Generation of the mutant library
We generated transposon insertion libraries in P. gingivalis using a Himar 1 Mariner mini-transposon system created for use in Bacteroides thetaiotaomicron[30]. The B. thetaiotaomicron promoter of BT1331 that drives expression of himar1c9a transposase is recognized by P. gingivalis, allowing us to use the B. thetaiotaomicron plasmid vector pSAM_Bt with modifications in growth media and antibiotic selection conditions. This mini-transposon is constructed with two translational terminators downstream of the gene for antibiotic selection, thus eliminating read-through downstream from the insertion.
We performed mutagenesis using pSAM_Bt with P. gingivalis strain ATCC 33277. The 4.6 kb pSAM_Bt vector containing the Mariner mini-transposon cannot replicate in P. gingivalis and, in addition, the plasmid lacks sequence homology with the P. gingivalis genome. Therefore, after the plasmid enters P. gingivalis by transformation, transposition from the plasmid into the genome occurs without significant background insertion of the plasmid into the genome by illegitimate recombination. This system allows for single, stable transposition events since transposase activity is lost along with the plasmid. We collected 54,000 transposon insertion strains (individual colonies) from six separate transformation experiments. Variable colony sizes were observed among the mutants harvested and pooled following 14 days of growth. However, the majority of macroscopically visible colonies were similar in size to those of wild-type P. gingivalis strain ATCC 33277 after 14 days of growth. To confirm that the strains contained transposon insertions and not cryptic or full plasmid integrations, we performed PCR specific for the transposon (ermG) as well as for two portions of the vector backbone (himar1c9a and bla) (Figure 1A). Of 100 colonies that were screened, all showed positive PCR reactions for the transposon gene and negative reactions for the vector backbone, indicating ‘correct’ transposition. ‘Incorrect’ transpositions can include portions of the vector backbone inserting with the transposon, the vector being stably maintained within the bacterium extra-chromosomally or multiple insertions within the same genome; such transposition events were not detected in the subset of mutants tested (Figure 1). To determine whether the transposon inserted into different genes and not preferentially into genetic ‘hot-spots’, we performed nested semi-random PCR followed by sequencing which confirmed that insertions occurred in multiple locations across the genome (Figure 1D) [39]. This traditional sequencing method is effective for targeted sequencing a subset of mutants from the mutant library if massively-parallel high-throughput are neither desired nor necessary.
Validation of Tn-seq of the P. gingivalis library
Having confirmed via nested semi-random PCR and subsequent sequencing that the libraries contained different transposon insertions scattered throughout the genome, we identified the location of each insertion in the library by Tn-seq analysis [27, 28]. This method couples transposon mutagenesis with massively-parallel, next-generation sequencing technology to identify the location of each insertion and quantitate the relative abundance of each insertion mutant in the library.
Prior Tn-seq experiments using Mariner libraries have taken advantage of an engineered Mme I restriction site that cuts 18–20 base pairs away from the recognition site into the genomic DNA [27, 30]. This method has been successfully employed to evaluate library sequences in a variety of settings, however, it suffers from a number of disadvantages including: 1) Yielding small sequencing reads limited to 16–18 nucleotides in length. 2) Requiring the use of a mutant transposon and hence existing transposition vectors must be mutated. 3) MmeI generates 2 base pair 3’ overhangs at adjacent sequences to which adapters are ligated. It is possible that the enzyme cleaves these adjacent sequences with varying efficiency. Furthermore, T4 DNA ligase is likely to join adapters to these varying overhangs with differential efficiency (for instance GG should be more efficient than AA). Such variations in efficiencies, if they exist, would lead to unequal representation of sequenced insertions. An alternate method for sequencing from junctions in transposon libraries involves the ligation of adapters to sequences near transposon junctions [29]. However, the method is labor intensive, requires gel purification of ligated products, and is prone to the unintended creation of inhibitory adapter dimers and other side products.
Here we report a new method, without the abovementioned disadvantages, for the construction of high-throughput sequencing libraries from transposon, retrovirus or repetitive element insertions sites in any genome. For details see the Materials and Methods section. In brief, genomic DNA containing the insertion element of interest is sheared, and then Terminal deoxynucleotidyl Transferase (TdT) is used to add an average of twenty deoxycytidine nucleotides to the 3’ ends of all molecules. Two rounds of PCR using a poly-C-specific and an insertion element-specific primer pair are then used to amplify one of the two insertion element-genomic DNA junctions and append all user-defined sequences needed for high-throughput sequencing and indexing. This new method does not require a ligation reaction, does not produce adapter dimers, does not require gel purification and is compatible with long sequencing reads the size of which is only limited by the length of library fragments and the sequencing technology. Here, in contrast to the 16–18 nucleotide reads obtained with the MmeI method we used 50 nucleotide reads allowing for significantly more effective and precise mapping of sequences to regions with nucleotide repeats as well as genes that contain nucleotide homology (Figure 2A). This is particularly important given that the current Illumina HiSeq2000 base-calling algorithm gives poor quality scores for the first few bases (Figure 2A).
Two replicate samples derived from the same master mutant library, but processed separately for sequencing, were compared. Sequencing revealed 35,937 and 35,732 distinct insertions (mutants) respectively (Figure 2B). Of the total insertions, 7,230 and 7,193 in the respective replicate runs were in putative intergenic regions. After quality filtering sequencing reads an average 6,310,573 reads could be attributed to an average of 35,835 insertions mapped to the genome. Of note, during multiplexed Illumina sequencing runs between 10–20 percent of sequencing reads are ‘thrown out’ during quality control analyses. This level of ‘discarded’ read data is seen by all groups performing permutations of Tn-seq, RNA-seq, ChIP-seq and other massively-parallel adapted methods. Sequencing data removed during our quality control analyses was within the 10–20 percent range previously noted. The number of insertions per gene and the number of reads per gene when comparing the technical replicates gave R2 values of 0.989 and 0.998 respectively (Figure 2B). The similarity between the two technical replicates demonstrates that aliquot production from the master library, processing of the samples as well as sequencing and analyses are highly reproducible.
Identifying putative essential genes of P. gingivalis by Tn-seq
The genome of P. gingivalis strain ATCC 33277 comprises 2.35 Mb of chromosomal DNA and no plasmids. With a GC-content of 48.4%, there are 2,155 genes, 2,090 protein-coding sequences, 53 tRNA, and 12 rRNA [GenBank: AP009380.1] [9, 40, 41]. An important factor for Mariner transposition is that the genome contains 117,742 informative ‘TA’ sites, the only known specific motif ‘required’ for Mariner transposition. In previous studies and in agreement with our sequencing results, approximately 98% of Mariner insertions occur at TA sites (unpublished results).
The presence in our library of a mutant bacterium, in which a gene or intergenic region has been interrupted by insertion of the Mariner transposon, would indicate that it is unlikely that the gene or region is essential for growth in vitro on blood agar plates, provided that the insertion was likely to have inactivated the function. Insertions into the first or last five percent of a gene have a higher likelihood of generating a functional gene product relative to insertions in the middle portions of a gene, therefore these mutants were eliminated from consideration. In addition, we required a minimum of three sequencing ‘reads’ of the mutant locus be present in both technical replicates to exclude nonexistent insertions introduced by mapping of incorrectly sequenced reads, and lower rates may be due to mis-assignment by the reference assembly software. By these criteria, we determined that 1,639 genes are non-essential for growth in vitro. Sixteen of these genes contained 100 or greater insertions, notably the proteinases/adhesins kgp (310), rgpA (300), rgpB (152) and hagA (340) as well as the minor fibrilin mfa1 and four of the twelve 23S rRNA genes. Eighty-eight genes contained 50 or more insertions and 837 contained 10 or more insertions, with a median number of 10 insertions per gene. There is a direct, but not completely exclusive, correlation between number of insertions and sequencing reads as 9 of the top 10 highest reads from the library belong to genes with more than 100 insertions; thus kgp, rgpA, rgpB and mfa1 are in the top ten most-read genes. Nine hundred and twenty genes had transposon insertions in at least 25% of their reported TA sites, while a remaining 528 genes had insertions in at least 10% of their reported TA sites. The average number of TA sites per gene, when including all 2155 genes (CDS, tRNA and rRNA), is 55. A total of 87 genes were fully saturated with at least one mutant insertion into every available TA site in the gene. Full saturation results in a TA insertion ratio (actual number of different insertions into the TA sites of a gene divided by the theoretical maximum number of different insertions into the TA sites of a gene) of 1. A TA site to insertion ratio of greater than 1 demonstrates that at low frequency Mariner will insert into non-TA sites, most likely due to medium composition such as salt concentration that alleviate transposon specificity, local DNA structure, nucleotide composition and/or DNA-binding proteins. Of the 87 fully saturated genes having on average have 49 TA sites, 64 are present in multiple copies throughout the genome (Additional file 1: Table S1) and include rRNA genes, ISPg1, ISPg3, gingipains, and hypothetical proteins (Additional file 1: Table S1). All of the rRNA genes (12 in total) are located in spatially separated clusters of three and are fully saturated. Efforts are currently underway to determine whether there is conservation among non-TA insertion sites.
For the remaining putative essential genes we applied the following rules to identify those most likely to be essential for in vitro survival. The rules are similar to those used in previous essential gene analyses of other bacteria and contend that [27, 29, 30, 42]: 1) A gene must contain at least 10 TA sites. Genes with less sites could be under-inserted due to random chance. Of the 204 genes in the P. gingivalis ATCC 33277 genome with less than 10 TA sites, 189 (93%) are annotated as hypothetical. 2) Genes found to have an actual to theoretical insertion ratio of 50-fold or greater under-insertion were considered putatively essential (actual:theoretical ≤ 0.020). Applying these rules, out of a total 2,102 genes in the ATCC 33277 genome (all protein coding sequences and rRNA genes combined minus the 53 tRNAs), we identified 463 (22.0%) genes as putatively essential for in vitro survival (described below) (Figure 3) (Additional file 1: Table S1). Twenty-two percent of a 2.35 Mb genome containing 2,102 genes is within the range of essential genes determined by transposon mutagenesis, single gene deletions and in silico analyses of other bacterial genomes, as described below [29, 30, 32–34, 42–51].
Prior to applying any cutoffs described above we found that 273 putative essential genes contained zero insertions. Given that these genes were a minimum of 200 base pairs in length and contained at least 10 TA sites the confidence level for concluding these as essential is high. Of the remaining 190 putative essential genes, 64 were found to have a ratio of between 0.001-0.010, 100-fold or less under-inserted, and 76 had a ratio between 0.010-0.020, 50-fold or less under-inserted. In most cases these genes had a single insertion over a gene length of 1.5-3.0 kb. Fifty genes had ratios between 0.020-0.050, however, these insertions were found to fall under the constraints outlined above and also met our qualifications for putative essentiality as well. Of note, of these 50 genes the majority (72%) have homology to genes of other bacteria identified in previous essential gene studies [35, 37].
In addition to identifying the essential nature of a gene, more detailed analysis, specifically mapping domains of proteins and intergenic regions, can provide valuable information about protein functional domains, promoter regions, mis-annotations, operon structure and regulatory RNAs (Figure 4/Figure 5). Simply mapping the insertions onto the genome to view saturation of specific genes provides a qualitative understanding of library complexity (Figure 4A). Annotations of genomes identify gene/coding-sequence start and stop codons, spatial relationships to other genes, operon structure, number of possible amino acids and amino acid composition. Such bioinformatic analyses are not perfect because they are based on coding-sequences from model organisms, e.g. Escherichia coli, and not adapted to less well-known bacterial species. Detailed insertion mapping allows for the determination of essential genes on a visual scale (Figure 4B). In addition, transposon mutagenesis mapping may reveal previously mis-annotated start and stop sites for genes as well as putative internal start sites, providing information on potential operon structure. Furthermore, essentiality of function domains can be determined (Figure 5A/5B). Although intergenic regions are far less abundant in prokaryotic genomes, mapping of insertions, or a lack thereof, to a specific intergenic region within the genome can provide insights on regulatory features within non-coding DNA sequences.
Comparison of P. gingivalis essential genes to core genome and transcriptome
The core genome of P. gingivalis previously proposed by Brunner et al. was derived from hybridization analysis of 10 different strains to a DNA microarray of annotated genes from strain W83 [38]. Of note, any gene ‘missing’ from strain W83, even if present in all other strains of P. gingivalis, would be considered not part of the ‘core’. Since both strains W83 and ATCC 33277 are now fully sequenced, it is known that 8 genes in the P. gingivalis ATCC 33277 essential list are missing in W83. Five of these genes have been identified in a third sequenced and annotated strain, TDC60. There was nearly complete overlap between putative essential genes determined by Tn-seq with the P. gingivalis core genome, 434 of 463 (93.7%) putatively essential genes overlapped (Additional file 1: Table S1) [38]. Also, several gene probes were left out of the core genome analysis due to low hybridization signals or redundancy; two of these are identified as essential in our study. Nearly half (12 out of 31) of the essential genes not found in the P. gingivalis core genome had BLAST matches in the Database of Essential Genes (DEG) [35, 37]. The remaining small difference may be explained by the hypothesis that certain essential genes are strain- and not species-specific, and thus may not be identified in a core genome analysis. In the circular genome representation of the base genome P. gingivalis ATCC 33277 (Figure 3) essential genes are depicted in arrows denoting directionality (blue) and homologous coding sequences are shown as tick marks in strains W83 (red), W50 (green) and TDC60 (blue). The map also shows that areas of genetic aberrance between P. gingivalis strains are areas devoid of essential genes (Figure 3). This would be hypothesized as essential genes should be conserved throughout a species unless duplication or gain-of-function mutation occur that can rescue the essential role of a give gene. As more P. gingivalis strain genomes are sequenced, bioinformatic analyses that provide mapped read-outs will delineate putative essential, core and accessory genetic regions, thus giving insight into strain-based differences within the species. Such differences may be useful to identify strain phylogeny and aid in clinical treatment regimens based on knowledge of genotype-to-phenotype virulence attributes (eg. antimicrobial resistance and gene transfer).
Chen et al. performed RNA-seq analysis of mRNA expression by P. gingivalis strain W83 from which 455 of the possible 463 ATCC 33277 essential genes were assayed [53]. This analysis demonstrated that 452 of 455 P. gingivalis ATCC 33277 essential genes were expressed during growth on blood agar medium (Additional file 1: Table S1). The 3 genes not expressed on blood agar plates as determined by RNA-seq are annotated as ‘hypothetical’ proteins. Transcriptome analyses were also performed on P. gingivalis grown on minimal (MIN), tryptic soy (TSB) and blood agar (BA) media, however, no essential genes were expressed solely on BA and not TSB or MIN despite some differences in levels of expression between the three media.
Comparison of P. gingivalis essential genes with other essential gene analyses
Of the 463 putative essential genes in P. gingivalis, 364 (78.6%) have known essential gene homologues determined by BLASTP interrogation of the DEG (http://tubic.tju.edu.cn/deg/), version 6.8, updated on November 4, 2011 (Additional file 1: Table S1) [35, 37]. The DEG curates a searchable list of “Essential genes [that] are those indispensable for the survival of an organism, and therefore are considered a foundation of life”. P. gingivalis essential genes were determined to have DEG homologues based strictly on BLASTP similarity. BLASTP similarities that resulted in e-values of 1x10-8 or less were considered matches. Homologies were found in at least one of the following species which had previously undergone essential gene studies: Bacillus subtilis, B. thetaiotaomicron, E. coli, Francisella novicida, Haemophilus influenzae, Helicobacter pylori, Mycobacterium tuberculosis, Mycoplasma genitalium, Mycoplasma pulminous, Saccharomyces cerevisiae, Salmonella Typhimurium, Staphylococcus aureus, Streptococcus pneumoniae and Vibrio cholerae[29, 30, 33, 34, 43, 45–47, 50, 51, 54–61]. For more than half of the 364 BLAST-matching essential genes there was homology within two or more species. In cases where only one other species contained a BLASTP match to a P. gingivalis essential gene it was most frequently to a gene in B. thetaiotaomicron, H. influenzae or H. pylori, which are the most closely related species to P. gingivalis both based on phylogeny and ecology.
The remaining 21.4% of putative essential genes that have no known homologue in the DEG may be essential in a species-specific or niche-specific manner. These 99 genes, many of which are functionally classified as containing known Pfam protein motifs, ‘conserved domains’ or ‘hypothetical’ proteins, may reveal important aspects related to metabolism and physiology of Porphyromonas species and closely related organisms. Of the 46 annotated as hypothetical proteins, 42 are among the 99 P. gingivalis essential genes not previously known to be essential from other studies.
Of the organisms for which an essential gene set has been identified, H. influenzae, F. tularensis, Acinetobacter, M. tuberculosis, Salmonella Typhimurium, S. aureus and B. thetaiotaomicron are the most relevant based on genome size, ecological niche and genetic relatedness to P. gingivalis. The determined essential genes of the above species were 1,657 genes with 462 essential (28%); 1,719 genes with 390 essential (23%); 3,307 genes with 499 essential (15%); 3,988 genes with 614 essential (16%); 4,314 genes with 353 essential (8%); 2,892 genes with 351 (12%); and 4,902 genes with 325 (6.6%), respectively [30, 34, 44, 47, 56, 57, 62].
P. gingivalis is a member of the Bacteroidetes, and before reclassification was known as B. gingivalis. There are no Bacteroidetes species or other anaerobes represented in the DEG, however, a putative list of B. thetaiotaomicron strain VPI-5482 essential genes is available from the supplemental material of Goodman et al. 2009 [30]. B. thetaiotaomicron strain VPI-5482 was originally isolated from human feces. The strain contains a 6.26 Mb chromosome and 0.03 Mb plasmid (NC_004663.1/NC_004703.1) with 4,864 genes (chromosome) and 38 (plasmid), 4,778 protein coding sequences (chromosome) and 38 (plasmid), 71 tRNA and 15 rRNA genes [63] [GenBank: AE015928.1 and AY171301.1]. In comparison, P. gingivalis ATCC 33277 has 43% (numerically) of protein coding sequences in a genome 37% of the size of that of B. thetaiotaomicron VPI-5482. It was estimated that B. thetaiotaomicron VPI-5482 contains 325 “candidate essential genes” [30]. Maintaining a larger genome and gene set provides more opportunities for functional redundancy and alternative pathways which can lead to a relatively smaller number of essential genes. Thus, 268 of 325 (82.5%) of B. thetaiotaomicron ‘essentials’ have BLAST homologues in P. gingivalis strain ATCC 33277 and of these, 78% (209 of the shared 268) are also essential in both organisms (Additional file 1: Table S1). Fifty-nine B. thetaiotaomicron BLAST matches are not essential in P. gingivalis and 57 have no BLAST match at all in the organism (Additional file 2: Table S2). A significant number of the shared essential genes (25 of the 209) are not characterized in the DEG (Additional file 3: Table S3) and of these 25 Bacteroidetes-specific essentials, three are annotated as permeases and two appear to be regulatory. Three essentials, PGN_1026, PGN_1481 and PGN_0249, are likely associated with capsular polysaccharide biosynthesis based on PGN_1026 and PGN_0249 being involved in the dolichol pathway and PGN_1481 functionally annotated as polysaccharide biosynthesis related. Parsing out essential genes of specific groups of species, in this case Bacteroidetes and/or anaerobes, can allow for specific drug targeting or directed nutrient supplementation.
In agreement with multiple previous studies on essential genes of bacteria, in P. gingivalis a significantly greater number of essential genes (248 or 53.6%) are found on the negative DNA strand, and 215 (46.4%) are found on the positive DNA strand (Additional file 1: Table S1) [64]. Similarly, there is a greater than expected proportion of enzymes, especially those within multiple functions or involved in multiple pathways, within the essential gene groups [65].
Using the Cluster of Orthologous Groups (COG) functional class designations (NCBI), we identified significant enrichment of essential genes within groups ‘D’ (cell cycle control/cell division), ‘I’ (lipid transport and metabolism) and ‘J’ (Translation/Ribosome); and a lack of enrichment was seen in ‘S’ (function unknown), ‘P’ (inorganic ion transport and metabolism) and ‘N’ (motility) (Figure 6A/6B) [66–68]. Enrichment (or lack thereof) of essential genes in these categories has been reported previously, however, essential gene enrichment in specific COG categories appears to be a species-specific characteristic.
Based on operon prediction and known essentials contained in the DEG it was determined that 25 of the 463 putative essential genes of P. gingivalis identified by Tn-seq may be the result of polar effects of the transposon insertion on downstream essential genes (Additional file 1: Table S1). Specifically, these 25 genes were identified as being upstream and potentially in an operon with one or more known essential genes, and additionally do not have BLAST matches in the DEG. Further study of each of these genes would be required to confirm their essentiality.
Bringing the DEG and P. gingivalis core genome together in relation to P. gingivalis gene essentiality, we have determined that 369 genes within the core genome, ones not identified as essential in our study, have BLAST matches to genes within the DEG (Figure 7)(Additional file 4: Table S4). Within our mutant libraries we were able to identify transposon insertions into these genes such that they do not qualify as essential in P. gingivalis. Reasons for these genes being identified as essential in other species could be due to multiple variables such as in vitro selection media, species-specific essentiality, transposon type, library complexity, sequencing method, and criteria for essentiality. Such information gives importance to the distinction between a core gene set and an essential gene set as well as possible limitations of essential gene analyses based solely on in silico methodology.
Characterization of P. gingivalis essential genes
Metabolic pathways that lack redundancy or have critical functions have been identified previously through essential gene studies. In our analysis of P. gingivalis we noted the presence of entire pathways as well as specific parts of pathways that are essential to P. gingivalis and to all other bacterial species. A subset of P. gingivalis-specific essential genes, possibly related to the ecological niche of the species, have also been identified (Additional file 1: Table S1). Of pathways involved in ribosome function we identified the rpsA, rpmA, rplB and rimP systems, which encode for 30S, 50S and maturation of ribosomes, respectively. The three major protein translation regulatory pathways of infB, tsf and prfA, as well as translational machinery pathway rpoA were found to be essential in our study. DNA replication, recombination and repair pathways of dnaA and ruvA as well as cell division pathways mreB, parA and ftsA were also found to be essential. Multiple pathways involved in LPS, CPS, fatty acid and murein biosynthesis, including lpxA, rmlA, manA, fabD and murA were also judged to be essential, as well as genes involved in secretion and chaperone pathways such as secD, groES/EL and surA. Pathways involving nrfA, etfA, sufB, nadD and ribE associated with oxidation-reduction reactions, were found to be essential in P. gingivalis, which is not surprising for an anaerobic bacterium. Major metabolic pathways purA, pyrB, coaA, accB, pdxA, ispA, thiF, serA and dapA, which encode nucleotide, amino acid and co-factor building blocks, respectively, were determined essential under our in vitro conditions. All of the aforementioned systems and pathways have previously been identified as essential for in vitro growth of other bacterial species, which is not unexpected given that replication, transcription, translation, cell division, membrane stability and central metabolism are key to survival [37].
Hypothetical genes are an often-overlooked group within any bacterial genome, including those of ‘model’ organisms. In our study we determined that approximately one-tenth of the essential genome of P. gingivalis encoded ‘hypothetical’ proteins, a few of which were homologous to other hypotheticals contained within the DEG. The majority of essential hypothetical genes are large and not within operons, suggesting that they encode functional proteins and are not essential due to a polar effect on a downstream essential gene. The finding that certain hypothetical proteins are essential will stimulate the search for protein motifs, structural bioinformatic and spatial organizational data and studies to define their function.
Although the notion that an essential gene within a given strain is likely to be essential in the species as a whole, intraspecies differences are known and often result in different phenotypes. For example, in strain ATCC 33277 we found no insertions into ragA and thus this gene was considered essential. Previous investigators also had difficulty making directed knockouts of ragA in strain W50; however, these investigators were successful in deleting ragA from strain WPH35. It is possible that ragA is only essential within specific strains and those strains in which it is non-essential compensate for loss of its function through the presence of other genes.
Limitations of essential gene analysis
Limitations to essential gene studies should be addressed regardless of systems and methods utilized for their identification. First, several studies have relied exclusively on in silico bioinformatic analyses to determine essentiality. These analyses were based on programs designed to combine information from previous in vitro and in vivo mutagenesis studies with genome annotation and composition scripts without having carried out actual mutagenesis studies. Thus, any limitations of these experimental studies will be carried over into the new analyses and magnified by any inaccuracies of the program design itself. Second, in insertional mutagenesis methods to determine gene essentiality, genes may be misidentified as essential due to transposon insertion ‘cold-spots’. There is no ideal transposon identified as yet that completely lacks any nucleotide specificity and which can create completely random and saturating mutant libraries. Thus, no matter what type of transposon is used, Tn5, Tn7, Tn10, a cryptic construct or Mariner, all studies will have regions of the genome where fewer insertions occur. Third, genes that are actually non-essential but when mutated cause severe growth defects may be scored as essential due to practical limits to the depth of sequencing of transposon insertion junctions. These ‘sick’ mutants could potentially be represented at levels below 1000-fold a neutral mutant due to the number of replications it could go through prior to being pooled from mutagenesis plates into the library. Fourth, non-essential genes immediately upstream of and co-transcribed with essentials may be incorrectly scored as essential due to polarity of the transposon insertion. Last, practical limits to library complexity can result in some genes that fail to get disrupted by the transposon and so are misidentified as essential. This is particularly a problem for small genes or genes that are within cold spots for the transposon. Several studies, based mostly on the genome size of the species under investigation and the type of transposon, have attained different levels of saturation prior to analyses for essential genes. The possible limitation of our library when combining the type of transposon and library complexity relates to genes that contain less than 10 TA sites in their coding-sequences. Of the 204 genes with fewer than 10 TA sites, 60 could potentially be scored as essential based on having zero insertions, but do not qualify, given our stringent criteria (Additional file 5: Table S5). Adding confidence to the notion that many of these are non-essential is that 24 of the 60 genes encode proteins of less than 35 amino acids in length. Since these are all characterized as ‘hypothetical’ and are rather short to encode functional proteins, we believe that some of these may simply be artifacts of annotation programs and thus not true protein-coding genes.
Even complete gene deletion, non-transposon based studies of essential genes have limitations. The Keio collection of single and double gene deletions in Escherichia coli is considered the most comprehensive essential gene study to date [43, 69]. Genes that could not be deleted were scored as essential, however, failure to delete a gene is not a guarantee of essentiality and there a are few genes identified as essential in the Keio collection that were successfully deleted by other labs. Furthermore, a handful of genes labeled non-essential were actually essential. The Keio deletions of those genes have second site suppressor mutations that compensate for the loss of the essential gene.
The best understanding of essential genes is likely to come from combining different modalities to confirm their essential nature and comparison of these databases both within and between species.