Escherichia coli Residency in the Gut of Healthy Human Adults

Abstract

Escherichia coli is one of the most well-studied bacterial species, but several significant knowledge gaps remain regarding its ecology and natural history. Specifically, the most important factors influencing its life as a member of the healthy human gut microbiome are either underevaluated or currently unknown. Distinct E. coli population dynamics have been observed over the past century from a handful of temporal studies conducted in healthy human adults. Early studies using serology up to the most recent studies using genotyping and DNA sequencing approaches have all identified long-lived E. coli residents and short-lived transients. This review summarizes these discoveries and other studies that focused on the underlying mechanisms that lead to establishment and maintenance of E. coli residency in healthy human adults. Many fundamental knowledge gaps remain and are highlighted with the hope of facilitating future studies in this exciting research area.

INTRODUCTION

Escherichia coli was first isolated and described by the Bavarian doctor Theodor Escherich in 1885 (1). As a pediatrician, Escherich was interested in the role of microbes in digestion and intestinal disease, primarily because diarrhea was a leading cause of infant mortality in Europe at the time (2). In an 1885 article, Escherich wrote, “…it would appear to be a pointless and doubtful exercise to examine and disentangle the apparently randomly appearing bacteria in normal feces and the intestinal tract, a situation that seems controlled by a thousand coincidences.” Despite the apparent cynicism, Escherich cultured, isolated, and used microscopy to describe bacteria present in neonatal stool samples at birth and throughout the first few months of life. As a result of his “pointless and doubtful exercise,” Escherich used Christian Gram’s newly developed staining technique to describe a Gram-negative, rod-shaped bacterium that appeared to be common among neonates (1). This bacterium, which he called Bacterium coli commune, was capable of growing on simple nutrient sources to form “a massive luxurious deep growth” (1, 2). Bacterium coli commune was renamed Escherichia coli in 1919 to honor Theodor Escherich (3), and somewhat remarkably, perhaps the first E. coli isolated by Escherich was recently rediscovered and sequenced (4). NCTC86, the original strain designation at the National Culture Type Collection (London, UK), was free of known pathogenicity and virulence genes and belonged to one of the four main phylogenetic lineages described for E. coli (phylogroup A). The strain also belonged to multilocus sequence type 10, which is still commonly observed in healthy children and adults (5–7). Given its relevance to health and its ability to be easily cultivated and manipulated in the laboratory, E. coli has served as a model organism to understand biology at its deepest levels.

E. coli is a species of almost exclusively nonpathogenic bacteria. In cross-sectional studies of human adults, E. coli is a member of the intestinal microbiome of over 90% of individuals (7). However, E. coli ecology has been biased heavily toward only a few pathogenic and antibiotic-resistant strains, and remarkably little is known about the biotic and abiotic factors influencing the abundance and distribution of nonpathogenic representatives of the species. E. coli is a pioneer of the human gut, being one of the first bacteria to colonize neonates at birth (8). As a facultative anaerobe, E. coli may help deplete oxygen along the gastrointestinal (GI) mucosal surface, creating a hospitable environment for strict anaerobes to colonize and become dominant (8). Given its role in the gut of neonates, its capacity to produce vitamin K (9), and its ability to confer colonization resistance (protection against pathogens) (10), human-associated E. coli can by definition be considered a mutualist even though much of the literature refers to the species as “commensal” (11). With this in mind, several research groups over the past century have begun to ask important ecological questions about E. coli, including “How do populations change through time?” and “What causes these changes?” Remarkably little is known about such dynamics, due in part to the above-mentioned emphasis on pathogens but also due to the use of cross-sectional rather than temporal study designs, the former of which captures only a snapshot of population-level diversity. More data and different approaches are needed to demystify E. coli dynamics. Here, we review current information regarding E. coli membership in the gut of healthy adult humans, focusing on experimental results that shed light on residency.

EXPOSURE VERSUS COLONIZATION

Humans are constantly exposed to microorganisms that are seemingly capable of colonizing the gut (i.e., producing offspring to the extent of overcoming the rate of gut transit). Ingestion of uncooked foods (raw meat and fresh fruits and vegetables) and drinking water are well-known routes of exposure to bacteria and E. coli. Other common fomites include currency (e.g., E. coli isolated from 8 to 40% of bills or coins [12, 13]) and cell phones (e.g., E. coli isolated from 2 to 14.3% of those surveyed [14, 15]). Given the prevalence of E. coli in the human environment and the high probability of exposure from multiple sources, humans almost certainly ingest many different types of E. coli throughout their adult life. In addition, pathogenic E. coli epidemiology has clearly linked diseases and/or “carriage” with a variety of environmental exposures (pets, food, water, etc.). With these factors taken together, it seems reasonable to assume that humans “randomly sample” different E. coli from their environment.

The identification and characterization of bacterial “immigrants” including E. coli, into the human gut is an important but understudied area of human health. Presumably, all immigrants are not equally likely to colonize, but how many cells are required for colonization is virtually unknown for individual members of the human microbiome. For pathogenic E. coli, like Shigella sonnei, human volunteer studies suggest that only a few cells are necessary for disease (16). Whether the same is true for naturally occurring, nonpathogenic E. coli remains to be experimentally evaluated. Double-blind, placebo-controlled trials of probiotic E. coli suggest that these strains are poor colonizers of the human gut even when exposure levels are high. For example, E. coli Nissle 1917 (Mutaflor), was detected in only 2 of 12 volunteers tested, even though they orally ingested 10⁹ to 10¹⁰ CFU twice daily (17). Finally, the presence of the same E. coli clone in cohabitating individuals and their companion animals suggests that repeated exposure may promote colonization (18, 19), although to what extent and under what conditions has yet to be determined. Even though more work is needed in this area, especially regarding naturally occurring, nonpathogenic strains, studies published to date suggest that exposure alone does not adequately or fully explain E. coli colonization of the human gut.

In stark contrast to immigration and reviewed throughout the remainder of this manuscript, E. coli exiting the human gut (i.e., “emigrants”) are not observed at random, and long-lived colonizers are phylogenetically distinct from short-lived transients. This again suggests that exposure alone cannot fully explain whether one E. coli versus another occupies a human intestinal niche. It should be noted that few studies, if any, have tested the above-mentioned random sampling hypothesis by comparing the diversity of E. coli in a person’s environment (exposure) to those cultured from stool, and again, especially with respect to nonpathogens. Here, we assume random sampling or at least exposure to many diverse E. coli throughout human adulthood and focus on events that take place over time after E. coli has successfully immigrated into the gut of human adults from whatever route, vehicle, or vector brought them there. We refer to in situ events as “dynamics” and emphasize factors that likely promote sustained colonization, or residency.

TEMPORAL E. coli DYNAMICS IN THE PRE-PCR ERA

The study of natural dynamics in bacteria first requires the ability to identify the same asexually reproducing bacterial clone in different samples, and a variety of phenotypic and genetic techniques have been developed for E. coli with this purpose in mind. E. coli was originally identified using serotyping, which differentiates isolates based on agglutination reactions between antisera (usually raised in rabbits) and bacterial antigens of surface-exposed oligosaccharides (O), capsule (K), and flagella (H) (20). As of 2016, 186 O-antigens and 53 H-antigens had been described to serotype E. coli, but due to technical difficulties, K-antigens are rarely used (20). An E. coli naming convention based on unique combinations of O- and H-antigens was long used for clinical and epidemiological purposes. More recently, DNA sequencing revealed that serotyping was not as discriminant as originally believed and that recombination in genes encoding for O- and H-antigens masked phylogenetic relatedness in certain cases (e.g., distantly related E. coli may have the same O:H designation) (21, 22). Regardless, studies of temporal E. coli dynamics in humans that utilized serotyping first described trends still observed today with more advanced methodologies (e.g., genome sequencing).

The second requirement for studying natural bacterial population dynamics is the ability to culture and isolate the individual(s) under investigation. DNA sequencing techniques can theoretically identify individuals without the necessity of culture. However, this is often not the case, because individual bacteria of interest are either present below detection limits or become so at some point during the study period. Cultivation is still needed in most situations to enrich or select for bacteria of interest, and especially when the point of the study is to identify individual clones as opposed to higher taxonomic groups (e.g., phylogenetic groups, species, genera, etc.). Thus, it is almost exclusively through the lens of cultivation that E. coli dynamics have been studied.

Enteric (gut inhabiting) bacteria such as E. coli are adapted to many conditions in the human GI tract that act as barriers to other bacteria, such as stomach acidity (the baseline pH in healthy humans is approximately 1.5 [23]) and bile (a complex mixture of reactive metabolites and detergents produced by the liver and secreted by the gallbladder into the small intestine [24]). Alfred MacConkey took advantage of this physiological filter to create a selective medium containing bile and crystal violet to inhibit Gram-positive bacteria (25, 26). Furthermore, MacConkey included a pH-sensitive dye to determine whether bacteria growing on solidified (agar-containing) medium fermented lactose (4, 26). This selective and differential medium, now known as MacConkey agar, still serves as an important cultivation technique for E. coli and nearly all currently named members of the Enterobacteriaceae family.

The earliest study of temporal E. coli dynamics in human adults that we are aware of was published in 1902 by K. Totsuka, a Japanese scientist in residence at the Institute for Infectious Disease in Berlin under the direction of Robert Koch. Totsuka self-collected stool samples on a weekly basis for four months and evaluated up to 32 isolates per sample. Totsuka first tested for differences in motility, indole production, and acid production and the ability to coagulate milk between different isolates (14). Using an early form of serotyping, he then described gradual shifts from one antigenic type to another (27, 28). The emergence of different serotypes over the study period that Totsuka observed was in contrast to the serotype stability/homogeneity observed in a temporal study in infants a few years earlier (1899) (28, 29). Little work was performed to follow up on the Totsuka study until 1943, when Wallick and Stuart described E. coli serotype dynamics in a single man over a 14-month period (28). They found at least four distinct serotypes that persisted for months at a time, some of which overlapped (i.e., simultaneously colonized). Of the 650 isolates evaluated, 85.3% were antigenically identical to 1 of 10 isolates used to produce antisera, suggesting that the bulk of population-level diversity was the result of a handful of closely related isolates. Interestingly, they observed that some serotypes emerged, became dominant for several months, and then disappeared, thus providing the first evidence that closely related E. coli stably colonize the human gut but are inevitably replaced. Wallick and Stuart also discovered that E. coli isolated from the study subject and E. coli isolated from a family member on the same day were antigenically identical, suggesting that transmission between individuals in close contact with one another may explain at least some of the observed dynamics. Moreover, one E. coli serotype observed in the study subject matched an isolate cultured from 100 unrelated individuals (1 isolate considered per person), which on face value may seem like this serotype was shared between distantly related individuals. However, the authors found multiple examples where isolates belonging to the same serotype were phenotypically distinct (i.e., utilization of 6 different carbon sources differed), and so it is possible that this single example of a serotype match was not due to widespread sharing of an E. coli lineage. Since the single isolate from the nonrelative was not evaluated using biochemical testing, it is impossible to know which of these scenarios was more likely. Regardless, this study provided fundamental information regarding E. coli in the adult human gut. Wallick and Stuart’s work was conducted at Brown University (Providence, RI, USA) and was interrupted by World War II when the study subject (and lead author) was drafted into the military. As an interesting and noteworthy historical coincidence, two German scientists, Kauffman and Perch, reported similar results the same year as Wallick and Stuart (1943) with data from two study participants, each followed for four months (30).

Sears et al. further explored E. coli serotype dynamics in five adult participants in two separate studies (31, 32). Participants were sampled for between 3 and 30 months, and the observed dynamics were again similar to those reported previously. Importantly, the authors defined E. coli serotypes persisting for “relatively long periods of time” as “residents” and those persisting for short periods (days to weeks) as “transients,” thus coining these terms that are still used today. In addition to confirming previous findings, Sears et al. also found that, in general, only one to two serotypes were present in an individual at any one point in time, meaning that most individuals host at least one resident E. coli. The authors attempted to determine the cause of transitions from one resident to another and found that diarrhea, gastroenteritis, and usage of certain antibiotics did not explain transition events in their study. Some participants even volunteered to ingest cultured isolates of known serotypes multiple times at high doses (10⁶ to 10⁷ CFU), but interestingly, none of these became established and sometimes were not even transiently observed. One of the participants even ingested 10⁸ CFU of a resident that they had previously hosted and lost. Again, the isolate did not become established. These studies clearly and elegantly demonstrated that E. coli residence in the adult human gut is not explained simply by exposure and suggest that other important factors are involved.

Sears et al. also tested the hypothesis that residents produced antibiotic substances (now known as bacteriocins) that prevented invasion of other clones “by possessing a wide range of antibiotic activity against other E. coli strains.” The authors made no comment on whether any of the clones were lysogens or if they produced viral plaques against other isolates. After screening several hundred strains collected during the study with a test similar to the interference assay used by Halbert (33)—a method similar to the double-layer plaque assay in which an antibiotic producer is dropped onto a lawn of target bacteria—notably, the authors found that only two residents produced widespread antagonistic activity. They concluded that the residents, as a whole, did not produce antibiotic activity more often than transient strains. Moreover, the most antagonistic isolates observed were, in fact, two transients. These results call into question the extent to which bacteriocins contribute to resident turnover.

Harriette Robinet added significantly to the work of Sears et al. in 1962, when she followed E. coli serotype dynamics in six adults and collected serum samples from each participant at the same time as stool samples (34). Robinet wanted to test the hypothesis that host antibody production caused resident E. coli turnover. To do so, she used a hemagglutination test to quantify serum antibodies active against resident isolates from the same individuals. Robinet concluded, “The study revealed that serotypes were lost and acquired in the bowel of hosts irrespective of high or low antibody levels specific for the antigenic types” (antigenic types = serotypes). Although not definitive proof, this study suggested that host antibody production does not explain resident E. coli dynamics. Subsequent to this study, Robinet’s group explored the relationship between resident turnover and bacteriocin production with the same isolate collection (35). They tested whether isolates collected during a 1-month period were active against E. coli collected the previous month. They also tested whether the isolates could inhibit a bacteriocin-sensitive control strain (E. coli phi). They found that residents produced bacteriocins more often than transients and that serotype diversity within samples was lower when residents produced colicins and higher when residents did not produce colicins. Although the authors commented on the fact that some of the E. coli inhibited clones present in the previous month’s sample, they did not state the frequency at which this was observed.

Shooter et al. questioned previous results based solely on O-antigen typing by including H-antigen typing to increase discriminatory power. In a 3-month study of nine middle-aged adults (22, 36), the authors found that many residents with an identical O-antigen carried different H-antigens, suggesting that earlier studies may have grouped together isolates that were actually different. Ultimately, this study illustrated the need for more discriminant techniques to identify E. coli clonal lineages (cells belonging to the same asexually reproducing population).

To increase discriminatory power and begin dissecting population genetic structures of bacteria, a technique called multilocus enzyme electrophoresis (MLEE) was applied to a unique E. coli collection in 1981. Using MLEE, Caugant, Selander, and Levin provided the first glimpse of the genetic structure of resident and transient E. coli from a single human adult over an 11-month period (37). Temporally collected E. coli were defined through the lens of multilocus genotypes and not serotyping, thus allowing the clonal structure of these bacteria to be quantified. Overall, results confirmed previously observed resident and transient patterns, but more importantly, the study ended an ongoing debate at the time as to whether recombination or the acquisition and loss of individual clones explained the large amount of genetic diversity observed among human E. coli (38–40). Caugant et al. demonstrated beautifully that clonal diversity was generated by the gain and loss of novel clones in the same host (37, 41). They also evaluated changes in plasmid and/or episomal DNA content using agarose gel electrophoresis within what they believed were identical clones isolated over time. Interestingly, extensive gain/loss events were observed. However, recent comparative genomics in our lab has shown that residents belonging to the same multilocus genotype but not the same clonal lineage (i.e., not belonging to the same asexually reproducing population) can cooccur and/or replace one another in human adults (5). So, in retrospect, and consistent with Shooter et al. (22), at least some results reported by Caugant et al. may still have been due to the lack of appropriate discriminatory power and not actual biology.

MLEE was later used to characterize E. coli dynamics in members of a British Antarctic Survey residing at the same research station (42). Study participants (n = 16, all males) lived in the highly controlled setting of an Antarctic base with no additional human contact over a 32-week period. E. coli bacteria were isolated from self-collected fecal swabs every two weeks, and a total of 1,447 isolates were collected. However, only 592 isolates grew after being transported to a laboratory in the United Kingdom, strongly suggesting isolation bias. Similarly, of the 592 isolates cultured, only 269 were identified as E. coli using biochemical testing (API-20E test kit), which again, is not consistent with other studies using similar isolation methods (i.e., the majority of isolates from MacConkey agar plates used in this study should have been E. coli [5]). These isolates were characterized by MLEE, assigned genotypes, and profiled for plasmid content. Resident clones (present in more than one sample) were observed in 14 of the 16 participants, and half of the individuals were colonized for most of the study period. None of the clones were observed for the entirety of the study, which again may have resulted from technical difficulties but is consistent with previously observed resident clonal turnovers (5, 28, 31, 37). Regardless, there was extensive sharing of clones between participants in this study, suggesting that interindividual transmission was common. Interestingly, shared clones were almost always resident in one participant and transient in other participants where they were observed, highlighting the possibility that they were uniquely adapted to one individual’s gut but not others. It should be emphasized that this study utilized methods identical to those of Caugant et al. (37), so the clonal resolution was similarly limited. Consistent with this potential limitation, the researchers found that isolates belonging to one of the clones observed in several subjects had three different plasmid profiles. While it is possible that this clone acquired new genetic material while residing within an individual or while spreading between individuals, it seems equally likely that what was believed to be a single, widespread clone was actually three separate ones.

TEMPORAL E. coli DYNAMICS IN THE PCR ERA

PCR significantly enhanced the discriminatory power of bacterial genotyping and began replacing MLEE for E. coli in the 1990s. E. coli phylogenetic groups, or phylogroups, originally defined by MLEE (40, 43, 44), were later resolved in greater detail with Sanger sequencing (multilocus sequence typing [MLST]) (45, 46) and, most recently, by whole-genome sequence-enabled “phylogenomics” (47, 48). As a result of these comparative phylogenetic approaches, one of the most often-used techniques for differentiating E. coli was a multiplex PCR protocol that effectively bins isolates into one of seven well-supported phylogroups (49, 50). Repetitive DNA analysis via PCR has also received considerable attention since the discovery of repeated DNA sequences interspersed between genes in bacteria (51, 52). A 35-base pair palindrome (51, 52) dubbed the REP (repetitive extragenic palindromic) sequence was shown to directly precede or flank certain protein-coding genes. Because (i) the REP sequence is distributed hundreds of times throughout nearly all E. coli genomes but at variable locations and (ii) short oligonucleotide primers successfully anneal to REP sequences (53), a single PCR produces dozens of amplicons that, when sized (agarose gel electrophoresis) and compared, can “fingerprint” isolates with discriminatory power approaching and often achieving clonal resolution. The REP sequence was the first of many intergenic DNA repeats detected in bacterial genomes (54), and other PCR protocols were developed (53, 55). For example, ERIC (enterobacterial repetitive intergenic consensus) sequences and simple trinucleotide repeats (GTG-5) have been used as fingerprinting PCRs (53, 55, 56). GTG repeats are the most common trinucleotide repeat found in E. coli genomes (57, 58), and the corresponding PCR has been shown to be the most discriminant to differentiate E. coli (55). All of the above PCR-based assays have been extensively used for tracking clones in temporal E. coli studies because of their relative ease of use and low cost.

Although fingerprinting PCRs provided a means to discriminate between clones, results are not easily comparable between institutions and do not correlate well with phylogenetic relationships (59). MLST overcomes at least some of these drawbacks and was first applied to differentiate phylogenetic lineages of Neisseria meningitidis (45) and, soon after, to E. coli (46). Isolates with identical sequences, called sequence types (STs), have been used extensively for epidemiologic purposes, and MLST databases are now available for different sets of genes or ST definitions (60–63). In general, MLST provided a much sought-after balance between discriminatory power and phylogenetic resolution.

Until the advent of (relatively) inexpensive and reliable full-genome sequencing, pulsed-field gel electrophoresis (PFGE) was the gold standard for epidemiologic source tracking during infectious disease outbreaks because of its ability to discriminate clones, reproducibility, and portability. This method utilizes restriction enzymes to cleave an isolate’s chromosome in specific places which are then size resolved with gel electrophoresis. PFGE is noted to be time-consuming and technically challenging (64, 65), and due to these drawbacks, this method is rarely used in temporal studies of human-associated E. coli.

Johnson et al. and Damborg et al. characterized temporal E. coli dynamics from adults and children of cohabitating family members and their respective pet dogs (19, 66, 67). Johnson et al. used MLST and PFGE to identify 14 E. coli clones from a family of five and their dog over a 3-year period. Clone dynamics within this study were difficult to discern given the infrequent sampling regimen (2- to 103-week gaps). Still, Johnson et al. found four clones that occurred in consecutive samples, suggesting that they were resident over a 4- to 45-week period. They also found that five clones were recurrent within an individual during the sampling period (i.e., the clones appeared in nonconsecutive samples), suggesting that recolonization was either common or that these clones fell below the limit of detection from time to time. Johnson et al. also observed clones that were shared among family members. Notably, they detected a clone shared between the family’s dog and four of five human family members. This particular clone caused a urinary tract infection (UTI) in the dog and belonged to ST73 and the phylogroup B2, both of which are commonly associated with UTIs in humans. The mother in the same household experienced a UTI during the study period, but the causative clone was another clone, ST95 (also phylogroup B2). This clone was also observed in samples from other family members, suggesting that it may have been transmitted.

Damborg et al. (67) temporally collected fecal swabs from 5 households (18 humans, 13 dogs) to describe E. coli clone dynamics within and between individuals living together. Ten fecal swabs were collected over a 6-month period with sampling intervals ranging from 3 to 36 days. After streaking swabs onto MacConkey agar, a single colony was randomly selected and characterized using amplified fragment length polymorphism (AFLP) to differentiate and group isolates and using the Clermont multiplex PCR to identify phylogroups. AFLP is a technique that uses restriction enzymes to digest chromosomal DNA, which is then ligated to oligonucleotides that act as a template for PCR amplification. These PCR products are then sized using gel-capillary electrophoresis. Damborg et al. identified 154 unique AFLP profiles (presumptive clones) from 322 isolates analyzed. Among the humans, they found that isolates belonging to phylogroup A were most often observed (32% of isolates). In addition, 21 clones were found to be resident, defined as present in a participant for greater than 3 weeks. Damborg et al. also observed clone sharing among family members and their respective pets and found that shared clones belonged primarily to phylogroups B2 and D, which are both commonly associated with UTI. Both studies (Johnson et al. and Damborg et al.) therefore provide evidence that clones may be acquired from cohabitating family members and animals. Whether shared clones are more likely to cause UTI remains an important epidemiologic question to be answered.

McBurney et al. assessed residency dynamics of Enterobacteriaceae from two adults (a 25-year-old female and a 50-year-old male) over a 1-year period (68). In this study, stool samples were collected on a monthly basis, and 10 to 15 colonies were picked randomly from MacConkey agar. The authors observed a large range of Enterobacteriaceae abundance (10² to 10⁹ CFU per gram of stool) and used ribotyping to understand the genotypic diversity. Ribotyping is a general genotyping technique that takes advantage of conserved nucleotide sequences in rRNA-encoding genes. Briefly, DNA was extracted, digested with restriction enzymes, and electrophoresed on an agarose gel. DNA was then transferred onto a nylon membrane and radiolabeled with oligonucleotide probes specific to 16S rRNA-encoding genes. The resulting banding patterns were visualized using autoradiography and grouped into “ribotypes” based on pattern similarity. Previous work by the group using PFGE supported that the ribotyping protocol was sufficient for identifying clonal E. coli populations. Ribotyping of isolates revealed that the male subject hosted a single resident clone, whereas the female subject hosted many different clones. Both individuals hosted resident and transient E. coli. Interestingly, the female subject underwent antibiotic chemotherapy (amoxicillin for 7 days) to treat a respiratory infection, which appeared to significantly impact Enterobacteriaceae diversity. Prior to the antibiotic, representative E. coli clones were found to be sensitive to ampicillin, doxycycline, and trimethoprim. After the antibiotic treatment, the E. coli clones present in her stool had changed and were resistant to these drugs. The authors concluded that amoxicillin selected for resistant E. coli, and while this is a possibility, amoxicillin did not cause the observed turnover of resident E. coli in this subject because all Enterobacteriaceae were below detectible levels for at least 4 weeks prior to chemotherapy. This study highlights the potentially important influence of antibiotics on temporal E. coli dynamics but was not designed to directly test for these effects. It should also be noted that ribotyping as described in this study has not been used in any other temporal E. coli study that we are aware of, so it is impossible to know how these results compare to those of other studies.

Following the work of McBurney, a Ph.D. dissertation describing the dynamics of antibiotic-resistant E. coli in healthy adults was published by Sashindran Anantham (36). We recognize that, as a dissertation, this work was not peer-reviewed to journal standards, but since it presents results that are relevant to temporal E. coli dynamics in human adults, we feel it is appropriate to highlight some of the more important findings. Anantham collected fecal samples from 11 adults over a 4-year period with the primary goal of determining whether antibiotic-resistant E. coli persisted in healthy humans even in the absence of antibiotic use. Most of the study participants provided fewer than 5 samples over the study period, but in contrast to McBurney et al., antibiotic-resistant residents were observed in the absence of antibiotic treatment. The long-term residents belonged primarily to the phylogroups A, B2, and D, and of the 73 clones identified, 34 (47%) were resistant to at least one antibiotic. Long-term residents also belonged to ST10 and ST95, which is similar to results from our lab discussed below (5, 69). Anantham also quantified the carriage of genes expressing so-called virulence factors, or factors associated with E. coli pathogens. It is important to note that just because a factor is cited or referred to as a virulence factor, only a subset of these factors have been experimentally validated to influence virulence in humans and most are defined simply due to the fact that pathogens carry the gene(s) or that they influenced virulence in an animal model. In other words, these factors likely play important roles beyond (and independent of) virulence. Regardless, over 80% of the long-term residents identified by Anantham carried genes encoding for fimbriae, capsule biosynthesis, and a siderophore receptor previously associated with E. coli pathogens, suggesting that these particular factors may be better referred to as residency as opposed to virulence factors per se. Remarkably, one of the study participants traveled abroad for a year, was treated with the antibiotic tetracycline, and yet did not lose the tetracycline-sensitive resident they originally hosted. Since both antibiotic treatment and travel were found to displace residents in previous studies (32, 70), it is possible that the study subject was either recolonized by this particular resident or that travel and tetracycline have minor effects on clonal turnover.

Work in our lab recently focused on temporal microbiome and Enterobacteriaceae dynamics in a cohort of eight healthy adults (5). Participants were sampled on a ∼biweekly basis for 6 months to 2 years. Stool samples were plated onto MacConkey agar, and 95 isolated colonies were analyzed per sample (both lactose-fermenting and nonfermenting colonies were included). Our protocol identified (i) all (or nearly all) resident clones in each of the eight subjects, (ii) the species-level identity of all resident clones, and (iii) the phylogroup that each resident E. coli belonged to. A few important results from this study are as follows. Of the 32,470 isolates evaluated, 87% were E. coli, which clearly demonstrated the dominance of this species over other Enterobacteriaceae in healthy human adults. Isolates belonging to phylogroups A, B2, and F were more likely to be residents, suggesting that these phylogroups are better adapted to the human gut. Surprisingly, residents often did not ferment lactose, which is noteworthy because lactose-negative isolates on MacConkey agar are often not recognized as E. coli. Finally, and perhaps most significantly, the amount of clonal turnover varied markedly between study participants. For example, one participant simultaneously hosted three distinct residents for over 400 days, whereas another subject hosted primarily one resident at a time, sometimes with 30- to 50-day gaps between any E. coli being observed at all, and none of the residents persisted longer than 54 days in this individual.

Temporal E. coli studies in nonhuman mammals are rare, especially at the bacterial clone level and using data where phylogenetic relationships can be drawn. Such studies hold the potential to help understand the impacts of residency in the face of different host genetic, environmental, and ecological factors. Consistent with our study in humans, a capture-recapture study of brushtail possums found that E. coli residents primarily belonged to phylogroup B2 (71). Based on the identification of residents in subsequent samples (i.e., observed in the same individual in subsequent time periods ranging between 1 and 7 months), Blyton et al. (71) reported that just over one-third (36%) of clones detected within an individual at one time period were reisolated the following time period. Based on this result, they concluded that E. coli organisms were rapidly gained and lost. This result is not entirely inconsistent with results we reported in humans over roughly the same time periods, and it is interesting to point out that 65% of the individual possums (11 of 17) sampled at all four time periods ended the study with the same clone they began the study with. In short, one could draw significant parallels between results found in humans and at least this nonhuman mammal.

In summary, resident E. coli have been identified in temporal studies of human adults dating back well over 100 years and using a variety of microbiologic and molecular methodologies. It is becoming clear that E. coli strains are not equal with respect to their ability to colonize and reside in the adult human gut and that phylogroups A, B2, and F are better adapted to this environment. Interestingly, STs often associated with urinary tract infections are commonly observed in healthy humans in cross-sectional and temporal studies (5, 72). For instance, STs 73, 95, and 131 are commonly found in healthy human adults without causing overt symptoms of disease (72, 73). Residence time and residency dynamics differ markedly between individuals, and potential explanatory factors, such as travel, antibiotic use, phage/bacteriocin production, and acute episodes of diarrhea/gastroenteritis do not always lead to clonal turnover. Clearly, more research is needed to understand this important aspect of microbial ecology.

THEORETICAL CONSIDERATIONS FOR AND MECHANISMS OF CLONAL TURNOVER

Many factors seem to alter the genus and higher taxonomic composition of the adult human gut microbiome, including antibiotics, heavy metals, diet, travel, and disease (70, 74, 75). However, and although not investigated in nearly enough detail, resident E. coli clones often turn over without an obvious cause, leading to the question, “Is resident turnover a random process?” Some studies suggest turnover is not random and that competition between different clones is an important factor (76, 77). In the most general terms, the competitive exclusion principle, or Gause’s law, proposes that two organisms cannot coexist on the same limiting resource (78), and for the purposes of an ecological framework for clonal interference, this law seems like a reasonable place to begin. That said, this principle establishes both the type (negative) and outcome of the interaction (one excludes the other), but it does not identify the mechanism(s) involved. Both direct and indirect “competitive phenotypes” exist in bacteria (79), where direct competition includes damaging or killing competitors (e.g., via phage or bacteriocin production) and indirect competition includes passive resource utilization (e.g., the more rapid use of a carbon source or occupancy of colonization sites). While there is little consensus as to which competition mechanism, if any, is the most important for resident turnover in the adult human gut, we present evidence for the competitive mechanisms that have been most commonly explored. It should be noted that multiple modes of competition may play a role simultaneously in turnover events in the gut. We also discuss possible environmental selection unrelated to competition that may be relevant to clonal dynamics. For example, it is possible that resistance to small molecules/compounds produced by noncompeting members of the microbiome and/or resistance to abiotic factors are most important. These possibilities are also discussed.

NUTRIENT COMPETITION

The nutrient niche hypothesis proposed by Rolf Freter suggests that competition for nutrients selects for species that most efficiently use them, making these more likely to colonize and persist (76, 80, 81). This hypothesis was supported by experiments performed in continuous-flow reactors seeded with a mouse cecal microbiome (80, 81). The primary force controlling the abundance of E. coli within reactors was the concentration of a specific nutrient that members had specialized on (81). Inhibitory substances produced by other microbiome members were found to have an effect on E. coli growth but did not appear to be the primary driving force affecting overall abundance (81). Thus, in general terms, Freter believed the overall presence-absence of E. coli and other species (i.e., richness) of the gut microbiome was explained by interactions between inhibitory substances and nutrient availability.

E. coli does not readily colonize the gut of conventional lab mice, most likely due to an innate function of a mature, climax microbial community, termed “colonization resistance.” Streptomycin pretreatment ameliorates this resistance, allowing Enterobacteriaceae to colonize and in some cases to cause disease (82). Cohen and Conway used a streptomycin-treated mouse model comprising 24-hour access (ad libitum) to streptomycin in drinking water to introduce E. coli strains into the gut of CD-1 mice and study nutritional requirements of individual bacterial strains. They demonstrated that carbohydrate metabolism predicted gut colonization success (83). Several studies published by this group showed that E. coli colonized the mucus layer of the murine epithelium, and much of their work focused on catabolism of sugars available in mucus, such as arabinose, fucose, galactose, gluconate, hexuronates, lactose, mannose, N-acetylglucosamine, N-acetylgalactosamine, N-acetylneuraminate, ribose, and sucrose. Mutant strains lacking specific catabolic pathways were generated and competed in vivo against isogenic wild-type counterparts (83). These experiments showed that only a subset of the total carbon sources an E. coli clone can possibly catabolize are actually utilized in the gut, suggesting that the realized niche is indeed narrower than the fundamental niche of E. coli (84, 85). Cohen and Conway also showed that a nonpathogenic strain could competitively exclude a pathogen only if the strains utilized the same carbon sources. Another important suggestion from this body of work was that E. coli’s metabolic niche is relatively stable because mucus is host derived and not subject to the compositional changes due to shifts in diet.

Although Cohen and Conway’s work significantly advanced current understanding of nutrient competition in the gut, it is possible that host response played more than a minimal role in the observed outcomes. For example, it has since been recognized that streptomycin causes inflammation in the murine gut, which alters the expression of inducible nitric oxide synthase which, in turn, alters levels of secreted sugars in the epithelial mucus layer (86, 87). Under normal conditions, the levels of oxidized sugars such as galactarate and glucarate were low compared to levels following streptomycin treatment. Furthermore, streptomycin-induced alteration of epithelial and immune cell function (akin to low-grade inflammation [87]), resulted in increased levels of the E. coli electron donor, lactate, and electron acceptors, oxygen (aerobic) and nitrate (anaerobic). This shift in nutrients allowed E. coli and other Enterobacteriaceae to outcompete strict anaerobes and increase in abundance. Such alteration to the metabolic landscape raises legitimate questions as to whether the streptomycin mouse model is the most appropriate model for studying natural clonal dynamics in the context of homeostasis (i.e., nondisease conditions). It should be noted that the above-mentioned streptomycin-induced host responses have only been reported in C57BL/6 mice following a one-time oral gavage. Even though mice receive comparable levels of streptomycin in both models, it is possible that the CD-1 mice used in many studies by Cohen and Conway responded differently (e.g., did not develop an inflammatory response). Regardless, more studies are warranted in both models to determine whether antibiotic-induced host responses confound or preclude natural clonal E. coli dynamics observed in humans.

The famous E. coli strain, K-12, is widely used as a representative “commensal” clone, and the use of such strains minimizes variability between studies, making results more generalizable. K-12 was initially derived from a human host (a convalescent diphtheria patient in 1922 [88]), but prior to the widespread use of lyophilization and freezing, many mutations emerged in different laboratories and clone collections. For example, preservation in stab cultures is documented to promote mutations by insertion sequences/transposable elements (89). Among labs and between strain collections, mutations have accumulated that may mask their true (original) phenotypes. For example, MG1655 (a K-12 sublineage) carries a mutation that reduces the expression of genes associated with pyrimidine metabolism, ultimately resulting in decreased growth in minimal medium lacking uracil compared to other K-12 sublineages (90). It is at least worth noting that many groups rely on K-12 as a stand-in for a “commensal” strain, which may or may not contrast with primary isolated, “wild” E. coli from human subjects.

Competition for trace metals has also been demonstrated to be important in competitive exclusion in the gut. For instance, one of the mechanisms of action of the probiotic E. coli Nissle 1917 is suggested to be production of siderophores (91). Siderophores are small molecules produced by polyketide synthases that capture free iron. Since iron is an essential micronutrient for nearly all living cells, it is feasible that levels in the gut either are or transiently become low enough to become limiting, especially during periods of stress (92). In murine models of colonization, E. coli isolates with siderophore genes removed were found to have reduced ability to colonize compared to their wild-type counterparts (93). Anantham and Wold found that the ability to produce siderophores was common among resident E. coli in humans (36, 94). Thus, it appears that competition for iron may play an important role in resident turnover.

Finally, the nutrient niche hypothesis has been criticized because it neglects how variation in the concentration and composition of nutrients in the gut over time affects niche stability (76). Since nutrient composition of the gut changes dramatically throughout the day (between meals) as well as between days, it is difficult to understand how a single nutrient could be responsible for long-term residency. Little is known about the spatial heterogeneity (patchiness) of the gut within and between individuals, but Pereira and Berry suggested that successful residents must have the metabolic flexibility to utilize different nutrients as they become available (76). Alternatively, Cohen and Conway proposed “the restaurant hypothesis,” which suggests that E. coli forms cooperative relationships with strict anaerobes, such as Bacteroides thetaiotaomicron, which break down complex polysaccharides present in the GI environment (83). With these findings taken together, whether and/or the degree to which competition over limiting nutrients drives population-level diversity in the human gut requires additional investigation.

COMPETITION FOR PHYSICAL SPACE

At the population level, Freter suggested that physical space was a key determinant of clonal diversity in the gut (80). Results from mouse experiments showed that single E. coli strains could not invade or persist when orally inoculated into conventionalized germfree mice (i.e., germfree mice colonized with a full microbiome by cohousing with other, conventional mice), whereas the same strains readily colonized when inoculated into germfree mice two days prior to conventionalization (80). Since the E. coli strains being evaluated were able to colonize/establish, the colonization resistance observed in conventionalized mice cannot be easily explained simply by competition of limiting nutrients (i.e., it is assumed the same nutrients existed in all conventionalized mice). The evidence that bacterial “adhesion sites” were the determining factor came from a mathematical model that showed that fitness for a limiting nutrient had little, if any, influence on colonization when physical space was unavailable. In fact, Freter’s model predicted that even less-fit clones (i.e., clones with lower fitness than the clone already residing in the gut) would be able to invade and persist if adhesion sites were present. It should be noted that Freter provided no in vitro or in vivo data to support the role of adhesion sites.

Although the surface area of the human GI tract is roughly 32 square meters (the equivalent of over seven and a half ping pong tables) (95), space in the gut has long been argued to be a limiting resource for microbes. Similar to siderophores, attachment factors, such as adhesins and fimbriae, are enriched in human-associated E. coli (36, 94). Remarkably, application of a high-affinity, small-molecule inhibitor of mannose binding by the type 1 adhesion pilus, FimH, appears to be sufficient to dramatically reduce at least some E. coli in the murine gut (96). Adherence sites in the gut are likely more diverse than simply sites on epithelial cells. Similar to nearly all environments where microorganisms are found (97), the human gut likely supports a substantial component of microbial diversity in an epithelial-associated biofilm, or a surface adhered structure composed of microbes embedded in a matrix. Biofilm dynamics are not the same as those of free-living, planktonic cells, and it is well known that biofilm-associated bacteria (i) are more resistant to stressors (e.g., osmotic imbalance, antibiotics, predation), (ii) stratify into subpopulations with markedly different physiologic and metabolic states, (iii) increase horizontal gene transfer, and (iv) are protected from physical forces such as shear. It is difficult to study intestinal biofilms in vivo, so the contribution of these structures to human health remains controversial. Still, shifts in polymicrobial biofilm dynamics seems to be associated with GI diseases such as colorectal cancer and inflammatory bowel disease (98). Certain models of the gut also call into question the relevance of GI biofilms because the rate of gut transit, mucus production, and epithelial cell turnover may be too great to allow a stationary structure like a biofilm to replenish itself before being washed out (98). That said, complex biofilms have been observed in the appendix of healthy individuals and perhaps, due to the lack of flow through this anatomical site, the mature biofilm that forms there acts as a “seed bank” to repopulate the gut after acute disruption (e.g., diarrhea, inflammation, antibiotic treatment) (98, 99). Biofilm-like structures have been reported in many mouse models, and Cohen and Conway suggested that life for nonpathogenic E. coli in the human gut is as a biofilm (83).

Goblet cells in the human epithelium produce and secrete a matrix, called mucus, that might promote establishment of a polymicrobial, mature biofilm (76, 100). Interestingly, the number of goblet cells increases in the distal GI tract, so their abundance correlates well with overall bacterial abundance. In the colon, two distinct layers of mucus have been described, where the inner layer, directly overlying the epithelium, is thick (>200 µm in humans) and impenetrable to bacteria, while the outer layer is more diffuse, compositionally different, and amenable to microbial colonization (101, 102). Defects in mucus production resulting in decreased thickness in the colon lead to increased bacterium-epithelium contact and inflammatory diseases. Interestingly, there is evidence that the overall structure-function of the mucus layer depends on the taxonomic structure of the microbiome (103, 104). The taxa involved, molecular cross-talk, and this overall phenomenon have not been well studied in humans. Regardless, if it is true that the human colon is protected from bacterial colonization by a thick, impenetrable mucus layer, then attachment to epithelial cells would play a minor role in microbiome residency, and attachment to sites on the epithelial cell surface (e.g., mannose) would not be conducive to survival. More research is needed to understand this fundamental aspect of the gut ecosystem.

Microorganisms also produce biofilm matrix components that may be important for a gut-biofilm lifestyle. Bokranz et al. screened for the production of biofilm components in a collection of 52 E. coli clones isolated from 16 adults and 5 children (105), including extracellular polymeric substances (cellulose) known to confer structural cohesion and amyloid structural proteins (curli) known to mediate adhesion between cells and surfaces. Bokranz et al. used LB agar lacking salt and supplemented with Congo red (an amyloid protein stain) and calcofluor white (a fluorescent dye that binds cellulose) to identify E. coli with these phenotypes along with Western blotting and matrix-assisted laser desorption ionization–time of flight (MALDI-TOF) to confirm the presence of curli. This was a cross-sectional study, and many of the participants were cohabitating family members that shared E. coli clones (identical PFGE types). Representative isolates of some of the shared clones showed different curli and cellulose expression, suggesting that these traits may have changed during residence within and/or transmission between individuals. Because biofilm assays were performed in vitro, these results say little about in vivo relevance. However, this study did support a link between E. coli from healthy human adults and biofilm formation. Interestingly, clones belonging to the phylogroup B2 (a phylogroup commonly observed as human residents) have been found to form biofilms more quickly than clones belonging to other phylogroups (106), a phenotype that could potentially explain residency.

White et al. (107) used a similar assay to screen for biofilm production in natural populations of E. coli (84). One objective of their study was to determine whether biofilm production may be a signature of host adaptation, and E. coli isolates from several sources were compared (humans, clinical samples and healthy donors; animals, birds and nonhuman mammals; and contaminated water). Contrary to results from Bokranz et al. (105), the human E. coli isolates in the White et al. study had a reduced capacity to produce biofilm components (curli/cellulose) compared to isolates from nonhuman sources. Interestingly, isolates belonging to phylogroup B2 from humans produced biofilm components less often than B2 isolates from nonhuman sources. Such an observation that accounts for phylogenetic relatedness is powerful because of the large interisolate differences in phenotypic/genotypic composition of E. coli. Results from White et al. (i.e., human-adapted E. coli have decreased biofilm-forming capacity) suggest that perhaps mucus production by the human epithelium provides sufficient matrix material for the microbiome and that a beneficial adaptation by gut bacteria may be to downregulate or otherwise shut off genes in this pathway. More experiments are needed to test this intriguing hypothesis, but if supported, mucus production may indeed turn out to be the most important determinant of the human gut-microbiome symbiosis.

DIRECT DAMAGE AND/OR KILLING OF COMPETITORS

Negative microbe-microbe interactions in the human gut microbiome are diverse and lead to variable outcomes. Unlike the indirect or passive competition mechanisms discussed above, bacteria produce factors that directly inhibit other bacteria, having both narrow and broad spectrums of activity. As such, at least some of these factors in at least some contexts have little to do with true competition because cells producing them and cells being affected by them do not interact and thus do not compete. Some factors seemingly made for the explicit purpose of inhibiting competitors include diffusible toxins (bacteriocins), contact-dependent toxin delivery systems (type VI secretion systems [T6SS]), and phages (79, 108–110). Remarkably little is known about the overall influence and outcome of these factors in E. coli populations over short and long time scales and, specifically, in the context of resident turnover. Here, we focus on the potential influence of bacteriocins and prophage, as these two factors have been at least superficially explored with respect to human-associated E. coli. Not surprisingly, and as briefly mentioned above, reported results are not in agreement (31, 35), and more studies are needed to understand the true impact of direct inhibition in structuring microbiome populations and communities.

Bacteriocins

Bacteriocins are antimicrobial peptides that target and inhibit other members of the same species or members of other species of bacteria. E. coli bacteriocins can be divided into colicins and microcins (109, 111), but it is important to note that the nomenclature is inconsistent and, at times, contradictory. Size is one property used to differentiate colicins (30 to 80 kDa) and microcins (<10 kDa) (112), but due to their genetic/functional diversity, these rules do not always apply (109, 112).

In 1925, Gratia discovered that a heat-labile product of E. coli V could kill E. coli ɸ (113, 114). Later, Gratia and Fredericq showed that these factors were composed of polypeptides and called them colicins (114, 115). Colicins are released by cell lysis or active secretion, and due to associated fitness costs (111), only a subpopulation produces them at any point in time (i.e., expression is tightly regulated). Stress such as increased temperature and DNA damage can induce colicin production in vitro and in vivo (111, 114, 116). Once released, a colicin will bind to a receptor (typically a nutrient transporter) located on the target cell surface such that resistance due to receptor modification often results in reduced fitness (111). After binding, colicins either depolarize their targeted cell by pore formation or inhibition of cell wall synthesis or are transported into the cytoplasm, where they degrade nucleic acids (114). With colicin E1, a single cell induced and then lysed is estimated to release 100,000 colicin peptides, and other colicins release far more (111). Once these peptides find target cells, their capacity to kill is great, as their activity follows single-hit (i.e., one colicin-one target cell) kinetics (111, 114).

Some microcins are induced under low-iron conditions and released by active transport such that host cell lysis is not required for dissemination. Under such conditions, host cells upregulate expression of siderophores and siderophore receptors to capture free iron (112). Microcins selectively bind to siderophore receptors of competitors and kill them by depolarization or DNA/RNA degradation (112). Microcin production has been reported to be essential for the probiotic activity of E. coli Nissle 1917 and competition in an inflamed gut environment (92, 117). Sassone-Corsi et al. found that in mice with dextran sodium sulfate (DSS)-induced colitis, Nissle-produced microcins were capable of killing nonpathogenic/mouse-adapted E. coli, adherent-invasive E. coli, and Salmonella enterica (117).

Estimates of bacteriocin production among E. coli range from 10% to >70%, depending on the isolation source (111). E. coli in low-cell-density environments such as freshwater appear to have a lower frequency of bacteriocin production than E. coli in high-cell-density environments such as the gut (111). It is possible that competition is increased in high- versus low-density populations, but low-cell-density environments may also be nutrient poor, thus making competition fiercer. Alternatively, diffusible toxins such as bacteriocins are more likely to contact target cells when population density is high, so this overall pattern may simply be due to minimizing the costs and maximizing the benefits of production.

Gordon et al. screened for bacteriocin production in a collection of 266 E. coli isolated from 266 Australian children and adults (116). Only 30% of participants (n = 77) were asymptomatic university students, and the remaining (n = 189) were hospitalized patients with symptoms of gastrointestinal disease. Thus, results were heavily biased toward pathogenic E. coli and/or nonpathogenic E. coli in patients with active gastroenteritis. With this in mind, bacteriocin production in mitomycin C-treated cells (i.e., under DNA-damaging stress) was quantified using two susceptible K-12 derivative strains and further differentiated from phage production using at least four different comparative approaches (trypsin treatment, freeze-thaw, size filtration, and dilution). Finally, bacteriocins were phenotyped and genotyped to evaluate the distribution of individual types. Of the 266 isolates evaluated, 38% (n = 102) produced at least one bacteriocin. Interestingly, of the isolates that produced bacteriocins, less than half (42%) produced only a single type, suggesting that most bacteriocin-producing E. coli produce bacteriocin “cocktails.” If individual bacteriocins in cocktails overlap in their specificity for target cells but use different receptors, this would theoretically decrease the likelihood that competitors evolve resistance. However, E. coli capable of producing a bacteriocin are also protected from its activity, so it is also possible that bacteriocin cocktail production is a result of this selective pressure (111). More experiments are needed to clarify this interesting pattern of activity. Finally, E. coli belonging to the phylogroup B2 were more likely to carry multiple bacteriocins compared to other phylogroups, and since more human residents belong to this group, it seems reasonable that stability in the gut is associated with bacteriocin production. As above, this hypothesis requires additional testing.

In vitro experiments and theoretical modeling suggest that bacteriocin production is an effective means to invade structured environments such as biofilms, more so than unstructured environments such as a shaking culture (118). In these studies, if the concentration of bacteriocin was not high enough, it was ineffective against its competitors (119). Perhaps resident E. coli of the human gut living in biofilms kill off competitors by producing bacteriocin cocktails at sufficiently high concentrations, whereas bacteriocin-producing would-be biofilm invaders fail to do so because their killing activity remains too dilute. Controlled gnotobiotic mouse modeling with engineered/bacteriocin-defined E. coli may help clarify these possibilities.

Phages

Bacterial viruses (bacteriophages or phages) were independently discovered by Twort and d’Hérelle in 1915 and 1917, respectively (120–122). Although Twort was the first to observe hallmark phage plaquing phenomena, d’Hérelle identified the parasitic, self-replicating nature of phages. During World War I, while investigating a particularly severe outbreak of dysentery (123), d’Hérelle mixed cultures of the dysentery-causing bacterium and filtrate from the stool of symptomatic patients. After plating, he observed glassy, bacteria-free zones (plaques) that he could propagate with naive, susceptible bacteria that further replicated the phages.

Although perhaps an oversimplification, phages can be separated into two groups: lytic and lysogenic (120). Lytic phages undergo a cycle of infecting host cells, replicating, lysing host cells, and releasing to find new host cells. Lysogenic (temperate) phages also infect, replicate in, and lyse host cells but can also integrate their DNA into the host’s chromosome or otherwise be maintained inside host cells as extrachromosomal (episomal) DNA. In this state, phages lie dormant until induced to replicate and lyse once more. As with bacteriocins, stress (e.g., UV irradiation, oxidative stress, nutrient deprivation) is an often-cited inducing event for lysogenic phages, and host cells carrying the same or similar lysogenic phage (lysogens) are immune to reinfection.

Furuse et al. (124) showed that the prevalence and type of E. coli phages differed markedly in stool samples from healthy adults and patients with leukemia. Using the phage-sensitive indicator strain, E. coli C, they found that 14% (15/109) of the leukemia patients shed >10⁵ phages (quantified as plaque-forming units [PFU]) per gram of stool, compared to only 1.6% (6/371) of healthy adults. They also found that phages in stool samples with low phage density (PFU/g stool) were mostly lysogenic phages, while phages from high-phage-density stool were mostly lytic. More recent work supported this observation (i.e., patients tend to have higher phage titers in their feces) (110). Furuse et al. also collected stool samples from 19 healthy adults every 2 weeks for ∼4.5 months and found that phages isolated from filtrate were primarily lysogenic. They found that phages were antigenically identical within individuals and remained at a relatively consistent density. They also isolated individual E. coli from some samples, UV irradiated them to induce lysogenic phages, and compared induced phages to those recovered from filtrate. They found that induced phages were antigenically similar to those recovered from the stool, suggesting that healthy adults normally shed a low level of lysogenic E. coli phages. Lysogenic phages may act as “self-replicating weapons” to kill competitors (77, 108), and like bacteriocins, they may be effective killing agents of would-be invaders (118). Somewhat unlike bacteriocins, experimental and theoretical work indicates that lysogenic phage production is an effective means to invade complex communities even when the lysogen is at a low concentration (118). Thus, current evidence seems to support that phages are more likely to play a role in clonal turnover in the gut, while both phages and bacteriocins likely have a role in restricting invasion by incoming transients.

RESPONSE (OR THE LACK OF) TO CELLULAR STRESS

Biofilm (curli and cellulose), bacteriocin, and phage production, as well as scavenging for nutrients (e.g., iron), carbon source utilization, and response to low pH, are stress-associated phenotypes that in E. coli and the Enterobacteriaceae are regulated by the alternate sigma factor, RpoS (125). Contrary to the hypothesis that human-resident E. coli have an increased ability to respond to stress, interestingly, White et al. (107) found that less than half of E. coli isolated from humans (n = 51 of 115) displayed a fully active RpoS phenotype (negative for glycogen production and catalase activity) compared to 90% (n = 151 of 169) of isolates from nonhuman sources (cows, birds, other nonhuman mammals, and water) (84). While this observation (fewer E. coli with active RpoS phenotypes from humans versus nonhuman sources) held across all phylogroups, isolates belonging to phylogroup B2 were the most disparate (33% in human isolates versus 91% in nonhuman isolates) and statistically significant. These results, as White et al. discussed, suggest that an RpoS-driven stress response is deleterious for long-term survival in the human gut. With respect to biofilm production, if bacteria rely more on host mucus as the primary biofilm matrix, there may be a distinct advantage to mitigating rpoS expression and shutting off production of bacterial matrix in the gut. In addition, some E. coli with wild-type rpoS alleles have reduced metabolic capabilities (i.e., number of carbon sources utilized) compared to their isogenic rpoS-null counterparts (126), so shutting off rpoS expression may broaden their nutrient niche. All things being equal, the human gut is more stable than environments outside the gut (e.g., temperature), making global stress responses more costly with respect to overall fitness. White et al. did not consider temporally collected E. coli, which will be an important consideration for future studies.

CONCLUSIONS AND FUTURE DIRECTIONS

In the 135 years since E. coli was first identified, only a handful of studies have attempted to describe the temporal dynamics of E. coli populations in healthy human adults. As this research area continues to emerge for E. coli and other microbial inhabitants of the human gut, it is important to organize/compare observations and experimentally identify/validate mechanisms governing the gut ecosystem. To this end, and from our perspective, we list fundamental hypotheses according to the amount/quality of evidence supporting them with the hope that hypotheses under each category are added, pruned, and/or revised with future research. Finally, we summarize a few important areas where experiments are needed to help clarify important aspects of E. coli population dynamics in the adult human gut.

Well-supported hypotheses are the following:

• E. coli is either resident and able to establish in the gut or transient and lost at the rate of gut transit.

• E. coli residency is dynamic, and resident clones turn over at different rates in different individuals.

• Resident E. coli clones are shared among cohabitating humans and their pets but are not necessarily resident in all individuals.

• Resident E. coli clones are rarely shared among noncohabitating individuals.

Hypotheses requiring further testing are the following:

• Residency in the human gut is not a random process and is determined by definable factors.

• Long-lived resident E. coli are adapted to the gut of adult humans and belong to phylogroups A, B2, and F.

• Competition for nutrients is the most important factor determining E. coli residency in the adult human gut.

• Competition for space is the most important factor determining E. coli residency in the adult human gut.

• The ability to directly damage and/or kill competitors is the most important factor determining E. coli residency in the adult human gut.

• Adaptation to environmental factors is the most important factor determining E. coli residency in the adult human gut.

NEEDED AREAS OF RESEARCH

Conceptually speaking, an organism’s decision to migrate or reside is a function of environmental condition-dependent fitness costs and benefits (127, 128). Specialization to conditions in the gut may limit a resident’s ability to colonize new hosts due to the loss of fitness in environments and conditions between them. Moreover, the population dynamics of E. coli appear to be very different from other microbiome species, such as Bacteroides fragilis, which is highly uniform with little or no turnover (129). Additional research is needed to clarify whether and how often prominent members of the human gut reside or migrate to other environments.

Several temporal studies have documented sharing of E. coli clones among cohabitants, including pets (19, 31, 42, 67). Even though there is little evidence that sharing leads to resident turnover, cohabitants represent a potentially important source of new residents that need to be further investigated (130). It is also possible that new residents are acquired from other sources (e.g., water, food, or contact with pets/other humans), but studies comparing E. coli inside and outside the gut consistently report marked differences (genotypic and phenotypic) in the populations found there, which decreases the likelihood of environmental acquisition. Although it is difficult to assess, more studies are needed to understand whether human-resident E. coli are acquired from other humans, other mammals, or other sources altogether.

Genome sequencing will play an important role in future temporal studies. Only two studies (131, 132) to our knowledge have sequenced E. coli clones isolated temporally from healthy human adults. Both studies found that the mutation rate was relatively low (2.26 × 10^–7 to 6.90 × 10^–7 mutations per base pair per year) and reported little to no evidence of selection, suggesting that random drift may explain the bulk of accumulated population genetic diversity in the gut environment. This suggestion is in stark contrast to recent comparative genomic results reported for B. fragilis (129), where polymorphisms were interpreted to be adaptative. Future comparative genomics studies promise to shed much-needed light on the impact of selection on residency and overall population genetic structure.

Many studies reviewed above, including our own, found that resident E. coli in human adults belong to the phylogroup B2. Our group also found E. coli of phylogroups A and F to be long-lived residents in human adults. Additional studies are needed to understand whether shared versus clone-specific traits result in the statistically significant overabundance of these phylogroups in the human gut. For example, there is evidence from studies of infants and adolescent children that virulence factors in pathogenic E. coli are important “residency factors” in nonpathogenic E. coli (94, 133, 134). Gene association studies are warranted in this context to better define residency-associated traits.