We compared the newly sequenced genome of F t subspecies novicida strain U112 with the published sequence of the genomes of F t subspecies tularensis strain Schu S4 25 and that of F t subspecies holarctica strain LVS (Chain and coworkers, unpublished data). Some general properties and features of the three genomes are summarized in Table 1, in which the extent of the similarity between the three subspecies is apparent. The genome of U112 is 17 kilobases (kb) larger than the Schu S4 genome and 14 kb larger than the genome of LVS. Few strain-specific regions were detected in this three-way comparison: the genome of U112 carries about 240 kb of sequences not found in the two other strains; the genome of Schu S4 carries 17.3 kb of strain-specific regions; and the genome of LVS does not contain any specific regions. The origin of replication of the U112 chromosome (around position 1) was predicted according to one of the switching points of the GC skew and by searching for DnaA-binding sequences. It is consistent with the predicted origin of replication of the chromosomes of Schu S4 and LVS, suggesting a common genome backbone for the three subspecies. The estimated nucleotide sequence identity is 97.8% between the sequences common to the U112 and the LVS genomes, 98.1% between the sequences common to U112 and Schu S4, and 99.2% between the sequence common to Schu S4 and LVS. The proposition based on physiologic experiments and DNA-DNA re-association 20 that novicida may be classified as a subspecies of tularensis is supported by the nucleotide identity between genomes.

Although no official genomic criteria exists to classify strains into species, Konstantinidis and coworkers 31 found that almost all 70 strains in their study set that reside in the same species exhibited greater than 94% average nucleotide identity (ANI). They also showed that the classification based on ANI correlates with classifications performed with 16S RNA sequences, DNA-DNA re-association, and mutation rate. In comparison, the few sequences of the other Francisella species available in Genbank, namely Francisella philomiragia, exhibit an ANI of 91.66% with the genome of U112. The ANI corroborates the proposition that novicida arose by diverging from an ancestor common to the subspecies tularensis and holarctica, and that the subspecies tularensis and holarctica subsequently diverged from a common ancestor 3132. Based on the average level of nucleotide identity between the three genomes, it is possible to estimate the rate of substitution in the genomes of holarctica and tularensis after their divergence. The genomes of holarctica strains are estimated to have evolved at an average rate of 0.55 base pairs (bp)/100 bp from the common ancestor, whereas the genome of Schu S4 diverged at the lower rate of 0.25 bp/100 bp.

A recent study using paired-end sequencing 24 indicated that the organization of the genomes of holarctica strains and tularensis strains is not conserved. However, the organization was highly similar for the genomes of the 67 holarctica strains analyzed. Similarly, the genome of holarctica strain OSU18 is collinear with the genome of the holarctica strain LVS, but it is organized differently than the genome of Schu S4 16. These findings extend the phylogenetic and molecular evidence that the strains are mostly clonal in the subspecies holarctica and that their genome is relatively stable 18323334. The subspecies tularensis can be divided into two distinct groups (type AI and AII) 1835. According to amplified fragment length polymorphism and restriction fragment length polymorphism analyses, genomes in the subspecies tularensis are organized differently but are similar within groups 3334. Hence, the genome of LVS is representative of all genomes in the subspecies holarctica, whereas the genome of Schu S4 represents genomes in the type AI group.

Sequence alignment of the U112 and Schu S4 genomes reveals 59 chromosomal segments with the same gene content and gene order in both organisms, but arranged differently throughout both genomes (Figure 1). Chromosomal segments with the same gene content and gene order in two bacterial genomes are hereafter termed 'syntenic regions'. The discrepancy in the order of the chromosomal segments between the two genomes suggests that regions have been moved, in one genome or the other. Hence, there are a total of 118 genomic breakpoints when comparing the two genomes. Similarly, 59 syntenic regions are arranged differently when comparing the genomes of U112 and LVS, and 51 are arranged differently between the genomes of Schu S4 and LVS (Figure 1), which is the same amount as found when comparing Schu S4 and OSU18 genomes 16. Twenty-eight out of the 59 syntenic blocks (47%) are nearly identical in the genomes of Schu S4 and LVS relative to the genome of U112. However, the order in which the blocks are arranged differs greatly. This suggests that these syntenic blocks formed before differentiation between both human pathogenic subspecies, but moved independently later in one or both genomes. The rest of the syntenic blocks in LVS and Schu S4, in comparison with U112, differ both in content and order (Figure 1), which suggests that they formed after differentiation of the two subspecies.

Six types of IS elements were identified in the three genomes. Five of them are present in the three genomes at least in a remnant form, whereas one, ISFtu5, is only present in the subspecies holarctica and tularensis. As shown in Table 1, the number of each IS element varies greatly in the three strains. The difference in numbers of ISFtu1 and ISFtu2 elements is particularly large. It suggests that ISFtu1 has transposed and proliferated in the genomes of the subspecies tularensis and holarctica, or in the genome of their common ancestor. ISFtu2 exhibits more proliferation in the holarctica genome. ISFtu1 appears to have been replicated essentially in the ancestor of holarctica and tularensis strains becuase 46 out of 53 elements are bordered by the same sequences in both genomes. Nine ISFtu1 elements exhibit the same bordering regions on both sides in the two subspecies genomes. However, 37 other ISFtu1 elements share only one side with an element in the other genome, indicating rearrangements specific to each subspecies. About 13 ISFtu2 elements may have transposed in the ancestral genome of tularensis and holarctica, as indicated by common bordering sequences, but have undergone subsequent rearrangements because ten ISFtu2 elements have only one common side.

These findings strongly support the proposition that genomic rearrangements occurred in the genomes of the tularensis and holarctica strains by homologous recombination at ISFtu1 and ISFtu2 elements 16. This proposition is also supported by the fact that 82% of breakpoints of LVS-Schu S4 syntenic blocks are bordered by an IS element within 100 bp (Figure 1). Similarly, 60% of the breakpoints in LVS-U112 and Schu S4-U112 syntenic blocks are bordered by IS elements in the genome of the human pathogenic subspecies (Figure 1). This lower incidence may be due to transposition of IS elements subsequent to the initial rearrangement. IS elements appear to play a prominent role in rearrangement events, further corroborating that these events took place in the ancestor of holarctica and tularensis. Indeed, 88% of the Schu S4-U112 syntenic blocks are bordered by an IS element at one extremity or both in the genome of Schu S4. On the other hand, the location of IS elements in the genome of U112 exhibits association with breakpoints for merely four ISFtu2 elements. This suggests that the IS elements did not play a prominent role in the evolution of the strains that are not pathogenic to humans.

In summary, comparative analysis using the genome of U112 revealed that the complex evolutionary scenario of the three F tularensis subspecies involves the transposition of ISFtu1 (tularensis and holarctica) and ISFtu2 (novicida, tularensis, and holarctica), accompanied by replication of these elements and genomic rearrangements at the location of these elements at distinct steps in genome evolution.

In the genome of U112, 1,731 protein-coding genes, 14 pseudogenes, and seven disrupted genes encoding an IS element transposase were identified. The coding regions (1,751,817 bp) represent 91.72% of the entire genome. Thirty-eight tRNA genes were identified, representing 30 anticodons encoding the 20 amino acids as well as three operons encoding the 5S, 16S, and 23S ribosomal RNAs and tRNAs for alanine and isoleucine. The same RNA genes and operons are found in the genomes of tularensis and holarctica. Overall, 1,813 distinct genes (excluding IS element genes and 33 hypothetical genes that we believe are noncoding) were found in at least one of the three genomes. Out of these 1,813 genes, a total of 1,572 gene sequences (functional or disrupted) are common to the three genomes. Hence, the core gene set may represent about 86.4% of all distinct genes identified in the three genomes (Additional data file 1).

In addition to this core gene set, the genomes of LVS and Schu S4 contain 41 genes whose sequences are absent from the genome of U112, and thus may play an important role in the virulence of holarctica and tularensis for humans. Thirteen are single genes found within sequences common to the three subspecies, and the remaining 28 are distributed in specific regions containing two to six genes (Table 2). Even a small number of acquired genes can cause specific differences in pathogenicity 36. It is interesting that U112 is not virulent for humans but is nonetheless able to colonize human macrophages in vitro. This indicates that the strain encodes virulence factors that are important for the infection of human macrophages but that it lacks specific factors that make human infection possible for the holarctica and tularensis strains. Hence, it is possible that some of the 41 genes that are specific to human pathogenic strains but are lacking in U112 could confer the ability to infect humans. The genome of Schu S4 contains nine additional protein encoding genes and two pseudogenes (Table 3) that are absent from the other genomes, which reduces the list of known tularensis specific genes 3738. An 11.1 kb region (FTT1066-FTT1073) has been shown to be present in all the strains of the subspecies tularensis and was named RD8 37. It is possible that some of these specific genes contribute to the greater virulence of the tularensis strains compared with the holarctica strains. In addition to specific genes, the genome of Schu S4 contains 20 duplicated genes and the genome of LVS has 34 duplicated genes, found as single copies in the genome of U112. Because they are identical copies, the duplicated genes could be responsible for a novel gene expression pattern and could therefore represent a gain of function for the human pathogenic strains.

Fourteen pseudogenes have been identified in U112 (Additional data file 1). In contrast, the original annotation of Schu S4 listed 201 pseudogenes 25. Using the genome of U112 as a reference, 53 additional pseudogenes were predicted in the genome of Schu S4 (Additional data file 1) following a procedure described in Materials and methods (see below), most of which were annotated as multiple open reading frames (ORFs) in the published genome. Because the strain LVS was artificially attenuated, it is expected to contain mutations that are not found in any other holarctica genome. Indeed, 11 pseudogene-causing mutations were found to be specific to the LVS genome 39. We ignored these 11 pseudogenes for the following comparative analysis, because they do not represent a loss of function in the holarctica subspecies as a whole.

When compared with the genome of U112, analysis of the genome of LVS revealed 303 pseudogenes in addition to those contained in IS elements (Additional data file 1). OK The number of protein encoding genes in the genome of LVS and the subspecies holarctica in general may therefore be about 1,400. The higher mutation rate observed in holarctica genomes as compared with tularensis could explain the greater number of pseudogenes. In addition, at least eight genes present in novicida and holarctica were lost by the strain Schu S4, and ten that were present in novicida and tularensis were lost by LVS. A set of 160 genes were inactivated in both LVS and Schu S4. Taking into account gene deletion and inactivation, U112 encodes 164 functions that are no longer active in both holarctica and tularensis strains. Similarly, 18 functions are specific to the strain Schu S4 and potentially to the subspecies tularensis in general (Table 3).

A total of 160 genes are inactivated in the genomes of both subspecies holarctica and tularensis. Upon alignment of their sequences, 53% of pseudogenes common to LVS and Schu S4 exhibit at least one common mutation that may have led to their inactivation, whereas 32% of the pseudogenes common to both subspecies share no common variations. The sequence of the remaining 15% is too divergent to determine a potential common inactivating mutation (Additional data file 1). This indicates that at least 53% have arisen in the genome of the human pathogenic ancestor. These 82 pseudogenes bearing common mutations are more likely to be located directly at breakpoints than the pseudogenes not sharing any common mutation (Figure 2b). In addition, the IS insertion is the only inactivating common mutation found in 19 out the 82 pseudogenes from the ancestral strain. This suggests that IS insertions or subsequent sequence rearrangements contributed to at least 22% of the earliest gene inactivations that took place in the emerging human pathogenic strain.

When directly compared with the genome of U112, most pseudogenes in the genomes of Schu S4 and LVS appear to result from small indels (1 or 2 bp) or nonsense mutations. In tularensis and holarctica genomes, genes within 1 kb from a genomic breakpoint are twice as likely to be inactivated as were genes in other genomic locations (Figure 2a). The proportion of genes that are within 1 kb from a genomic breakpoint and are inactivated is 28.5% in the genome of Schu S4 (57 out of 200), whereas the global proportion of inactivated genes is 12.6%. Similarly, 24.9% of genes within 1 kb from genomic breakpoints are inactivated in the genome of LVS, whereas the global proportion of inactivated genes is 16.3%. Figure 2a shows that, to a lesser extent, the genes within 3 kb from a breakpoint are also more likely to be inactivated than are the genes in the rest of the genome. In Schu S4, 15.4% of genes between 1 and 2 kb from a breakpoint are inactivated and 17.1% are between 2 and 3 kb. Similarly in LVS 18.8% of the genes between 1 and 2 kb from a breakpoint and 22.1% between 2 and 3 kb are inactivated. It is unlikely that genomic rearrangements could directly have caused mutations as far as 3 kb from the breakpoints. It is more likely that the rearrangements disrupted the transcriptional unit to which these genes belong. If these genes are no longer transcribed, then their sequences are no longer subjected to selection and evolve by neutral genetic drift, eventually causing the disruption of the ORF through mutation.

In agreement with this conjecture, predicted operons located at breakpoints are more likely to contain more than one pseudogene, in Schu S4 by 4-fold and in LVS by 1.4-fold. An additional argument in favor of the inactivation of some genes by genetic drift is the uneven distribution of pseudogenes across functional categories (Figure 2c). Pseudogenes and absent genes of the holarctica and tularensis genomes have been assigned to functional categories based on the annotation of their functional counterpart in the genome of U112. For example, 41.2% of the genes predicted to be involved in amino acid biosynthesis in the genome of novicida are inactivated in the genome of one or both of the other subspecies. Similarly, 43.1% of the genes predicted to encode transporters are inactivated in the genomes of holarctica and tularensis. Remarkably, the distribution in functional categories is the same for genes inactivated in one genome and those inactivated in both. Likewise, it was previously observed in the genomes of Salmonella typhi and S paratyphi that the pseudogenes were different but appeared to belong to the same pathways and operons 11. The over-representation of pseudogenes in certain functional categories suggests a loss of function associated with specific pathways, resulting in the decay of multiple genes in these categories 40. Following the disruption of a biologic process by the inactivation of one gene, other genes involved in this process are no longer subjected to selective pressure.

This example illustrates the proposed model of evolution of Francisella human pathogenic strains: initial inactivation of a gene in the ancestor of the subspecies tularensis and holarctica (potentially pathoadaptive) and further gene inactivation in regions no longer subjected to selective pressure before and after subspeciation.

In the genome of U112, the genes involved in leucine and valine biosynthesis are organized in two operons: one contains leuB, leuD, leuC, leuA, and ilvE; and the other one contains ilvD, ilvB, ilvH, and ilvC. All genes are expressed in rich medium (Rohmer and coworkers, unpublished data). In the tularensis and holarctica strains the leucine, isoleucine, and valine biosynthesis pathway is inactivated. Based on the organization of the two regions depicted in Fig