoo TalAL target (Figure 4, 6). This gene was also previously hypothesized to be a target of the TALE class TalAV (Erkes et al., 2017). We re-constructed Xoc TalAV1 and tested the direct recognition of the target promoter in a GUS-assay (Supplementary Figure S13). This experiment shows that the promoter of OsLsi1 is directly recognized by TALE classes TalAL from Xoo and TalAV from Xoc. The two TALE classes bind different positions in the promoter, suggesting a co-evolution. PXO142 might even be adapted to different versions of this promoter, as it has two members of TALE class TalAL with minor differences in their RVDs (Supplementary Table S5). According to the daTALbase, there is a known variant of the OsLsi1 promoter with the altered TALE box TAG[C/T]TAGCTCGATCTCCCT, but the SNP does not correspond to differences between the TALE class TalAL members.

These two examples demonstrate that Xoo and Xoc have more common infection strategies than previously believed.

The strains ICMP 3125^T and PXO142, which were sequenced in this study, expand the variety of fully sequenced Asian Xoo to 16 strains originating from India, the Philippines, Japan, Korea, and Taiwan (Ochiai et al., 2005; Salzberg et al., 2008; Streubel et al., 2013; Booher et al., 2015; Quibod et al., 2016; Char et al., 2018; NZ_CP011532.1). The isolations span a time from 1965 to 2006 for ICMP 3125^T and PXO602, respectively (Ochiai et al., 2005; Salzberg et al., 2008; Streubel et al., 2013; Booher et al., 2015; Quibod et al., 2016; Char et al., 2018; NZ_CP011532.1). With these resources, a spatiotemporal view of Asian Xoo strains can now be established. In order to get a comprehensive picture of differences and similarities between strains, a variety of tools was applied. At first, we characterized the TALomes of all strains with AnnoTALE by assigning TALEs into 38 TALE classes. The TALE classes represent groups of TALEs with similar RVD sequences and subsequently common binding sequences and target genes (Grau et al., 2016). Next, we assessed the frequency of each TALE class in those 16 strains. We identified 12 core TALE classes present in over 80% of strains, 18 rare TALE classes found in fewer than 20% of strains and eight intermittent TALE classes occurring in 20–80% of strains. Core TALE classes have a high prevalence and are therefore likely to be generally important for the infection, as they are conserved over a long time and different geographical regions. At the same time, rare TALE classes might be involved in adaptations to different climatic regions or regional rice cultivars. One example is the rare TALE class TalBM targeting OsTFIIAγ1, which only promotes virulence during infection of rice cultivars with a xa5 genomic background (Sugio et al., 2007). Additionally, we found some rare TALE classes that are closely related to core TALE classes, which might circumvent resistances by altering existing TALE genes to evade detection or to overcome promoter mutations.

In order to investigate the targets of the TALE classes present in the strains ICMP 3125^T and PXO142, we combined in silico target predictions with in vivo transcriptomic data. Our analyses revealed 36 potential target genes that were TALE-dependently induced and contained a predicted TALE box in their promoter region. Of these, 12 could be verified and attributed to individual TALEs. Among these target genes were the three previously published targets OsSWEET14, OsTFIIAγ1, and OsTFX1 for TALE class TalBH, TalBM, and TalAR, respectively (Sugio et al., 2007; Streubel et al., 2013). Additionally, we provide experimental data to support the relationship between five more TALE classes with their previously hypothetical target genes OsLsi1 (TalAL), OsHEN1 (TalAP), OsPHO1;3 (TalAO), OsNPF6.3 (TalAE), and OsDOX-1 (TalAQ) (Grau et al., 2013; Cernadas et al., 2014). Finally, we introduce four new TALE target genes, that have not been described before: OsPL (TalAB), OsWAK51 (TalES), OsHLS1 (TalBA), and OsFBX109 (TalAD). Because different members of the TALE classes were able to induce the same target genes, the consistency between TALE classes and common target genes could be shown. Taken together, this broadens our knowledge of Xoo TALE target genes significantly with now 60% of all functional Asian Xoo TALE genes having a known target (Yang and White, 2004; Yang et al., 2006; Chen et al., 2010; Streubel et al., 2017; Tran et al., 2018). Especially the targets of core TALE classes are brought to light, as we expand the number of experimentally confirmed targets from one (OsTFX1) to seven (OsTFX1, OsPL, OsFBX109, OsNPF6.3, OsPHO1;3, OsHEN1, OsDOX-1) out of eleven. Addressing targets of core TALEs might be the most sustainable way to provide resistance against a variety of Asian Xoo strains.

It can be expected that several of the confirmed and postulated target genes have relevance in Xoo infection of rice. In fact, some of them can be grouped into known pathogenicity hubs often manipulated by Xanthomonas. In the following, possible virulence implications of selected TALE targets are discussed.

Most known susceptibility targets of X. oryzae TALEs were identified by changes in lesion length based on presence/absence of the corresponding TALE in the Xanthomonas strains. However, in most cases only one or two TALEs of Xoo or Xoc strains show any influence on lesion length (Cernadas et al., 2014). With the knowledge that a lot of TALEs without obvious growth phenotype belong to a highly conserved core TALE class, it seems unlikely that none of them are important for infection. Instead, it is conceivable that these TALEs are benefitting infection without influencing symptom formation and bacterial growth under artificial inoculation conditions. For example, susceptibility factors like PthXo7 or iTALEs/truncTALEs only show their beneficial effect under very specific conditions (Sugio et al., 2007; Ji et al., 2016; Read et al., 2016). In order to identify the virulence function of TALE classes without obvious phenotypes, a targeted approach based on the potential function of the target genes which we have outlined here, will be needed.