Xanthomonas transcription activator-like effectors (TALEs) enhance bacterial virulence by acting as transcription factors within plant cells. TALEs contain a modular DNA-binding domain with a predictable and designable specificity1,2. The DNA-binding specificity of TALEs is determined by a central domain of near-perfect 33 to 35-amino acid-repeats. Each repeat aligns with one base pair of the bound DNA sequence, with two variable amino acids at position 12 and 13 (termed the ‘RVD’, for repeat-variable diresidue) acting as the determinants of base-recognition specificity1,2. X-ray crystallography of TALE-DNA complexes has revealed that only amino acid 13 actually contacts the DNA base and is the main determinant of specificity3,4, while the amino acid 12 forms inter-repeat contacts that contribute indirectly to the binding efficiency3,4. Certain RVDs (NI for A, HD for C, NG for T and NN for G or A) are predominantly used in nature, but several others occur less frequently5. The DNA-specificities of all 400 theoretically possible RVD di-amino acid combinations have been determined experimentally6,7,8,9,10 and have confirmed that position 13 controls DNA recognition specificity. In addition, the so-called ‘strong RVDs’, HD and NN, form the strongest interaction with the DNA bases and are required for an overall efficient binding of TALEs to DNA9. The one base-one repeat specificity of TALEs is extended by one 5′ thymine1,2 that is specified by the N-terminal domain of TALEs. This domain also contributes to efficient DNA interaction11,12,13 and likely facilitates initial DNA contact. The unique modularity of TALEs has led to their widespread use as laboratory reagents for genome engineering and transcriptional reprogramming14,15. By choosing the appropriate number of repeats and RVD-types, practically any desired DNA-binding specificity can be engineered. In contrast to the well-documented use of TALEs in biotechnology, their natural diversity in plant-pathogenic Xanthomonas spp. bacteria is not well understood.

TALEs and related proteins are bacterial virulence factors found primarily in plant-pathogenic Xanthomonas spp. and Ralstonia solanacearum bacteria5. TALE proteins are secreted by the bacterial type III secretion system into plant cells where they localize to the plant nucleus and activate the expression of specific target genes. The N-terminal portion contains the type III secretion signal and the C-terminal portion contains the nuclear localization signals and an acidic activation domain; all three are highly conserved among TALEs5. In contrast, the number of DNA-binding repeats and their RVDs vary greatly, suggesting that TALEs can activate a wide variety of different plant target genes.

Rice-pathogenic Xanthomonas oryzae pv. oryzae (hereafter termed ‘Xoo’) and Xanthomonas oryzae pv. oryzicola (Xoc) are serious threats to the agricultural production of rice16. Both pathogens typically harbor large TALE repertoires which, in case of Xoo, differ in number and nature between Asian (15 to 26 TALEs), African (8 to 10 TALEs) and North-American (no TALEs) strains17,18,19,20. Although some TALEs contribute significantly to pathogen virulence, the physiological role of most remains unclear21. Computational algorithms have used the RVD-based target specificity to predict possible TALE target sequences in rice promoters22,23,24, but biological confirmations of these predictions are for the most part lacking.

Three complete Xoo and one Xoc genomic sequences including TALE genes have been generated by classical Sanger sequencing25,26,27,28 and two draft genomes by Illumina sequencing19. Next-generation sequencing (NGS) methods like Illumina yield short reads unsuitable for the assembly of the highly repetitive TALE sequences. Recently, two genomic TALE genes were correctly assembled following Illumina sequencing of a Xanthomonas translucens pv. cerealis strain29, but the large number of TALE genes present in most X. oryzae strains precludes a similar strategy. The TALE repertoires in Xoo and Xoc are highly variable. At present we are unable to establish their diversity although this is essential for understanding the contribution of these potent virulence factors to pathogen performance.

The nomenclature of Xanthomonas TALEs has been non-uniform. Historically, individual TALE genes have been sub-cloned and assigned TALE names based on their reaction in resistant hosts (names starting with “Avr” for avirulence), their contribution to pathogenicity (names starting with “Pth” for pathogenicity), or their sequential location in the genome (e.g. Tal1, Tal2, Tal3, etc. or Tal9a, Tal9b, etc. if part of the same genomic locus). Because of frequent genomic rearrangements in different strains of Xoo28, the order of genomic loci and thus TAL names, is not consistent. Consequently, the current TALE nomenclature fails to reveal similarities between TALEs that could indicate common host targets and is profoundly confusing, because completely different TALEs can have identical names and near-identical TALEs can have different names (e.g. Tal9c of strain Xoo PXO99A and AvrXa27 are synonyms and 99.4% identical on DNA-level to Tal1c of strain Xoo MAFF311301828).

Here, we demonstrate that the PacBio NGS method30 is uniquely suitable for sequencing Xanthomonas genomes that harbor a large number of TALE genes. This new approach enables rapid characterization of TALE repertoires revealing a pathogen’s complete virulence arsenal. To assist in our analysis, we have developed a suite of applications for predicting TALE genes in sequenced genomes and for grouping TALEs into classes that indicate possible functional relationships to known TALEs. We envisage that a large number of TALE sequences will emerge from PacBio sequencing in the near future and we propose a unified TALE nomenclature based on this classification. Our work contributes to a deeper understanding of plant-pathogen interactions and could help reduce the damage wrought by Xanthomonads on domestic crops.

Results and Discussion

Selection of rice-pathogenic Xanthomonas oryzae pv. oryzae strains from the Philippines

Rice is grown in many areas of the world and is the predominant food crop of Asia. Accordingly, Xanthomonas oryzae pv. oryzae (Xoo) is a worldwide scourge and different lineages exist in Africa, Asia and North-America18. Rice lines resistant to specific isolates of Xoo have been cultivated and several of these (e.g. Xa3, Xa7, Xa10, Xa23) stem from altered responses to TALEs. This, in turn, has led to selective pressure on Xoo to diversify its repertoire of TALEs.

For the study reported here, we focused on Xoo strains from the Philippines, a relatively confined region with a long-standing history of Xoo-rice interaction research31,32. Philippine Xoo strains have been classified into eleven races based on their pathogenicities on near-isogenic rice lines harboring individual resistance loci18,31. We chose six strains belonging to four races that are distantly related to each other as well as to the archetype sequenced strain, Xoo PXO99A (Table 131). All six strains cause disease in rice (Oryza sativa ssp. japonica cv. Nipponbare) (Fig. 1a). The TALE genes in these strains were initially analyzed by Southern blots of BamHI-digested genomic DNA (Fig. 1b). BamHI digestion distinguishes TALEs by size, mainly reflecting the varying number of repeats. This analysis revealed a non-identical, but apparently partially overlapping TALE repertoire.

Table 1 Xanthomonas oryzae ov. oryzae strains used in this studya.
Figure 1
figure 1

Phenotypes and race-specific TALE gene pattern of Philippine Xoo strains.

(a) Leaves of 3- to 4-week old rice cultivar Nipponbare plants were infiltrated with 10 mM MgCl2 or Philippine Xoo strains PXO35 (race 1), PXO132 (race 1), PXO83 (race 2), PXO142 (race 3), PXO143 (race 3), PXO183 (race 5) and PXO99 (race 6), respectively. Plant reactions were documented five days post inoculation. (b) Southern blot analysis. Genomic DNA of the Philippine Xoo strains PXO35 (race 1), PXO132 (race 1), PXO83 (race 2), PXO142 (race 3), PXO143 (race 3), PXO183 (race 5) and the reference strain PXO99 (race 6) was digested with BamHI, separated on an agarose gel and transferred to a nylon membrane. TALE-containing fragments were detected using a DIG-labelled probe corresponding to 500 bp of the 3′ part of talC. Known repeat numbers of TALEs and their corresponding fragments in the sequenced Xoo strain PXO99 are indicated at the right side. The approximate sizes of marker fragments estimated from the ethidium bromide-stained gel are indicated at the left side. The large fragments in lanes 3 and 5 (PXO83 and PXO143) are due to an absence of the typically conserved BamHI sites in certain TALEs.

Decoding genomic TALEs using PacBio

TALE genes have traditionally been cloned via cosmid libraries, but this is a time-consuming method and does not guarantee that all possible TALE genes are identified. It is possible for TALE genes to be correctly assembled following Illumina sequencing, provided the number of TALE genes is small and the inter-repeat differences are substantial29, but this is not feasible for Xoo harboring numerous similar TALE genes. PCR-amplification and sub-cloning of individual TALE open reading frames is possible29, but encounters two problems. First, it cannot be excluded that truncated amplification products mis-prime on different TALE templates and in different repeat regions, generating artificial chimeric products. And second, PCR-amplification requires the primers to be complementary to conserved regions - typically the 5′ and 3′ regions of TALE genes - where they cannot amplify the flanking genomic context. Therefore, this method does not reveal genomic TALE loci, omitting an interesting piece of information to unravel evolutionary events during TALE genesis.

We chose PacBio sequencing to characterize and compare the TALE repertoire of Xoo strains, because this method is uniquely able to produce long reads from singular templates30. As proof of concept, we sequenced the genome of Xoo PXO83, a race 2 strain that clusters with race 3 and race 5 strains in RFLP analysis31 and which differs from the sequenced race 6 Xoo strain PXO99A. PacBio sequencing of approximately 170x coverage resulted in a single contig despite the presence of numerous repetitive elements (Table 2). Trimming contig ends and circularization yielded a final PXO83 chromosome of 5,025,428 bp (Genbank accession no. CP012947).

Table 2 Xoo PXO83 genomic parameters compared to other Xoo genomes.

To assess the quality of the final PXO83 assembly, we performed a re-sequencing experiment using the same PacBio reads that were used for the assembly. We found a generally homogeneous coverage of the assembled chromosome (Supplementary Fig. S1), with a mean coverage of approximately 182. We investigated the stability of the results by artificially reducing the set of reads by sub-sampling (Supplementary Fig. S2). We found that almost the complete chromosome is covered by at least one read using only 5% of the PacBio data, while 99.89% of the chromosome is covered by at least 100 reads using all of the data (Supplementary Fig. S3). We assessed the concordance of base calls for these reduced sets (Supplementary Fig. S4) and the potential ambiguity of base calls (Supplementary Fig. S5) and found that 30% of the PacBio reads would have been sufficient to yield 100% concordance to the final chromosome with low ambiguity.

As we were especially interested in the TALE genes of PXO83, we additionally considered the mean (Supplementary Fig. S6) and position-specific (Supplementary Fig. S7) coverage of TALE repeats and found that all TALEs are sufficiently covered, that coverage is largely uniform in successive repeats (Supplementary Fig. S8) and, importantly, that no divergent patterns are present for RVD-coding codon pairs. We did not observe a divergent pattern for RVD-coding positions regarding the ambiguity of base calls (Supplementary Fig. S9). Finally, we asked whether a sufficient number of PacBio reads completely spanned each TALE - important because of their repetitive nature - and found that each is spanned by at least 10 continuous reads (Supplementary Fig. S10).

A progressiveMauve33 alignment of the Xoo PXO83 genome with the Xoo strains PXO99A, MAFF311018 and KACC10331 demonstrated that PXO83 is most closely related to PXO99A with distinct rearranged regions (Fig. 2).

Figure 2
figure 2

Comparison of Xanthomonas oryzae pv. oryzae genomic regions.

ProgressiveMauve alignment of four fully sequenced genomes of Xanthomonas oryzae pv. oryzae (Xoo) strains. Similarly coloured areas represent genomic regions with significant synteny. The genomes of Xoo KACC10331 and Xoo MAFF311018 are shown in reverse complement to simplify the view, because of extensive genomic rearrangements. To indicate this, the origin and the orientation of dnaA is indicated by a black dot with horizontal arrow. Black arrowheads point to the positions of the genomic TALE clusters. The numbering is according to Fig. 5.

Computational extraction and annotation of the genomic TALE repertoire

We developed a novel suite of applications, called “AnnoTALE”, for 1) identifying and analyzing TALEs in Xanthomonas genomes; 2) clustering TALEs into classes by their RVD sequences; 3) assigning novel TALEs to existing classes; 4) proposing TALE names using a unified nomenclature; and 5) predicting targets of individual TALEs and TALE classes (Fig. 3a). The suite of applications is available at “” as a JavaFX-based stand-alone application with graphical user interface for interactive analysis sessions. In addition, we provide a command line application that may be integrated into other pipelines. Both are based on the open-source Java library, Jstacs34 and use identical code for the actual analysis, ensuring consistent results between both versions. In the following section, we use the newly sequenced genome of race 2 Xoo strain PXO83 and its TALE repertoire to exemplify the steps of the AnnoTALE pipeline (Fig. 3a).

Figure 3
figure 3

Workflow for AnnoTALE.

(a) Diagram of input parameters, workflow and program parts. The newly assigned names of TALEs consist of Tal (black), TALE class (XY, light blue) and allele number of TALE within a class (123, purple). Yellow boxes represent input parameter, orange boxes show generated data and green boxes indicate the program parts that perform the individual job. To assign TALEs to existing classes the current class definition can be downloaded and imported into the program. Besides genomic sequences, individual TALE DNA sequences can optionally be loaded into the program. N, N-terminal region; C, C-terminal region; RVD, repeat variable di-residue. (b) Examples of the TALE Class Assignment tool. Two representative TALE classes, AD and AP, are displayed. RVD changes between TALEs are marked in black.

Predicting and analyzing TALE genes

AnnoTALE identifies TALE genes in Xanthomonas genomes based on the DNA sequence homology of individual TALE domains. The most prominent feature of TALEs is their DNA-binding domain, which comprises an array of highly conserved tandem repeats. Since the number of repeats is not known beforehand, we first scan the genome for individual repeats, which are then joined into contiguous stretches. At the 5′ and 3′ ends of these stretches, we search for occurrences of the TALE N- and C-domains, yielding a TALE open reading frame (ORF). For all three domains, we consider the homology on the DNA level to identify putative pseudogenes, which might be missed by comparing only amino acid sequences.

Finally, we refine the ORF so that the reported TALE gene starts with an ATG and ends with a stop codon. If this refined region is substantially shorter than the initial match, i.e., more than one third of the initial C- or N-terminal match is missing, we label it as a putative TALE pseudogene. Applying the “TALE Prediction” tool of AnnoTALE to the Xoo PXO83 genome, we found 16 full-length TALE genes and 2 pseudogenes (Table 3, Supplementary Table S1). Using the “TALE Analysis” tool, we split TALEs into their individual N- and C-terminal domains and repeats. This allows RVD sequences to be extracted and repeats of aberrant length to be identified35. Two of the 16 identified TALE genes of Xoo PXO83 have an aberrant long repeat and the two pseudogenes have an aberrant short repeat (Supplementary Fig. S11b,d).

Table 3 Xoo PXO83 TALEs.

Recently, the genome sequence of Xoo strain PXO86, a close relative of strain PXO83, was published36. The authors used command line programs termed PBX toolkit to perform first a local assembly of TALE gene reads from PacBio raw data and to extract RVD sequences prior to assembly of the whole genome36. In contrast, AnnoTALE is a user-friendly tool collection with graphical user interface for characterizing TALEs in already assembled genomes.

Building classes of known TALEs

To compare the Xoo PXO83 TALEs to those of the characterized Xoo and Xoc strains and to individually sequenced TALEs (Supplementary Table S2), we developed a pairwise measure of TALE divergence. The basic intent is to group TALEs with related target specificity together, since this could signify related functionality. Because TALE-specificity primarily depends on the sequence of RVDs, we chose a measure based on the alignment of RVD sequences. Within the RVD alignment, mismatching amino acids at positions 12 and 13 are given penalties of 0.2 and 0.8, respectively, reflecting the importance of amino acid 13 for target specificity and the minor, but measurable influence of amino acid 12 for activation efficiency3,4. Hence, if both amino acids of an RVD mismatch, we score this with a penalty of 1.0. Internal gaps are strongly penalized (gap opening penalty 5.0, extension 1.0), because these have a strong impact on target specificity. Comparing TALEs with differing numbers of repeats requires gaps at either side of the shorter TALE, which are allowed but also slightly penalized (opening 1.0, extension 0.1), because i) the additional RVDs of the longer TALE narrow target specificity and ii) very short TALEs might otherwise align to portions of very long TALEs with a good score just by chance. Using this measure of TALE divergence, we can now assign a divergence score to each pair of TALEs.

Alternatively, we could have compared TALEs by means of their predicted targets according to computational approaches22,23,24, which, however, will change with refined algorithms and compromise a stable TALE classification. In addition, a classification by RVDs has the advantage that to some extent it reflects evolutionary relationships. In contrast, two TALEs with similar target specificity can theoretically consist of different RVDs (signifying different evolutionary origin) due to the ambiguous specificities of certain RVDs1,2,5. In addition, we did not compare TALEs based on their full-length amino acid or nucleotide sequences, because it is one of the hallmarks of TALEs that novel specificities and thereby novel physiological functions, arise frequently by recombination and repeat diversification. Therefore, a phylogenetic classification37 might not necessarily reflect TALE functionality.

Based on this pairwise score of TALE divergence, we grouped TALEs into different classes. Technically, we use a hierarchical, agglomerative clustering approach. By this means, we group TALEs together such that the average divergence score of TALEs in a common class does not exceed a threshold of T = 5.0 (see Methods). Instead of the average divergence, we could have used the minimum or maximum divergence score of TALEs in a common class. However, the former has the disadvantage of long “chains”, where finally two TALEs in a common class might only be linked by several TALEs with low pairwise divergence, whereas these very two TALEs themselves are highly divergent. The maximum divergence, in turn, might lead to a situation where a new TALE is not assigned to the class containing its least divergent peer, because other TALEs in the same class show a greater divergence. The choice of the average divergence balances between these two extremes and often gives results that conform to intuition. The above procedure is implemented in the “TALE class builder” tool of AnnoTALE, which can be applied to cluster custom sets of TALEs into classes (Fig. 3).

A universal TALE nomenclature

The first TALEs identified were named according to the resistance reaction they induce (e.g., AvrBs3, AvrXa7, AvrXa10), or to their contribution to pathogenicity (e.g., PthA, PthB, etc.). Because in principle the same resistance reaction can be triggered by different TALEs, or by TALEs that are related but carry non-identical repeat regions (e.g., different pthXo2 alleles exist38), this naming scheme is not unambiguous. Subsequently, TALEs identified in whole genome sequences were named sequentially starting from the origin (dnaA) (e.g., tal1, tal2a, tal2b, etc.28), or according to size (talA, talB, talC, etc.39). Because large genomic rearrangements are frequent in Xoo28,40 and different TALEs can cluster at similar positions, this naming scheme can and indeed already has, given the same name to different TALEs in different Xoo genomes. To rectify this situation and to assign unambiguous names to the large number of TALE sequences we expect to be discovered in the future, we have developed rules for a unified nomenclature of TALEs.

This nomenclature is based on the assignment of TALEs into classes derived in the previous step. The name of each TALE starts with “Tal”, followed by a two-character identifier of its class (e.g., XY) and the number of the TALE within its class (Fig. 3a). While this nomenclature already uniquely identifies each TALE by its name, AnnoTALE also reports the strain origin of the TALEs within a class. Using this nomenclature, Tal9e from Xoo strain PXO99A is re-named TalAD1, pthA1 from Xoo KACC10331 is re-named TalAD3 and XOO2001 from MAFF311018 is re-named TalAD2 (Fig. 3b). Using this new nomenclature, it becomes instantly clear that these three TALEs, whose previous names were completely different, are in fact closely related. In this particular example, the RVD sequences of all three TALEs are identical (Fig. 3b), suggesting that they recognize the same target sequence in the host and trigger the same reaction.

Classification of PXO83 TALEs

The assignment of novel TALEs to existing classes is based on the same divergence score that was used to build these classes initially (see above). We first compute the average divergence between the novel TALE and each of the existing classes. If the average divergence score of the TALE and the most similar class does not exceed the threshold, the TALE is added to that class. Otherwise, it becomes the first member of a new class. This assignment is implemented by the “TALE Class Assignment” tool of AnnoTALE. Considering the TALEs of PXO83 identified in the work reported here, one TALE could not be assigned to any existing class and thus became the founding member of a new class and was assigned the name TalCA1 (Supplementary Fig. S11f). All of the other TALEs show sufficient similarity to previously identified TALEs and were assigned to one or the other of the existing classes (Table 3; Supplementary Sequences).

While this manuscript was in preparation, two additional complete Xanthomonas ssp. genomes were published: Xoo PXO86 and Xoc CFBP7342 (Genbank accession nos. CP007166.1 and CP007221.1). We applied the AnnoTALE pipeline to these genomes as well, yielding a total of 62 classes for 143 TALEs. Two examples of these classes are shown in Fig. 3b. Five of the TALEs of class TalAD are completely identical and cluster with a slightly divergent TALE from Xoc (Fig. 3b). The relatedness suggests that these TALEs target the same plant genes, making it one of the few examples in which Xoo and Xoc share similar TALEs. In the second example, class TalAP TALEs from Xoo PXO99A, PXO83 and PXO86 have identical RVD sequences, while TalAP2 from MAFF311018 shows amino acid substitutions at RVDs 6, 8, 9, 14 and 19 (Fig. 3b) resulting in a divergence score of 4.0, close to the threshold of 5.0. This example demonstrates the utility of our classification to reveal similarities that are not immediately apparent.

Predicting targets of TALE classes

Ultimately, we are interested in the function of TALE genes as virulence factors. To facilitate initial analyses of putative target genes of individual TALEs and common target genes of all members of a class, we incorporated target prediction into AnnoTALE. The “Predict and Intersect Targets” tool of AnnoTALE is based on the statistical model of TALgetter23 and uses speed-up techniques of TALENoffer41. It allows us to scan the rice promoterome for target sites of all 143 TALEs in less than five minutes on a standard PC. In contrast to TALgetter, however, our scan is limited to the top 100 target sites and does not include the computation of p-values of individual target sites. For a complete overview we used the full-scale option of the “Predict and Intersect Targets” tool and analyzed the rice promoterome target of the full “Class builder” (i.e., all TALEs in all classes). The list of targets can be found as Supplementary Table S4. Applying this tool to the TALEs in all 62 classes, we find, for instance, that the four TALEs in class TalAP share 11 target genes, of which the target gene OsHEN1 of TalAP1 (PthXo8) and TalAP2 (XOO1138)2,23 is predicted on rank 1 for all four TALEs. Interestingly, OsHEN1 is also the target gene of TalAK2 (Tal1c) from Xoc BLS256, but this TALE targets a different site within the OsHEN1 promoter using different RVDs.

The current list of TALEs and their classifications can be accessed at To reserve a name for a given TALE in the course of manuscript preparation, confidential requests can be sent to the corresponding authors of this paper. Following publication, these new names will be included in our updated web-based class list that is also used by AnnoTALE.

Conservation and substitutions in RVD codon pairs

The clustering of TALEs into high-confidence classes allows new insights into the evolution of TALE genes. Considering all repeats of all TALEs, the RVDs are typically perceived as the most variable residues. However, the clustering of TALEs into classes reveals that RVDs of related TALEs are highly conserved between different Xoo and Xoc strains as well. For example, the RVDs of all TALEs in class AD (Fig. 3b) are completely conserved between the Xoo strains PXO99A, PXO83, PXO86, MAFF311018 and KACC10331, while we observe only two differing RVDs (HD at position 7 and NN at position 10) for Xoc BSL256. Conservation of RVDs is also present at the nucleotide level for the codons of the RVD amino acids. We systematically investigated which codon pairs are used to code for RVDs and found that only a small subset in fact occur. For instance, the RVD NS (Asn-Ser) can be encoded in principle by 12 different codon pairs, but only two of these occur in the TALEs considered (Fig. 4a). We make similar observations for all other RVDs, where only one to three codon pairs are used (Supplementary Table S3).

Figure 4
figure 4

Changes in RVD codons.

(a) Network of codon pairs of RVDs connected to the RVD NS by single-nucleotide or double-nucleotide substitutions. RVDs are boxed in colors according to their DNA base specificities (green: A, orange: G, red: T) and list the observed codon pairs. Frequency in known TALEs is shown below each codon pair. Lines represent possible single-nucleotide (solid line) or double-nucleotide (dotted line) exchanges. Substitutions observed within classes are in red with labels indicating the number of occurrences. (b) Histogram of the number of observed nucleotide substitutions in aligned RVDs of TALEs in a common family.

To trace evolutionary lineage, we used our TALE classification to compare codon pairs of the aligned RVD sequences of TALEs within each class, where the majority of RVDs are completely conserved between all class members, even on the codon level. We plot a histogram of the number of nucleotide substitutions in these RVD codon pairs in Fig. 4b and find that frequency decreases with increasing number of substitutions in RVD codons. In 26 cases, we observe a single-nucleotide substitution; 25 of these lead to a non-synonymous substitution of one of the AAs in an RVD and one is a synonymous substitution. However, we also observe substitutions of a greater number of nucleotides in RVD codon pairs with surprisingly high frequency (e.g., 16 occurrences of 4 substitutions in RVD-coding nucleotides). As one specific example, we consider the RVDs NS and NN in the following. The network in Fig. 4a represents observed substitutions in the codon pair of RVD NS in all TALE classes. The only synonymous substitution appears in NN-codon pairs of TALE class BF (Supplementary Fig. S11e). Single-nucleotide substitutions are present between NN and NS in two cases and between NS and NI in eight cases. Substitutions of two nucleotides arise between NS and SN, NS and HN, NS and NI, NS and NN, NS and NG, NN and NG and NN and NI.

Finally, we examined some substitutions in the corresponding class-contexts. In class AA (Supplementary Fig. S11a), we found several single-nucleotide substitutions, including non-synonymous substitutions at repeat three, where four class-members have an NS repeat and the last two an NI repeat and in repeat 19, which is an NS repeat only in the last TALE of this class and an NN repeat in the other TALEs. In contrast, class AQ (Supplementary Fig. S11d) displays substitutions of two nucleotides in repeat 5, which is NS in three TALEs and NG in one TALE. For repeat 8, which has an S* repeat in the same three TALEs and an NS repeat in the other, we observe the deletion of a complete codon. These observations suggest that one way in which TALE specificities evolve is by direct base substitutions in RVD codons.

Allocating genomic TALE cluster

TALE genes are not scattered randomly in bacterial genomes, but occur instead at somewhat conserved loci28. These loci have previously been numbered, but with a different numbering scheme for different Xoo genomes28. To simplify phylogenetic comparisons, we propose renaming TALE loci (Fig. 5). Starting with the first sequenced Xoo genome of strain MAFF311018 and then with strains KACC10331 and PXO99A, we named the TALE loci with “T”-followed by roman numbers (Table 3). The same locus name in different genomes is assigned if at least one of the flanking genes (ignoring mobile elements) is conserved. These flanking genes of TALE loci are listed in Table 4. The TALE pseudogenes in cluster T-III and T-VII group together into the TALE class AI (Supplementary Fig. S11b) which indicates that they are of common origin and that genomic rearrangements have re-located them to different sites in the Xoo genome. The TALE cluster comparisons (Fig. 5) reveal that some clusters contain different TALEs at the same conserved location in the genome (e.g. cluster T-I; TALE PthXo7 in strain PXO99A and TALE PthXo2 in MAFF31101; Fig. 5 and Table 4) a feature that has been noted earlier28. The order of TALE genes in other loci (e.g. cluster T-IX) is highly similar, but the number of TALE genes differs due to deletions or insertions.

Table 4 Definition of genomic TALE loci.
Figure 5
figure 5

Distribution of TALE genes in Xoo strains PXO83, PXO99A, MAFF311018 and KACC10331.

Individual TALE genes are represented by arrows, which indicate the orientation of the gene. TALEs are labelled with their new name inside the arrow. Below the arrows, former names or gene numbers are shown in brackets. TALEs with prominent TALEs in their class are labelled in bold above the arrows. Pseudogenes are indicated with the Greek letter phi (Φ). Novel assigned TALE clusters (T-I to T-IX) are depicted above the scheme in roman numbers. Similar colors indicate which cluster the TALEs belong to. Colored bars between the genomes represent the regions on which basis the clusters are defined. TALE clusters T-IV and T-VII are unique for their respective genome. Grey bars represent chosen similar genomic genes in different genomes that were not used to discriminate TALE clusters.

The contribution of mobile elements to type III-effector mobility and recombination is well known42,43,44,45,46 and apparently also contributes to TALE diversification40. Future work will reveal the molecular mechanisms and selective pressures that shape TALE repertoires in Xanthomonas spp. genomes.


Here, we describe an improved way to identify TALE genes in genome sequences and to group TALE proteins into classes. The classification is based on the TALE repeat region and reflects similar DNA-binding specificities. TALEs within the same class are likely to induce expression of the same target genes in the host plant. The classification provides a quick and convenient way to assess possible functions of TALEs in novel genomes. In parallel, we propose an urgently needed nomenclature for TALEs that reflects these functional relationships. This will prove useful for classifying the large number of TALEs47 in Xanthomonas spp. genomes that we envision will be published in the future.


Bacterial growth and inoculations

Xanthomonas oryzae pv. oryzae (Xoo) strains were cultivated in PSA medium (10 g peptone, 10 g sucrose, 1 g glutamic acid, 16 g agar per liter H2O) at 28 °C. Oryza sativa subsp. japonica cv. Nipponbare was used for virulence assays. Leaves of 3- to 4-weeks-old plants were infiltrated with bacterial suspensions at an optical density of 0.5 (OD600) using a needleless syringe. Symptoms were documented five days post inoculation.

DNA preparation and Southern blot

Genomic DNA from Xoo was isolated using Phenol/Chloroform extraction, digested with BamHI and separated on a 0,8% agarose gel at 90V for 24h. The chemiluminescent detection of genomic TALE sequences on Southern blot was done using a digoxigenin-labelled (Roche Applied Science, Mannheim, Germany) probe derived from 500bp of the 3′ part of talC from Xoo BAI339 that hybridizes to TALE genes.

Next generation sequencing

Libraries were sequenced in five Pacific Biosciences SMRT cells, yielding a total of 163,617 reads with a median read length of 9,669 bp. Reads were assembled in the Pacific Biosciences SMRT Portal using the HGAP_Assembly.3 pipeline. This yielded an initial assembly comprising a single contig of 5,029,672 bases with a coverage of 169.77 and an average concordance of 99.95%. Overlapping ends of this contig were trimmed and the contig was closed to a circular chromosome of 5,025,428 bp. Position 1 of the contig was determined using G/C ratio and the location of genes (dnaA, dnaN, gyrB) typically located close to the origin of replication, such that dnaA is located at position 45 in forward orientation relative to the origin.

Automated annotation of TALE genes

For the automatic annotation of TALE genes, genomic DNA sequences of known TALEs from fully sequenced Xoo (PXO99A, MAFF311018, KACC10331) and Xoc (BLS256) strains were collected and each split into the regions coding for i) N-terminus, ii) tandem repeats that are part of the binding domain and iii) C-terminus. For each of these regions, separate hidden Markov models were learnt using HMMer48 based on their DNA sequences. Domain occurrences were detected using custom Java code to avoid dependencies on HMMer, allowing for downstream analyses in one stand-alone application. Since arrays of tandem repeats are a unique feature of TALE genes, genomes were first scanned for contiguous arrays of matches to the HMM for these repeats. For each array detected, upstream and downstream regions were scanned for matches to N-terminus and C-terminus, respectively, yielding an initial region for each putative TALE gene. Initial regions were further refined by i) choosing the longest open reading frame within this region and ii) adjusting start and end of the prediction to the closest start and stop codon within the reading frame, respectively. The annotation of truncated TALE pseudogenes, e.g., due to premature stop codons or frame shift mutations, is not straightforward. Here, a conservative approach was chosen, reporting the longest reading frame between ATG and a Stop codon in the initial region. For in-depth analyses of pseudogenes, AnnoTALE provides the full-length DNA sequence of the initial region and its translation in all three frames as an additional file.

Automated domain splitting of TALEs

From a multiple alignment49 of the protein sequences of N-terminal and C-terminal regions, the full tandem repeat and the last half repeat, the corresponding consensus sequences were extracted as the most frequent AA in each column of the alignment. If in one column the gap symbol was more frequent than any of the AAs, this column was removed from the consensus. Extracted genomic TALE DNA sequences (see above) were translated to the corresponding protein sequences in silico. As reference, prototypic TALE protein sequences were built by concatenating the consensus of N-terminus, k tandem repeats, the last half repeat and the C-terminus. For each number k of tandem repeats, a pairwise alignment between the protein sequences of the given TALE and the prototypic TALE sequence was performed (BLOSUM62 substitution matrix50, affine gap costs: gap open penalty 3, gap extension penalty 2). That k yielding the best alignment score was chosen. The TALE sequence was then split into those parts induced by the alignment, e.g., the DNA sequence coding for the third repeat of a TALE was extracted as those nucleotides that code for the amino acids aligned to the third repeat in the prototypic TALE sequence. The split repeat sequences were the input of the automated extraction of RVDs in the next step.

Automated extraction of RVDs

Given one repeat sequence extracted in the previous step, the corresponding protein sequence was aligned to the consensus repeat. In the consensus repeat, the RVD comprises AAs 12 and 13. Hence, those AAs in the given repeat sequence that aligned to those AAs were extracted. For this alignment, a BLOSUM62 substitution matrix and affine gap costs with gap open penalty of 3 and gap extension penalty of 1 were used, to account for the existence of repeat of aberrant lengths35.

Computing p-values of pairwise TALE alignments/distances

Conceptually, the p-value of a TALE distance (resulting from pairwise alignment) was determined as the probability of finding the same or a lower distance by comparing a TALE against a database of random TALE sequences. For the random data base, it was assumed that TALEs are generated by independently drawing a sequence of RVDs with probabilities given by the relative frequencies of RVDs in the input data set. Since the number of possible RVD sequences grows exponentially with the length L, a naive enumeration of TALE is intractable, while a random sampling procedure would only yield approximate and varying p-values. Hence, this problem was approached algorithmically by exploiting two specifics of the given setup: Consider one position in the alignment with RVD ‘ab’ in the TALE of interest. For all possible aligned RVDs ‘de’, we can observe only four different alignment scores: 0 in case of a match, i.e., ‘ab’=‘de’, 0.2 in case of a mismatch only at position 12, 0.8 for a mismatch only at position 13 and 1.0 for a mismatch at both positions. Hence, all RVDs can be grouped by their corresponding alignment scores and the relative frequencies of all RVDs with identical alignment scores were summed up. Second different alignments can be joined. For instance, all alignments with one RVD with a mismatch at AA 12 and one RVD with a mismatch at AA 13 (at the same or any other combination of RVDs) and all alignments with 5 RVDs having mismatches only at AA 12 yield the same pairwise distance of 1.0 and their probabilities can be joined. These aggregations on the RVD and the alignment level are combined in a dynamic programming-like approach to yield exact p-values in a runtime roughly linear (instead of exponential) in the number of RVD positions in the alignment.

Building classes of TALEs

Classes of TALEs were built such that within each class the average pairwise distance between member TALEs did not exceed a given threshold T. More specifically, agglomerative hierarchical clustering with average linkage (UPGMA51) was used to obtain a tree of all TALEs. Classes were then determined by cutting the resulting tree at height T, yielding a number of sub-trees, each constituting a TALE class. This process and, hence, the number and composition of resulting TALE classes, depends on the choice of the threshold T.

To find a reliable threshold, we developed a significance measure (see above) that assigns each pairwise alignment and, subsequently, each given class of TALEs a p-value. This p-value reflects the probability that any pair of TALEs is assigned to a common class just by chance. We then test a wide range of thresholds (Supplementary Fig. S12a) for the assignment of TALEs of the initial set (Supplementary Table S2) into classes. We find that all class assignments obtain significant p-values (α = 0.01) up to a threshold of 5.5 (p = 2.9 × 10−5), while the number of classes C continuously decreases with increasing threshold. Since the p-value dramatically increases for greater thresholds (p = 0.065 for T = 5.6), we finally chose a threshold of T = 5.0 yielding C = 52 classes for the initial set of 83 TALEs (instead of C = 51 for T = 5.5,). The majority of these classes contains one or two TALEs, but we also observe classes of up to 6 members in the final set of classes (Supplementary Fig. S12b).

Additional Information

Accession codes: The genome sequence of Xoo PXO83 can be found at Genbank accession number CP012947. The program suite and a user guide for the suite can be downloaded from

How to cite this article: Grau, J. et al. AnnoTALE: bioinformatics tools for identification, annotation and nomenclature of TALEs from Xanthomonas genomic sequences. Sci. Rep. 6, 21077; doi: 10.1038/srep21077 (2016).