We obtained the draft genome sequence of V. fischeri strain MJ11 and, as part of our initial analysis, we conducted a number of reciprocal BLAST analyses to compare its predicted proteome with that of the completely sequenced conspecific strain ES114 16. We used BLASTP 24 to identify orthologs between the two strains, using a modified reciprocal best-hit approach as outlined in the Methods. A surprising outcome from this analysis was the occurrence of over seventy protein-coding genes in MJ11 with reciprocal best-hits to two neighboring genes in ES114. At the time that we were performing this analysis, a handful of cases were being identified empirically in which neighboring genes in ES114 were actually one gene, and that the appearance of two genes resulted from frameshift or nonsense errors in the original sequence data. Examples that were identified independent of this work include ptsI 25, fnr (J.L. Bose and E.V.S., unpublished data), and acs (S.V. Studer & E.G.R., unpublished data).

Analysis of the suspect regions supported the hypothesis that there were a large number of loci in ES114 in which sequencing errors had led to the miscalling of one gene as multiple ORFs. In support of this hypothesis, we identified a number of genes that are essential in Escherichia coli and other bacteria, but that were split in version 1.0 of the ES114 sequence. These included dnaG, ftsQ, mukB, nusG, rplC, rplN, rplO, rpoB, rpoC, thrS, and tilS 2627, and the conditionally-essential rpoH 28. Second, we identified eleven ambiguous bases (i.e., "N" listed in the nucleotide sequence) that had been called in the original sequence, and the incidence of these bases correlated with the presence of suspect ORFs.

In addition to suspected frameshifts and substitutions, we also identified fourteen regions in which it appeared that extraneous sequence had been incorporated that was highly similar to neighboring sequence, with the size of the duplicated/extraneous region ranging from 318 bp to 1264 bp. In one case, pre-genomic sequencing of a suspect region did not identify any repeated sequence 29. Therefore, we hypothesized that these regions represented assembly errors in which the same stretch of DNA was mistakenly incorporated twice into the genome's sequence. These regions typically contained a few unique base pairs at either end – likely due to low-coverage sequencing – that led to the misincorporation, but were otherwise essentially a direct repeat of DNA that had the effect of introducing extra and/or truncated ORFs.

A list of the loci targeted for resequencing was assembled and each was assigned a "target number"; that number is used consistently in tables and figures so that the primer sequences used to analyze the data may be correlated with the resulting sequence and analysis.

In addition to the BLASTP-based identification of potential errors, we undertook a full-length visual comparison of the chromosomes of V. fischeri ES114 and MJ11. Given the prevalence of errors detected by identifying adjacent ORFs that likely represented a single ORF, we hypothesized that there were probably other cases of errors that would not have manifest themselves in this way. Examples of other suspected errors that warranted investigation included situations in which one of the fragmented ORFs was too small to be detected as an ortholog candidate by the BLASTP filters, or in which the second fragment did not lead to a predicted open reading frame. Using the program Mauve 30, we analyzed ORFs along the length of the chromosomes, identifying candidates that had suspect 5' or 3' ends. In some cases, these appeared to result solely from annotation differences, in which identical sequences had predicted translational start sites (5' boundaries) that were called at distinct points in the two annotations. In other cases, sequence differences underlay the unique ORF boundaries, and we targeted those for our analysis. Furthermore, there were three cases in which putative extraneous sequence was visually identified in intergenic sequence, which could not have been detected by BLASTP analysis in the absence of annotated ORFs (target nos. 130, 172, 178). These cases were added to the list of targeted loci. Finally, any remaining ambiguous bases in the sequence were targeted for resequencing.

We examined a total of 82 targets for resequencing. Our general approach involved amplifying across the target, and then sequencing the amplified product with the PCR primers. In cases where we were clarifying the sequence following a large detected "deletion" (missing sequence from what is predicted from the published sequence), we amplified a larger product and sequenced from a set of sequencing primers across both strands. For the oligonucleotide primers used for PCR and sequencing see Additional file 1. With one exception, all of the primer pairs amplified products in which there was a clear, predominant band, and thus served as satisfactory templates for sequencing. The primer pair that failed to amplify (target no. 180) included a primer that was in a region that does not exist in the true ES114 sequence, as clarified by our analysis of target no. 182. Therefore the absence of a band in this case supports the deletion that emerged from target no. 182.

Seventy-five of the 82 sets (91%) of resequencing targets examined were found to be in error in the published ES114 sequence. The errors, subsequent changes, new locus tags, and new annotations, are listed in Table 1. Conceptual diagrams of representative sequencing and other annotation changes discussed during this report are illustrated in Figure 1. Note that with this update, the locus tag format has been modified to the new NCBI format for locus tags (underscore following the "VF" prefix, which denotes V. fischeri ES114). As a convention, in cases of gene fusion, the locus tag of the 5'-proximal (N-terminal-encoding) fragment retained its locus tag identifier, while the identifier(s) for the remaining gene fragment(s) were deaccessioned.

It is thought that the creation of false large-scale genomic rearrangements such as insertions rarely occurs in microbial genome projects 311; however, we confirmed the presence of all fourteen predicted insertions by amplifying from the respective unique flanking regions, and demonstrating that the bands obtained are inconsistent with the previous sequence model (Figure 2). In each case, the bands observed were smaller than predicted, and the sequence obtained led to the precise deletion of the extraneous repeated DNA in the new model.

Most of the resulting changes led to the fusion of two – or in some cases three – neighboring ORFs, and/or the extension of ORFs at the 5' or 3' end (Figure 1A–C; Table 1). In one case (target no. 178), the deletion affected only an intergenic region that contains no annotated features. In another case (target no. 172) the deletion identified by visual analysis in Mauve affected what was believed to be a 1498-bp intergenic region between rluE and VF_1777. The corrected sequence revealed this region to be only 368 bp in length, but also that it contains a predicted lipoprotein conserved in V. fischeri MJ11. The new release reflects the sequence deletion, as well as the added annotation for this gene (VF_2633).

The resulting sequence corrections led us to propose a number of protein annotations that were consistent with our predictions. Based on the corrected sequence, many conserved genes now more closely resemble their orthologs in other species. In other cases, the domain structure of even poorly characterized proteins supported the accuracy of the corrections. For example, target no. 185 extended the 3' end of VF_1515 by correcting a frameshift mutation. Analysis of protein domains by conserved domain search (CDD; 31) identified an incomplete GGDEF (diguanylate cyclase) domain in the protein's C-terminus, and correction of the frameshift led to inclusion of the entire domain.

In some of the cases we confirmed the published ES114 sequence to be correct, and that the ORF boundaries (5' or 3' end, or the presence of two genes instead of one) were correct in ES114 version 1.0. Table 2 lists those five cases that we can now more confidently assume to be pseudogenes in ES114 because they appear to be nonfunctional given their predicted amino acid sequence. In each case, the indicated defect is predicted to interrupt a significant portion of the coding sequence required for function in well-characterized homologs. The N-acetylglucosaminyltransferase VF_A0466 has two (apparently functional) paralogs in the genome, and ES114 is capable of utilizing N-acetylglucosamine as a sole N+C source (data not shown): therefore, the appearance of a pseudogene at this location does not have obvious functional consequences for the cell.

There is little information about the remaining four pseudogenes, except for umuC. The transcriptional organization between the genes encoding the DNA polymerase V subunits umuD and umuC is conserved between ES114 and MJ11 (Additional file 5). However, umuC has uniquely degenerated in ES114, with both a nonsense codon and a 5-bp repeat expansion following the nonsense codon. DNA polymerase V is responsible for error-prone translesion synthesis (e.g., following UV-irradiation), which allows DNA synthesis to proceed despite a high rate of error incorporation 32, yet there are organisms, including V. cholerae El Tor, that apparently do not encode these functions 3334. Whether the situation in ES114 represents an evolutionary transition state, or instead this arrangement (umuD⁺umuC^-) has relevant functional implications remains to be determined.

Because examination of the intergenic region corrected by target no. 172 revealed a likely protein-coding gene, we asked whether there were other genes present within the ES114 sequence that were previously unannotated. Additionally, regulatory RNA genes had not been previously annotated in the V. fischeri genome, yet they are known to play important roles in V. fischeri and other diverse bacteria 3536. Therefore, we undertook an effort to systematically identify ORFs and regulatory RNA genes that had not been called in the published version 1.0 sequence.

To accomplish this search we took advantage of the annotations present in the J. Craig Venter Institute's Comprehensive Microbial Resource (JCVI CMR), which include ab initio gene-calls that can differ from those in the deposited GenBank flatfiles 37. We examined the list of approximately 150 novel genes identified in the CMR. We excluded candidates that were unlikely to be biologically significant – generally, short ORFs that were encoded on the opposite strand against much larger ORFs – and were left with 53 likely novel ORFs (Additional file 4, Basis code "A"). We also took advantage of the presence of the closely-related MJ11 strain as a source for novel gene annotations. Of the MJ11 proteins that did not have an ortholog in ES114, we examined those in which a TBLASTN query of MJ11 proteins against the ES114 genome yielded a percent amino acid identity value of at least 85%. We excluded candidates in which there was low biological basis for the assignment (as described above), or in which the open reading frame was not conserved in ES114. There were 72 novel ES114 ORFs assigned by comparison with MJ11 (Additional file 4, Basis "B"). 30 ORFs were called by both methods (CMR and MJ11 conservation), whereas 65 genes were called by only one method, for a total of 95 uncalled chromosomal protein-coding genes that we predict to be uncalled in ES114 (Table 3, Additional file 4).

In all of these cases, no sequence was changed in the genomic model, but annotations imposed on the sequence were added. Included in the list of new genes predicted from both approaches is biotin synthase (bioB). Because ES114 grows on minimal medium lacking biotin 38, BioB is likely expressed by the organism. Another example of a gene predicted from both approaches is the oxaloacetate decarboxylase gamma subunit (oadC), which is predicted to be encoded in an operon with the already-predicted alpha and beta subunits (new operon prediction of oadCAB).

In addition to genes that were identified from both the MJ11-comparative and JCVI CMR approaches, we believe that genes identified by only one of the approaches are still worthy of inclusion, subject to the filters imposed above. Genes that were identified by comparison with MJ11 have the support that the open reading frame is conserved in at least these two strains. A similar measure has been used to call genes in Saccharomyces cerevisiae for genome inclusion 39. Genes that were identified solely from the JCVI CMR annotation include a number of regions that are unique to ES114, such as a prophage that is present in ES114 and absent in MJ11 (Additional file 4, new loci VF_2640 through VF_2649), and therefore would not be expected to be called by comparison with MJ11. The coding density of the prophage was markedly increased due to the addition of the novel gene annotations, consistent with phage genome organization and supporting the assignments predicted by the CMR. It is clear that the consolidated results from both methods, though partially overlapping, identify a significant number of novel, bona fide gene annotations in ES114.

We added regulatory RNA genes to the annotation as identified from multiple sources (Table 3, Additional file 4). Prediction of CsrB regulatory RNAs has been pioneered using V. fischeri 36, and additional regulatory RNAs were added based on motifs found in the RFAM database 40. These methods identified a total of 10 regulatory RNAs and 1 operon leader peptide. Although operon leader peptide predictions are not typically found in databases, we speculated that additional such genes were present in ES114 and, using the Ecocyc database 41 as a guide, we manually searched for operon leader peptides in ES114 and identified five additional high-confidence members in the genome (Additional file 4, Basis "E").

In total, we called 95 new protein-coding genes, 8 regulatory RNAs, and 6 operon leader peptides, and we incorporated 2 protein-coding genes and 2 regulatory RNAs that were published previously, for a total of 113 new annotations incorporated into ES114 version 2.0 (Table 3, Additional file 4).

In the process of correcting sequence errors and adding missing annotations, we additionally took the opportunity to update the annotations of the genes in the ES114 genome and to establish a framework for future genomic and genetic studies in V. fischeri. To update the product annotations of V. fischeri, we assembled a database of V. fischeri genetic and genomic analyses from the the PubMed database 42. Our initial curated V. fischeri list included 545 unique gene-publication associations from 60 publications, encompassing 339 distinct genes represented in strain ES114. This list served as the core of the reannotation effort, which further gave us the opportunity to update a number of genes whose functions have been discovered since the initial genome publication.

For all genes in ES114, we additionally compared protein annotations from multiple sources: (1) Orthologous protein annotations in the recently reannotated Escherichia coli K-12 MG1655 sequence, and updates made subsequent to sequence publication through the ASAP and Ecocyc databases 4143; (2) Orthologous protein annotations in V. cholerae 34; (3) the JCVI CMR 37; and (4) UniprotKB 44. These comparisons allowed us to update the annotations and to make the annotations more consistent with current practice and NCBI guidelines.

We found the annotations of E. coli – though most distant phylogenetically – to be the most valuable empirically. The timeliness of the update and the availability of curated, referenced descriptions in the Ecocyc entries allowed us to improve a number of entries that appeared to lag behind the other data sources. As one example, we point to the case of yihY (VF_0100, ortholog of E. coli locus tag b3886). Previously annotated as encoding the ribonuclease BN 45, this annotation has been propagated through numerous sources, including most of the Vibrionaceae genomes. A subsequent report identified the E. coli rbn/elaC gene (locus tag b2268) as the gene that encodes RNase BN, and the most recent genome annotation for b3886 has been updated as yihY, "predicted inner membrane protein" 46. We compared data from the sources described above, as well as the literature described, and captured this update by calling VF_0100 as yihY with a product of "predicted inner membrane protein". In fact, V. fischeri, like most sequenced Vibrio spp., does not contain an rbn ortholog, and therefore having any product labeled as "ribonuclease BN" would have been misleading from the perspective of predicting genome capabilities. We note that the old annotation persists in major databases 31474849 and in most of the Vibrionaceae genomes available at the time of data submission. This example highlights the value and relevance of the E. coli K-12 update to this and related annotation projects, as well as our ability to capture the latest information about genes encoded in the ES114 genome.

Fine-scale annotation changes, such as those shown in Figure 1F, are detailed both in that figure and in the Methods. We also wish to highlight the updated entry for prfB (noted in Table 3), the peptide chain release factor RF-2. The programmed frameshift in prfB is not called correctly by machine-call algorithms, and this gene is improperly entered in the GenBank flatfiles of all of the previously submitted Vibrio spp. genomes.

With the blossoming number of sequencing projects, utilization of locus tags (e.g., VF_0001) as identifiers for both genes and their products has become commonplace as the increase in genomic characterization has outpaced genetic and biochemical characterization of gene products. Nonetheless, biological analysis in a genomic context depends on understanding gene function, and proper nomenclature has been adopted in a number of species to facilitate meaningful communication about genes and their products. In fact, we (and others) repeatedly refer to genes by their identifiers and, without tracking in a database, this practice can lead to incorrect conclusions 50. Therefore, whereas the previous ES114 version did not contain 3–5 character "gene" identifiers, we added those for approximately 1,995 genes in which the identity of the gene could be identified or inferred from published work in V. fischeri, or by orthologous genes in other organisms. Due to the availability of well-curated database resources, most of the names were derived from their orthologs in E. coli MG1655 414351.

We demanded that unique gene identifiers be a minimum of three lowercase letters (e.g., fnr), with an optional uppercase letter (e.g., dnaA), and/or an optional numeral (e.g., nagA2), for a maximum of five characters total. Such numeric suffixes were assigned to distinguish among members of paralogous families or genes of related function. For approximately half of the genes, no gene identifier could be assigned at this time.

Because three RNA polymerase subunit genes were affected by the resequencing (rpoB, rpoC, rpoH), we took a genomic inventory of the corrected ES114 transcriptional apparatus in a manner that was not possible prior to the targeted resequencing. The subunits identified in the genome are listed in Table 4 and include the core subunits α, β, β', and ω. Classification of the eleven identified sigma factors is described below and was performed by the scheme outlined in Gruber & Gross 52.

Group 1 sigma factors include regions 1.1, 1.2, 2, 3, and 4. This category includes only σ⁷⁰ in ES114. Group 2 sigma factors (regions 1.2, 2, 3, 4) include the closely-related σ^S subunits; as mentioned above, V. fischeri curiously contains two of these sigma subunits. In addition to a clear ortholog of rpoS (VF_2067), the gene encoding the stationary-phase sigma subunit (σ^S), ES114 also contains a gene that is expected to encode a σ^S-like subunit (VF_A1015). Transcript levels of this second σ^S-family subunit increase upon C8-homoserine-lactone (AinS-dependent) quorum-