We used the Illumina GA2x sequencing platform to generate genome-wide sequence data for 13 isolates of Xcm available from the National Collection of Plant Pathogenic Bacteria (NCPPB). We also sequenced a further three African isolates of Xvv, a pathovar that is very closely related to Xcm but non-pathogenic on banana. We also included in our analyses the genome sequence data from Xcm NCPPB4381 and Xvv NCPPB702 that we published previously [12]. Table 1 lists the sequenced isolates and the depth to which each was sequenced. We submitted all raw sequence data to the Sequence Read Archive [18]. Note that another isolate from Tanzania, NCPPB4393, is described in the NCPPB’s catalogue as Xcm, but we previously sequenced its genome and showed that it is actually Xanthomonas sacchari [19]. We generated genome assemblies de novo for NCPPB2005, NCPPB4379, NCPPB4380, NCPPB4384, NCPPB4392, NCPPB4394, NCPPB1326, NCPPB1381 and NCPPB206 using Velvet 1.1.04 [20]. These have been submitted to GenBank [21] with accession numbers AKBE00000000, AKBF00000000, AKBG00000000, AKBH00000000, AKBI00000000,

AKBJ00000000, AKBK00000000, AKBL00000000 and AKBM00000000. The most contiguous of these assemblies was for NCPPB4384. This consisted of 84 scaffolds, of which the 12 longest scaffolds were each at least 154 Kb long and accounted for more than 2.5 Mb; that is the N₅₀ length was 154 Kb for NCPPB4384. The N₅₀ lengths for the other Xcm assemblies were 56 Kb (NCPPB2005), 146 Kb (NCPPB4379), 147 Kb (NCPPB4380), 87 Kb (NCPPB4392) and 151 Kb (NCPPB4394).

The currently used method for identifying BXW is by isolation of bacteria from the infected plant and performing fatty acid and metabolic analyses [4]. However, this approach is only appropriate once symptoms become visible, by which time it may be too late to control or eradicate the pathogen. An alternative approach, amenable to the rapid detection and identification of bacterial plant pathogens is the use of the polymerase chain reaction (PCR). A specific assay for detecting Xcm has recently been proposed based on PCR amplification of the hrpB gene [22]. However, this gene is also conserved in Xvv and this assay was unable to distinguish between Xcm and non-banana-pathogenic isolates Xvv NCPPB702 and NCPPB1326 [22]. Another study [23] generated several PCR primer pairs that were highly specific for Xcm but this study did not utilize Xcm or Xvv genomic sequence but rather used sequences from a range of other xanthomonads and so the candidate primers had to be tested for specificity by trial and error. Another recent study [24] exploited our previous [12] Xcm and Xvv draft genome sequence data to rationally design primers specific for Xcm. However, this was based on genome sequence from a single isolate of Xvv and a single isolate of Xcm. Until now, little was known about sequence diversity among isolates. Therefore, we identified a set of genes that are conserved in all of the sequenced isolates of Xcm but are absent from all the sequenced isolates of Xvv and are therefore candidates for use in an Xcm-specific PCR assay. Examples of these genes are listed in Table 2. Note that this list of genes was not generated by aligning assembled genome sequences. We aligned raw unassembled sequence reads against the previously published Xcm NCPPB4381 reference genome sequence [12] using the Burrows-Wheeler Aligner BWA [25]. This approach has the advantage of avoiding assembly artifacts and problems arising from incomplete assemblies. In our BWA alignments of raw Illumina sequence reads versus the reference genome sequence, the full length of each gene was covered by reads (depth of one or greater) from all our Xcm Illumina sequence datasets. Each of these genes has no matches (i.e., zero depth of) to any sequence reads in our Xvv Illumina datatsets (as judged from the BWA alignments).

We identified SNPs among the Xcm and Xvv isolates based on BWA [25] alignments of our Illumina sequence data against the X. oryzae pv. oryzae MAFF 311018 reference genome sequence (RefSeq: NC_007705). Based on nucleotides found at 21,525 polymorphic positions we generated the maximum likelihood phylogenetic tree shown in Figure 1. This clearly groups all of the sequenced Xcm isolates into a single distinct clade that is closely related to but distinct from the sequenced Xvv isolates.

We next identified SNPs among the Xcm and Xvv isolates based on BWA [25] alignments of our Illumina sequence data against the Xcm NCPPB4381 reference genome sequence (RefSeq: ACHT00000000). Out of 2,908,042 nt over which there was no ambiguity, 2,907,999 were invariant across all isolates; that is the Xcm genomes shared at least 99.9985% identity. Based on nucleotides found at 243 polymorphic positions we generated the maximum likelihood phylogenetic tree shown in Figure 2. This clearly delineates the Xcm clade into two distinct sub-lineages, I and II. Sub-lineage I include isolates from Ethiopia, DR Congo and Rwanda whilst Sub-lineage II includes isolates from Uganda, Tanzania, Burundi and Kenya. The two sub-lineages are distinguishable from each other by 86 polymorphic positions. These are listed in full in the Supplementary Material and in part in Table 3.

The geographic locations of the sequenced Xcm isolates are shown in Figure 3. It has been widely assumed that the outbreaks in Uganda, and subsequent outbreaks in neighboring countries, ultimately originated in Ethiopia, with the pathogen perhaps being inadvertently transmitted via international trade. Consistent with this model, the Xcm isolates from DR Congo and Rwanda do indeed show extremely high levels of genetic similarity to Ethiopian isolate NCPPB2251. However, the isolates from Uganda, Kenya, Tanzania and Burundi show a distinct genotype, characterized by the 86 consistent SNP differences. This is not consistent with a single introduction from Ethiopia into East Africa.

Our data do not exclude the possibility that the current outbreaks can ultimately be traced back to Ethiopia; it is possible that both lineages I and II are endemic there and it is simply by chance that the two available isolates happen to belong to the DR Congo/Rwanda sub-lineage I. There is an urgent need to collect a range of isolates from Ethiopia and survey their genotypes to ascertain the level of genetic diversity in this pathogen’s presumed centre of origin. Genotyping new isolates should be possible and will be expedited by developing these newly discovered SNPs into PCR based molecular markers. Similarly, there is a pressing need to survey genotypes of a much larger collection of isolates from outbreaks in all the banana growing areas to uncover the routes of geographical spread at a much higher degree of resolution. Ideally a survey of genotypes should be conducted on isolates for which precise details are available on the date and the geographic location at which they are collected.

It should also be noted that although we have categorized the isolates according to the country in which they were collected, paths of transmission may be more influenced by geographical boundaries rather than by political ones. For example, although Rwanda shares two of its borders with Uganda and Tanzania it is somewhat isolated from them by lake and forest.

All of the available isolates from Uganda were collected in 2007 from sites in the central region geographically close to Mukono where the disease was first reported and probably all represent the same single outbreak. It would be interesting to compare these with isolates from outbreaks in Kabale (near Rwanda) or Kasese (near DR Congo).

All of the available isolates from Tanzania also belonged to sub-lineage II along with those from Uganda. The disease was reported in Tanzania shortly after it was discovered in Uganda and there has been unconfirmed speculation that it may have been inadvertently carried to Tanzania by banana alcohol traders from the Buganda region, close to where the sequenced Ugandan isolates were collected. Our molecular sequence data are consistent with this but do not provide definitive proof.

Although BXW was reported in DR Congo after it was reported in Uganda, the field pictures first sent to Uganda from DR Congo, showed greater devastation. It is not clear where banana Xanthomonas wilt occurred first: DR Congo or Uganda. There is a lot of movement of people and bananas from Congo to Rwanda and back, conflicts notwithstanding, and so it is perhaps not surprising that we observe isolates from these two countries belonging to the same sub-lineage I. However, the close relationship between an Ethiopian isolate and those in Rwanda and DRC is not so easily explained unless it is by sampling bias or by one-off international travel; if disease spread was primarily determined by movement of local people and bananas between countries, then we would instead expect isolates from Rwanda, Uganda and DR Congo to cluster together.

Most of the available isolates of Xcm were originally isolated from banana. The exception is NCPPB2005, which was isolated from Ensete ventricosum in Ethiopia in 1967. This isolate clearly falls within Xcm sub-lineage I (Figure 2). This enset-associated isolate differs from the banana-associated isolates NCPPB2251, NCPPB4387 and NCPPB4389 at 67 SNP positions listed in the Supplementary Material. Some examples of these differences are listed in Table 4 and include non-silent polymorphisms in several potential virulence genes (e.g., homologues of hrpF and a gene encoding a HopW1 T3SS effector). However, since Xcm is able to infect both banana and enset [2,3] it is not clear whether these differences have any biological significance. It would be interesting to survey a much larger sample of isolates from both banana and enset to search for any significant associations between genotype and host species that might reveal adaptation.

In addition to surveying SNPs, we also searched for loss or gain of genes. By aligning sequence reads against the previously published NCPPB4381 genome assembly and systematically comparing gene-coverage in each of the alignments, we were able to identify a genomic region (GenBank: GG699410.1) that showed differential coverage among different isolates of Xcm (Figure 4). This region shows significant similarity at the amino acid and nucleotide sequence levels to two previously sequenced phage: Xanthomonas phage Cfc1 (RefSeq: NC_001396.1) [26] and Stenotrophomonas phage phiSHP2 (GenBank: HM150760.1). Specifically, two Tanzanian isolates (NCPPB4392 and NCPPB4395) appear to have completely lost at least 17 genes from this region, while Ethiopian isolate NCPPB2005 has lost 11 of the same genes. Interestingly, another isolate from the same area of Tanzania (NCPPB4394) appears to have these genes intact as do all the Ugandan, Kenyan, Rwandan and Burundi isolates. Furthermore, there is a high concentration of SNPs in this genomic region. Therefore, it seems likely that this genomic region represents the relic of a phage or similar mobile element that is in the process of degenerating, convergently, in some members of both sub-lineages.

One of the major motivations for comparing these genome sequences is to provide a resource for generating molecular markers that can be used epidemiological and biogeographical studies on a much larger panel of isolates without the need for whole-genome sequencing using, for example, the polymerase chain reaction (PCR). Therefore, we used the results of our genome comparisons to design pairs of PCR primers that can be used to distinguish between the two sub-lineages of Xcm (Table 5). A similar approach could also be taken to assay other classes of SNPs identified from the genome sequence data.

To experimentally validate SNPs, we used an approach based on digestion of PCR products with restriction enzymes. Many of the SNPs that we identified are predicted to fall within restriction sites. For example, position 6150 in RefSeq accession NZ_ACHT0100081 is a G that falls within an AluI restriction site (AG↓CT). However, in NCPPB2005 and the other members of sub-lineage I, this G is substituted for an A abolishing the recognition sequence for the AluI restriction enzyme (see Supplementary Files for a figure illustrating this SNP and several others). There are no other AluI sites in the vicinity of this SNP. We generated a pair of primers flanking approximately 250 bp either side of the SNP and amplified this 500 bp sequence by PCR. The two alleles could then be readily distinguished by digestion with AluI (see Figure 5). We used the same approach to design assays for three other SNPs (see Table 5 and Figure 5).