How E. coli Rose to Prominence – Again, One More Addendum (2|2)

by Christoph
 

In the fourth and, for the time being, last episode of this Ecolimania in STC, I take a look around: Do we know the diversity of the species E. coli ? And what can we learn from analyses of its pangenome ? Reminder: Roberto looked into the past of How E. coli's Rose to Prominence, plus An Addendum, and I looked ahead in Just One More Ad­den­dum.
 

(click to enlarge)
Figure 2.1. Circular representation of the E. coli NCTC 86 chromosome. From the outside to inside, circle 1 shows the sizes in bp. Circle 2 marks the positions of regions of difference (RODs). Circles 3 and 4 show the positions of coding sequences (CDSs) transcribed in a clockwise and anticlockwise direction (the position of the replica­tion origin oriC is indicated). Circles 5 to 10 show the po­si­tions of E. coli NCTC 86 genes that have ortho­logues (by reciprocal FASTA analysis) in other E. coli strains: K-12 MG1655 (red), E24377A (ETEC; yellow), 042 (ETEC; green), Sakai (O157:H7; blue), IAI39 (Extrain­testinal Pathogenic E. coli; pink), CFT073 (UPEC; orange). Circle 11 shows a plot of G+C content. Circle 12 shows a plot of G+C skew. Note that what appear to be graphical glitches in each circle are local genomic inversions. Source. Frontispiece: Detail from Figure 2.4.

This is not about the diversity of E. coli in all its splendor. I will limit myself to the quantitative aspects of its genomic diversity. AllTheBacteria, the publicly accessible database just published by Hunt et al. (2024) and containing 300,000+ E. coli genomes, triples the number of the previ­ous "661k" collection. The sheer volume of these "stamp collections" should suffice to cover the ge­netic diversity of E. coli within the limits of the known sampling bias: most genome se­quences were obtained from pathogenic and commensal isolates from humans, domestic mam­­mals and birds.
 

A note in advance: the 'circular plot' images in this post are mainly illustrative and a look at the overall picture is suf­ficient. Clicking on the images enlarges them so that they can be read better, but the large number of data points in each image cannot really be grasped. All images represent compromises in re­gard to the reso­lu­tion (in pixels) that can be achieved in print or the file size for webpages. Furthermore, it is not yet technical­ly feasible to reproduce interactive images on websites – and is impossible in print – that would allow zooming in, which would be crucial for a detailed evaluation of the images.
 

Do we know the genomic diversity of the species E. coli ?

In 2017, Dunne et al. (2017) sequenced the genome of E. coli NCTC 86 and compared it to the genomes of six other strains (Figure 2.1). E. coli NCTC 86 is the strain originally described by Theo­dor Escherich as Bacterium coli commune, later renamed Escherichia coli, and in 1919 de­posited in the NCTC. Its genome has a size of 5.144 Mbp, is classified as belonging to phy­lo­­type A (see below for "phylotypes") and is re­lated to K‑12 strain MG1655 (4.625 Mbp), but is not at or close‑to the base of a (phy­lo­genetic) E. coli species tree (Figure 2.1b). Hence, it is not "the wild type" researchers could or should refer to.
 

The circular alignment of the seven E. coli genomes in Figure 2.1 offers an instructive and ex­em­plary overview of their similarity and likewise their diversity. The sequences are to a high de­gree co‑linear, resulting in a syntenic arrangement of the (colored) orthologous genes like pearls on strings (in the paper, the authors do not disclose the significance of the two color shades in each of the 6 circular plots. I assume this means differences in sequence identity). Dun­ne et al. (2017) count 41 regions of divergence (RODs) that coincide in position in all ge­nomes. They found differences in gain/loss of single or a handful of genes that are widely scattered through­out the genomes. Together, these ROSs and the numerous single gene gain/loss events con­tribute to the ~0.5 Mbp differences in genome size between the seven strains com­pared.
 

(click to enlarge)
Figure 2.1b. Phylogenetic relationships amongst se­quenced E. coli genomes. The genomes of E. albertii and E. fer­gusonii are included as outgroup sequences. The tree was obtained by maximum likelihood analysis of a concatenated alignment of 2173 genes, using the general time reversible (GTR/REV) model, with the CAT approximation of rate heterogeneity as implemented in RAxML version 8.1.3. The numbers on individual bran­ches indicate the percentage support from 100 non-para­metric bootstrap replicates, performed using the rapid algorithm implemented in RAxML. The branch length indicates the number of substitutions per site. The major E. coli phylogenetic groups (phylotypes) are indicated. ⚫ , strains in Figure 2.1. Source

If you look really close at Figure 2.1 you see what appear as glitches in the graphics are ac­tu­ally small localized genomic inversions. Ignored in the figure for clarity but mentioned in the text is a large 2.8 Mbp inversion in the E. coli NCTC 86 genome that is not present in the other six ge­nomes (see here). Intra‑genomic inversions of any size occur comparatively fre­quently in the ge­nomes of Enterobacteriaceae, usually coinciding with the pre­sence of multiple transposable ele­ments in the genome, and are not restricted to distinct re­gions. But since it is in most cases im­possible to deter­mine when inversions occurred during speciation of a lineage because they can be overlaid by later partial overlapping inversions at any time, their phylogenetic signal is fuzzy at best. The large inversion in the genome of E. coli NCTC 86 can only be classified as a late "acqui­sition" because it is not found in any of the other six genomes. (As mentioned here in STC, the large intra‑genomic inversion in E. coli K-12 strain W3110, which stands out when compared to the genome of K‑12 strain MG1655, is also such a late "acquisition".)
 

Ghomi et al. (2024) have recently taken a different approach to tackling the genomic diversity of E. coli. The authors compared the essentiality of genes (ORFs) in three E. coli strains with that in strains of closely related species. Of particular interest is their result of a genome alignment that clearly visualizes species boundaries (Figure 2.2). Orthologs are identifiable in the genomes of other Enterobacteriaceae species for the vast majority of E. coli genes (here irrespective of their essentiality). On average, however, the orthologous genes in the three E. coli genomes (the genome of E. coli BW2113 serving as reference) are significantly more similar to each other than the corresponding orthologs in the genomes of ten strains of four other species. This conclusion would not change if the identity classes (100%, >85%, >70%) had been chosen differently and with more grada­tions (but would be less easily to visualize in a figure).
 

(click to enlarge)
Genome alignment for the 12 genomes in this study com­pared to E. coli BW25113 from Keio collection, ge­ne­rated with BRIG. The genomes from the inner circle to the outer circle are E. coli BW25113, E. coli UPEC ST131 EC958, E. coli UPEC ST131, Salmonella Typhi­mu­rium A130, Salmo­nella Typhimurium D23580, Salmo­nella Typhimurium SL3261, Salmonella Typhimurium SL1344, Salmonella enteritidis P125109, Salmonella Typhi Ty2, Citrobacter rodentium ICC168, Enterobacter cloacae NCTC 9394, Klebsiella pneumoniae Ecl8, and Klebsiella pneumoniae RH201207, respectively. (Modi­fied by the author to show a scale (inner circle) for E. coli BW25113 and the approx. pos. of oriC in this ge­nome). Source

To get the most out of Figure 2.2, you need to know that although Ghomi et al. (2024) call it an "alignment" in the legend, the layout of the graphic is somewhat more complicated. It is a pro­jection of an ortholog from each of the other 12 genomes onto the position of this ortholog in the genome of E. coli BW25113, regardless of the position of this ortholog in its own genome (I wish I could describe it in a less twisted way). This be­comes im­mediately clear when one looks at the co-linearity of the genomes, which, although almost complete for S. Typhimurium LT2 and E. coli BW 25113 except for an inversion in the terminus region, is largely in disarray in the case of Citrobacter rodentium, for example (see Figure 2.2b). And then I see nine "gaps" in the circle for the reference strain E. coli BW25113, where the authors apparently took orthologs from the other E. coli strains as the reference (without explaining this).
 

Taken together, the genomes of the three E. coli strains stand out as a homogeneous group and the alignment visualizes species boundaries on the background of a high number of or­thologous genes (ORFs) in the other Enterobacteriaceae. However, species boundaries cannot be simply quanti­fied by an averaging calculation on this limited set of genome sequences, not the least since individual orthologs are by not equally conserved.
 

It has been known for more than a century that the species E. coli is phenotypically diverse, and that a simple di­vi­sion into "commensal" and "pathogenic" isolates is insufficient. Extensive attempts to dis­tin­guish and differentiate E. coli isolates based on morphological or metabolic criteria did not lead to a fully consistent differentiation/typification scheme; nor did sero­typing, which led to the known serotypes O, K and H, think of E. coli O157:H7.
 

Over time, phylotyping became the method of choice for differentiating E. coli iso­lates/strains by their genotypes. The method is based on PCR analyses, today also as digital PCR (dPCR) of a few selected genes, and was explored in the initial studies by Clermont et al. (2000). The number of E. coli phylo­types has grown from initially six to now eight, and these are sufficient to cover >90% of the genomic diversity of the species E. coli (see Figure 2.1b). All phylotypes contain vastly different numbers of "members", and this is primarily due to the (above mentioned) known sampling bias for E. coli, rather than to real differences in their occurrence in natural habitats. And in case you wonder, it is now confirmed by numerous phyloge­netic studies that Shigella, isolated by Kiyoshi Shiga in 1898 (see here in STC), belongs to the species E. coli and is therefore listed among the E. coli phylotypes. Only colleagues in me­dical micro­bi­ol­ogy want to keep Shigella as a (taxonomic) species because they are not inclined to abandon the terminology for established diagnostics and therapies of shigellosis.
 

What can we learn from analyses of the pangenome of the species E. coli ?

First of all, it must be noted that to date there is no published pangenome analysis on all ~10⁵ E. coli genomes in the above-mentioned "661k" collection. All published studies were performed on a significantly smaller sub­sets of genomes. The algo­rithms necessary for com­putational pan­ge­nome analyses have been, and are still being, de­vel­oped and evaluated on subsets. A major reason for the lack of com­pre­hen­sive pangenome stud­ies for E. coli is that the necessary com­puting times on hi‑perfor­mance computers are hard to come by. The main focus of bioinfor­ma­ti­cians is there­fore to make their algorithms ever faster without compromising on quality. It's not unlike wet-lab microbiologists using all the tricks to make their fastidious pets grow with genera­tion times of days rather than weeks or months.
 

Pangenome analyses (or "pan‑genome" as mentioned here in STC) have introduced a number of terms that are now in common use: "...the core genome com­prises the genes that are present in (nearly) all genomes analyzed. Some authors consider a softcore genome (>95% occurrence) to avoid dismissing gene fa­mi­ies due to sequencing/annotation artifacts. The shell genome consists of the genes shared by the majori­ty of genomes (10–95% occurrence). The gene families pre­sent in only one genome or <10% oc­currence are described as accessory or cloud genome" (adopted from Wikipedia).
 

Tantoso et al. (2022) found, when analyzing 1,324 genomes, that the E. coli pangenome is an "open pangenome", meaning that with every genome added the number of cloud genes in­creases without reaching a plateau while the number of core genes (or softcore genes) asymp­tot­i­cally flattens out to a minimum (see here). All the studies I am aware of found an "open pan­genome" for E. coli. And, according to current estimates, the E. coli cloud genome comprises ~80,000 genes/gene families, while the core genome comprises ~2,500–3,000 genes/gene fa­mi­lies.
 

With 10,667 E. coli genomes analyzed, Abram et al. (2021) come closest to a comprehensive quantitative pangenome study and the (presently) least blurry "snapshot" of the genomic diversity of the species. The authors obtained this subset by "cleaning", that is by removing duplicates and inconsistent "complete genomes" from the 12,602 E. coli and Shigella genomes in the then‑available NCBI RefSeq collection. Since it is virtually impossible to align and compare this number of genomes one-to-one and subsequently determine the degree of relatedness by conventional cluster analysis, the au­thors resorted to alignment-free tools that are routinely used in metagenomics (see here for a primer in STC). Briefly: the genomes were converted to "sketches" in an approach based on k‑mers (k-mer size of 21, sampling size of 10,000). (Note that it is in principle possible to "reverse engineer" the original sequence form such "sketches"). From these "sketches", a distance matrix for all 10,667 genomes was calculated using Python, with the accession numbers of the genomes as columns and rows. This matrix of was then clustered with Mash, using hierarchical clustering to produce a heatmap of 'Mash distances', which illustrates the population structure of these genomes (Figure 2.3).
 

(click to enlarge)
Figure 2.3. Heatmap representation of 10,667 genomes using Mash distances. The color bars at the top of the heatmap identify the phylogroups as predicted from the analysis. The scale to the left of the dendrogram corresponds to the resultant cluster height of the entire dataset obtained from hclust function in R. The colors in the heatmap are based on the pairwise Mash distance between the genomes. Teal colors represent similarity between genomes with the darkest teal corresponding to identical genomes reporting a Mash distance of 0. Brown colors represent low genetic similarity per Mash distance, with the darkest brown indicating a max. dis­tance of ∼0.039. Genomes of relative median ge­netic similarity have the lightest color. Source

As indicated in the legend to Figure 2.3, colors in the heatmap are based on the pairwise Mash distance between the genomes. Teal colors represent similarity between genomes with the darkest teal corresponding to identical genomes reporting a Mash distance of 0. Brown colors represent low genetic similarity per Mash distance, with the darkest brown indicating a max­imum distance of ∼0.039. Since you cannot read-out numbers from Figure 2.3, here they are: 70% of the 10,667 E. coli genomes belong to four phylogroups: B1 (28%), A (21%), B2-2 (13%), and Shig2 (8%). phylogroup C is represented by 5% and E2(O157) by 7% of the genomes. The remaining 18% are distributed unevenly over eight phylogroups and include the 9% "un­assign­ab­le" cases. The methodology used in this work is thus able to reproduce known phylogroups, as well as to identify previously uncharacterized E. coli phylogroups, with an acceptable amount of "bottom sediment" (the authors use the terms 'phylotype' and 'phylo­group' inter­changeably).
 

The authenticity of these 14 phylogroups is supported by three different lines of evidence: 1. phylogroup-specific core genes, 2. a phylogenetic tree constructed with 2613 single‑copy core genes, and 3. differences in the rates of gene gain/loss/duplication. It goes without saying that the analysis of entire genomes allows a much greater depth of analysis than the PCR-based method, which is limited to a few gene segments. Note that Abrams et al. (2021) counted "core genes", i.e. genes that are present in all genomes of a phylogroup, and not, as Ghomi et al. (2024) "essential genes" (see above).
 

To classify the 95,525 unassembled genomes, the authors took an interesting approach: rather than averaging the Mash values over a phylogroup, they defined a "medoid" as the genome that has the lowest average distance to all other genomes in its phylogroup when using the Mash values obtained for the entire set of 10,667 assembled genomes. By using the medoids of the 14 phylogroups as a proxy to classify the unassembled genomes, they found that two-thirds (67%) belong to phylogroups A (23%), B1 (15%), C (15%), and E2(O157) (14%). Strains belonging to phylogroups B1, C, and E2(O157) are of­ten pathogenic and of interest to medical research and epidemiology, whereas phylogroup A includes strains frequently used in the lab or genet­ically modified strains. Researchers working with these strains are often not interested in fully assembling the genomes, which could explain the uneven distribution across the phylogroups compared to that of the complete genomes. (the remaining ~31,000 unassembled genomes were distributed at lower percentages among the remaining phylogroups or did not meet the cutoff).
 
 

To come back to the initial question of whether we know the genomic diversity of the species E. coli, I should give a two-part answer:

yes, as far as the approximate size of the core genome is concerned (note that "core gene" and "essential gene" are not synonymous). There are about 2,500–3,000 core genes/gene families across all E. coli phylotypes, but since the function of their gene products is only known for a fraction of the genes, although a major fraction now, it is not pos­si­ble to say exactly how much redundancy there is in the core genome. And in our at­tempts to extrapolate from genomic to genetic diversity, we still lack knowledge of many regulatory pathways and knowledge of their respective redundancies.

Also a cautious yes for the approximate size of the cloud genome. Estimates based on the number of genomes available in 2021 suggest that there are about 80,000 cloud genes/gene families, with the tendency of a steadily slowing increase with more genomes added to the collections. Since E. coli has an open pangenome, we will have to be content with estimates also in the future.
 

Pangenome analyses carried out in recent years have already led to compelling results, despite the fact that this type of analysis is in its infancy. Or rather, is in pre-school age, to be fair. The approach of a Mash-based analysis of multiple genomes chosen by Abram et al. (2021) can, if the parameters are set accordingly, reproduce the known E. coli phylotypes to a satis­fac­tory extent and is suitable for identifying new phylotypes. Since genome se­quencing is now a lab rou­tine, it is not presumptuous to expect that this analytical tech­nique will replace PCR-based "Cler­mon­Typing" over time, especially when the "medoids" of the 14 phylo­groups are employed to reduce the computational demand. A Mash-based analysis may even help with the genomes of E. coli strains that defy PCR‑based phylotyping. It is not possible to predict whether the number of 14 phylogroups proposed by Abrams et al. (2021) will persist, whether they will need further differentiation, or whether new ones will be added when genomes of E. coli from previously underrepresented habitats are added to the "basket".
 

(click to enlarge)
Figure 2.4. Gene relationships in the E. coli pangenome. (a) An overview of the co-occurrence and on click (b )avoid­ance networks, coloured by connected com­po­nent. Selected connected components are numbered as in the Supplementary Data Files. Source

The sheer number of core genes/gene families plus the daunting number of cloud genes/gene families in the E. coli pangenome and our current (at best) spotty understanding of the func­tion(s) of the latter fraction are a strong reminder of what W. Ford Doolittle said in a 1999 paper, namely "that all prokaryotic taxa are in essence imprecisely bounded and ephemeral. We might thus realistically look at all prokaryotes as one ‘global superorganism’ ... divided into subpopulations – within and between which genes are exchanged at different frequencies." (exchanged by horizontal gene transfer (HGT), that is). One of Doolittle's 'subpopulations' is Es­cher­ichia coli, which, according to Kostas Konstantinides, is a "sequence-discrete population" (95–100% ANI) and should therefore be considered a taxonomic species. Doolittle's 'impre­cisely bounded' is reflected in the huge cloud genome of E. coli, and that this species is 'ephemeral', well, yes, this is what evolution is about after all.
 

By a pangenome analysis of a selected subset of 400 E. coli genome se­quences Hall et al. (2021) revealed numerous groups of genes that preferentially co‑occur together in a genome (Figure 2.4), and reciprocally, groups of genes that preferentially do not co‑occur with other gene groups in a genome (click here for Figure 2.4b). These authors have cut first deep trails into this seemingly im­pen­e­tra­ble jungle: there is structure in the E. coli pangenome, shaped by its evolution, and it's not simply a vast heap of core and cloud genes/gene families. Not that we already understand this structure well enough...
 

Do you want to comment on this post? We would be happy about it! Please comment on Mastodon, Bluesky, or on 𝕏 (formerly Twitter).