The first attempt to assess the genetic structure and diversity of E. coli was made by Milkman, who analyzed 829 isolates obtained mainly from humans (26). Using multilocus enzyme electrophoresis (MLEE) based on four loci, he determined that the average genetic diversity (H) of this species was 0.23. Subsequent studies that extended the work of Milkman were primarily concerned with the genotypic and phenotypic variation among strains isolated from the commensal fecal flora of humans and those responsible for neonatal septicemia, cystitis, pyelonephritis, and acute diarrhea (6, 38, 41, 47, 49, 50). Other studies have focused on various aspects of this species’ natural history, such as the turnover of strains in a single host (5), the sharing of clones among hosts (6), and the genetic structure and diversity of E. coli in its primary and secondary habitats (33, 46, 48). These and other studies have formed the basis of the clonal paradigm for the genetic structure of bacterial populations (10, 15, 16, 19, 25, 45) and have been reviewed by Selander et al. (42) and by Whittam (50).

Estimates of genetic diversity obtained by using strains isolated from human fecal samples range from 0.45 to 0.54 (42, 50). However, there is a caveat associated with this observation. The majority of strains from humans were isolated from people living in the developed countries of the West. Although some studies have included strains from Tonga, even these are suspect, since Tonga was used as a military base by both U.S. and New Zealand forces during World War II (4). It is therefore possible that the diversity of E. coli from humans is also underestimated. A high degree of genetic diversity (H = 0.61) has been found in E. coli isolated from sewage (33). These sewage samples undoubtedly represent two quite distinct sources of strains: those from the species’ primary habitat, the lower intestinal tract, and those from its secondary habitat, the sewage environment. Whittam and collaborators (46–48) have shown that sympatric E. coli populations inhabiting primary and secondary habitats are quite distinct from one another in terms of their clonal composition (50).

The objective of this study was to extend our understanding of the variation and genetic structure of E. coli by examining strains from a large variety of wild mammalian and avian hosts. We assessed genotypic diversity, phenotypic diversity, and genetic relatedness of 202 E. coli strains by using 11 loci in MLEE, plasmid profiles, resistance to six antibiotics, production of two toxins, and the utilization of 12 sugars. In this study, we analyzed how these genetic and phenotypic characteristics varied by host taxonomic group and geographic origin. E. coli was sampled mostly from wild mammals, and strains were taken from 81 species representing 39 families and 14 orders of mammals (31) in the Americas and Australia. We also included a sample from a reptile and 10 strains from birds from Mexico. As a reference, we also studied six strains from African baboons (35), 13 strains from the ECOR collection (32), and strain K-12 (2).

Bacterial strains. The complete list of hosts sampled by diet and geographic origin as well as their taxonomic classification by order and proximity to human environment is presented in Table A1 in the Appendix.

TABLE A1 Host characteristics and geographical origins of E. coli from wild mammals andbirds

Strain isolation. Strains collected in Mexico, Costa Rica, and Venezuela were recovered from captive or wild mammals and birds. Fecal samples were transported in swab-transport system containing Aimes media (Difco). Mexican samples were suspended in 1 ml of Luria broth and incubated; an aliquot was then streaked for single colonies on a minimal lactose plate. More than one isolate was sometimes taken from a single host individual. The Mexican samples were processed as follows. After incubation, the Lac⁺ colonies were tested for growth on minimal citrate plates. The Lac⁺ Cit⁻ colonies were then tested to confirm that they matched the biochemical characteristics of E. coli: gas production positive, H₂S negative, urea negative, methyl red positive, Voges-Proskauer negative (13). The Australian strains were collected and cultured by David Gordon. These strains were isolated from anal or cloacal samples or from fecal samples from wild mammals. A primary isolation of the strains was carried out by streaking the sample on a MacConkey plate to obtain single colonies. Only one isolate was taken from each individual host. Subsequently, single colonies from each plate were restreaked twice onto MacConkey plates. Colonies morphologically consistent with E. coli were then tested for growth on minimal lactose and minimal citrate plates. All Lac⁺ Cit⁻ colonies were tested to confirm that they matched the biochemical characteristics of E. coli: phenylalanine negative, H₂S negative, urea negative, indole positive, methyl red positive, Voges-Proskauer negative (13). In Australia and Mexico, all incubations were carried out overnight at 37°C. Following isolation, all strains were immediately stored at −80°C. All the strains classified by us as E. coli were confirmed as E. coli by serotyping in the Faculty of Medicine, Universidad Nacional Autónoma de Mexico.

In addition to the newly isolated strains, 13 strains from the E. coli reference collection ECOR (32), the sequenced K-12 strain MG1655 (2), and 6 strains from yellow baboons (Papio cynocephalus) were included in the analysis (35).

MLEE analysis. MLEE using cellulose acetate membranes was carried out in Tris-glycine buffer (pH 8.5) (17). Ten enzymes were selected based on previous studies (40): ADH (alcohol dehydrogenase), ARK (arginine kinase), G6PDH (glucose-6-phosphate dehydrogenase), IDH (isocitrate dehydrogenase), MDH (malate dehydrogenase), ME (malic enzyme), MPI (mannose-6-phosphate isomerase), PEP (peptidase), PGM (phosphoglucomutase), and XDH (xanthine dehydrogenase). Eleven loci were resolved with those enzymes because ME exhibited two loci. All of the strains were examined at least twice to confirm their electrophoretic types (Table 1).

Biotype analysis. All strains were taken from freezer cultures and grown on MacConkey plates. A single colony of each strain was tested for growth on minimal plates containing a 0.4% concentration (27) of one of the following sugars: adonitol, arabinose, dulcitol, inositol, maltose, mannitol, raffinose, rhamnose, salicin, sucrose, sorbitol, trehalose, and xylose.

Antibiotic resistance and toxin production. Resistance to each of six antibiotics was tested by using Luria broth plates supplemented with one of the following antibiotics: ampicillin (50 μg/ml), chloramphenicol (12 μg/ml), kanamycin (50 μg/ml), neomycin (50 μg/ml), streptomycin (50 μg/ml), and tetracycline (25 μg/ml). The concentrations were based on previous studies of E. coli (27, 35).

Strains were also tested for hemolysin production with heart infusion agar supplemented with blood (5%) and were tested for verotoxin (VT) production with Rainbow Agar O157 (Biolog, Inc.). However, Rainbow Agar O157 is not a direct test for VT production; rather, it detects a trait that has been found to be highly correlated with VT production in clinically pathogenic strains such as O157:H7 (44). This pathogenic strain is β-glucuronidase negative (black colonies); other VT-producing strains typically overproduce β-galactosidase relative to β-glucuronidase (blue, purple, or violet colonies), and it is the production of these two compounds that Rainbow Agar has the ability to detect. We took dark and bluish colonies to represent VT-positive strains. To determine the average number of colonies per host group, we scored for dark colonies as follows: 2 for black colonies; 1 for dark blue, violet, and purple; and 0 for red and white colonies.

Plasmid analysis. Isolated colonies were grown in Tris-borate medium and plasmids were extracted by the alkaline lysis procedure (12). Plasmid profiles were run in 0.7% agarose gel and stained with ethidium bromide. Megaplasmids were extracted directly in horizontal agarose gels by a modification of the procedure of Eckhardt (12). For purposes of statistical analysis, the observed plasmid bands were assigned to categories according to their size.

Statistical analysis. The isolates were grouped on the basis of the taxonomic order of the host from which they were isolated, host diet, and host geographic origin. Some of these groups are naturally confounded (most carnivora eat meat and most rodents are granivores), and it is not possible to separate their contribution to phenotypic or genetic diversity. For example, Australia lacks native primates and Mexico lacks monotremes. To partially compensate for these confounding effects, various subsets of the data were used in the analyses.

Average genetic diversity per locus was estimated as H = Σh_j/m, where m equals the number of loci scored and h_j = [n/(n − 1)] (1 − Σp_ij²), where p_ij is the frequency of allele i at locus j, and n is the number of multilocus genotypes (30, 40). Standard error of H was obtained with the ETDIV program (46). We used modified G_st statistics to analyze the data. For example, the proportion of genetic variation attributable to geographic effects is (H_T − H_G)/H_T, where H_G is the arithmetic average of the H’s calculated separately for electrotypes (ETs) from each locality, and H_T is the diversity of all strains regardless of locality (29, 30, 40). Standard error of G_st was also obtained with the ETDIV program (46). The statistical significance of G_st was analyzed with a χ² test of independence, using the formula χ² = nG_st(a − 1), where n is the number of individuals and a is the total number of alleles; degrees of freedom are (k − 1)(a − 1), where k is the number of subdivisions (16).

The frequencies of different biotype traits were compared using χ² tests or Fisher’s exact test when appropriate (37). To test for the association of phenotypic traits, a concordance analysis was done by using contingency coefficients (37). As contingency coefficients do not range from −1 to 1, as is the case for parametric measures of association, the absolute values of the coefficients are not presented.

A dendrogram was constructed using Nei’s genetic distances (30) and the neighbor-joining (NJ) method, using the Phylogeny Inference Package (PHYLIP, version 3.5c by Joseph Felsenstein, University of Washington).

Allozyme analysis. The 202 strains examined yielded 187 genotypes (i.e., ETs). The number of alleles per locus averaged 6.8 (Table 2; range, 4 to 9). Null alleles were detected at all loci except MDH. The locus with the greatest allelic diversity was ME2 (0.80) and G6PD was the least variable (0.441). The average allelic diversity (H ± standard error [SE]) for the 11 loci was 0.682 ± 0.034 for the 187 ETs, and for the 202 isolates, we obtained an H of 0.673 ± 0.034 (Table 2).

Locus	No. of alleles	H
ADH1	9	0.736
ARK	6	0.617
G6PD	5	0.441
IDH	7	0.778
MDH	6	0.730
ME1	4	0.531
ME2	9	0.800
MPI	8	0.786
PEP	8	0.697
PGM	7	0.727
XDH1	6	0.654
Avg ± SE	6.8	0.682±0.034

In Table 3, we present genetic diversity and genetic differentiation results at different levels of analysis. At the geographic level, the diversity (H) ranges from 0.489 in the human-related strains of the ECOR collection to 0.705 in the strains isolated from Mexican mammals. The genetic differentiation ± SE among those data sets (G_st = 0.047 ± 0.014; χ² = 630.9) is significantly different from zero (P < 0.0001).

Level of analysis	Origin of the strain or description of host	No. of strains	No. of alleles	H	Gst (SE)	P
Geographic	Australia	41	4.18	0.566
	Mexicoa	131	6.82	0.705
	ECOR	13	2.73	0.489	0.047(0.014)	0.00001
Host order	Carnivora	34	4.55	0.653
	Rodentia	51	4.82	0.657
	Marsupialia	28	4.27	0.603
	Primates	22	4.36	0.658
	Chiroptera	14	4.18	0.665
	Artiodactyla	11	3.09	0.511
	Perisodactyla	10	3.18	0.608
	Aves	10	3.36	0.630	0.075(0.017)	0.00001
Host diet	Omnivore	66	5.45	0.646
	Granivore	28	4.91	0.645
	Carnivore	12	4.27	0.671
	Herbivore	50	5.82	0.645
	Insectivore	23	4.36	0.672	0.025(0.007)	0.126
	Mexicob	110	6.74	0.698	0.044(0.012)	0.0052
	Australiab	41	4.18	0.566	0.01(0.01)	1
Rodents	Australia	17	2.91	0.515
	Mexico	34	4.73	0.639	0.098(0.03)	0.00001
Australia	Rodents	17	2.91	0.515
	Marsupials	21	3.55	0.552	0.036(0.011)	0.184

If we take into account the host order, the genetic diversity ranges from 0.511 in artiodactyls to 0.665 in bats; the genetic differentiation among the strains from different hosts (G_st = 0.075 ± 0.017; χ² = 898.56) is also significantly different from 0 (P < 0.0001; Table 3).

Considering host diet, the genetic diversity ranges from 0.672 in the strains associated with insectivores of different countries to 0.566 in the Australian strains with different diets. There is no genetic differentiation according to host diet (G_st = 0.025 ± 0.007; χ² = 323.64, P = 0.126). If we analyze only the host diet in Australia (G_st = 0.01 ± 0.01; χ² = 28.68, P = 0.999), we also do not find a significant difference. However, diet is significant in Mexico (G_st = 0.044 ± 0.012; χ² = 296, P = 0.005).

The genetic differentiation between rodents from Australia and Mexico (G_st = 0.098 ± 0.03; χ² = 234.85, P < 0.0001) was significantly different from zero (Table 3), with more diversity in the Mexican isolates (H = 0.639) than the isolates from Australian rodents (H = 0.515). However, partitioning the strains of Australia showed a nonsignificant differentiation (G_st = 0.036 = 0.011; χ² = 48.95, P = 0.184), because the genetic diversity is very similar in marsupials (H = 0.552) and rodents (H = 0.515).

The genetic relationship among 202 strains was analyzed with Nei genetic distances, and we constructed an NJ dendrogram (Fig. 1). We also performed other phylogenetic analyses using UPGMA (using both the number of mismatches and Nei distances as genetic distances) distance method and constructed a PAUP tree using parsimony (figures not shown). All the analyzed trees showed results analogous to those of the NJ dendrogram. Based on biochemical tests and the UPGMA dendrograms, we decided to root the NJ dendrogram with the reptile strain, which shows a long branch. If this root is correct, in the dendrogram we can define an ancestral group from which all the other strains are derived. In this supposedly ancestral group, we found several strains (10) from marsupials and rodents from Australia as well as 11 strains from a diverse group of mammals from Mexico and three strains from the cosmopolitan house mouse Mus musculus. In this clade, we also observed the strains from groups B1 and C of the ECOR collection, as well as one strain from the group A (ECOR8). The majority of the human-associated ECOR group A isolates are present in the most differentiated cluster (G), along with strain K-12 (MG1655, identical in our analysis to ECOR11) and a group of strains from Mexican carnivores, several from rodents, and all the nonpathogenic strains from humans. The rest of the ECOR isolates (no. 5 and 41) are in cluster C along with most of the bird strains and isolates from the old world and new world monkeys. In cluster B, we observed a diverse group of strains mostly from Lagomorpha and carnivores. Cluster D has a tight group of Australian strains from marsupials and rodents. Strains from Mexican rodents and bats are mostly in clusters E and F. Cluster F also contains strains from domestic carnivores and wild dolphins. It is evident in this tree that most of the strains from wild animals are grouped by host order. It is also interesting that most of the Australian strains are in two clusters (A and G) while Mexican strains are dispersed along the tree.

FIG. 1 Dendrogram depicting the strain similarities for the 202 strains of E. coli. The NJ tree was obtained from a Nei’s distance matrix derived from the multilocus electrophoresis data using the PHYLIP 3.5 program. For each strain, the name of the (more ...)

Biotype analyses. A minority of the strains could utilize adonitol, arabinose, dulcitol, inositol, or salicin, while most strains could ferment maltose, mannose, rhamnose, sorbose, trehalose, and xylose (Table 4). For 10 of the 12 substrates tested, the proportion of strains able to exploit a substrate varied significantly depending on the taxonomic group of the host from which they were isolated (Table 4). The ability of the strains to ferment maltose, rhamnose, and xylose was independent of host taxonomic group. Overall, strains isolated from hosts with diversified diets such as rodents, birds, and marsupials could ferment the greatest number of substrates, while strains from hosts with very specialized diets such as Monotremata, Cetacea, Xenarthra, and Sirenia could exploit fewer sugars.

Host groupa	No. of isolates	% Utilization of:													Total no. of sugars utilized
		Ado	Ara	Dul	Ino	Mal	Man	Raf	Ram	Sal	Sor	Suc	Tre	Xyl
Aves	10	20	10	50	40	80	80	40	90	50	100	80	100	90	13
Artiodactyla	12	0	0	8	0	100	25	50	75	0	58	33	91	75	9
Carnivora	34	6	0	39	0	100	76	67	91	0	91	70	100	100	10
Cetaceab	3	33	0	0	0	100	0	0	100	0	100	100	100	100	7
Chiroptera	15	29	0	79	29	93	100	79	93	50	86	100	100	93	12
Insectivorab	2	50	0	0	0	100	100	50	50	50	50	50	100	50	10
Lagomorphab	7	14	43	0	0	100	100	43	57	0	100	43	100	100	10
Marsupialia	28	4	4	11	11	100	86	39	86	4	89	71	100	100	13
Montrematab	5	0	0	0	0	100	0	0	80	0	100	60	100	100	6
Perissodactylab	9	33	0	0	0	100	56	33	67	0	89	22	100	78	9
Primates	24	5	5	14	0	100	54	50	68	0	55	64	91	91	11
Rodentia	51	10	4	10	2	90	64	50	78	10	84	70	90	90	13
Statistical significancec						NS			NS					NS
ECOR	13			32				39	81	0	39	11

^aDoes not include the single isolate from an elephant and a desert turtle.

^bAnalysis excluded those host groups were the strain number is lower than 10.

^cNS, not significant; , P < 0.05; , P < 0.01; , P < 0.001. Significance levels determined by χ² tests.

Geographic effects on the frequency of sugar utilization could also be detected. A comparison of strains isolated from rodents in Australia and Mexico revealed that for mannitol only 32% (n = 16) of the strains from Australia could ferment this sugar, compared to 85% (n = 30) of the strains from Mexico (P < 0.001). Similarly, 32% of strains from Australian rodents can ferment raffinose compared to 64% of the strains from Mexican rodents (P < 0.02).

Antibiotic resistance, toxin production, and plasmid profiles. Of the antibiotics tested, resistance to streptomycin was most frequent while resistance to chloramphenicol was rare (Table 5). Resistance to antibiotics was more frequent among strains from Mexico than among strains from Australia (9.6% of strains from Australia were resistant to one or more antibiotics compared to 41% of strains from Mexico). There was an insufficient number of strains with resistance to permit a detailed analysis of the distribution of antibiotic resistance as a function of host taxonomic group. However, considering the Mexican isolates alone, strains isolated from bats tended to have a much higher frequency of antibiotic resistance than other mammals. For example, 46% of strains from bats were ampicillin resistant, 100% were streptomycin resistant, and 15% presented neomycin resistance; none of those strains was resistant to the other antibiotics.

Antibiotic for which resistance was tested	% of strains resistant to indicated antibiotic			Level of statistical significancea
	All strains	Mexican strains (n = 135)	Australian strains (n = 46)
Ampicillin	13.4	18.5	5.8
Chloramphenicol	0.8	0.6	0	NS
Streptomycin	23.9	32.6	5.8
Kanomycin	2.1	3.3	0	NS
Neomycin	3.4	2.6	0	NS
Tetracycline	5.9	7.8	0

Overall, 44% of the strains produced more β-galactosidase than β-glucuronidase (i.e., were VT positive), and 2% of strains produced hemolysin. No difference in the overall frequency of hemolysins or VT production between strains from Australia and Mexico could be detected (Table 6; the average for Mexico was 0.64 ± 0.06 and the average for Australia was 0.57 ± 0.10 [not significant by Student’s t test]). The frequency of the marker for VT production varied with the host taxonomic group from which the strains were isolated (Table 6). Among the Mexican isolates, the frequency of positive strains ranged from 0.28 for strains isolated from Lagomorpha to 0.69 in strains isolated from rodents. Isolates from Australian rodents also showed a higher than average frequency of VT production compared to strains from marsupials. However, none of these differences are significant from zero.

Origin of host	Host order	No. of isolates	Avg no. of VT-positive strains ± SE	Frequency of VT-positive strains
Mexico and South America	All orders	121a	0.64±0.06	0.60
	Ungulata	18	0.39±0.14	0.33
	Carnivora	34	0.53±0.1	0.47
	Chiroptera	13	0.69±0.17	0.61
	Lagomorpha	7	0.29±0.17	0.28
	Primates	10	0.30±0.14	0.33
	Rodentia	34	1.09±0.14	0.69
Australia	All orders	46a	0.57±0.1	0.45
	Marsupialia	21	0.48±0.16	0.33
	Rodentia	17	0.82±0.15	0.66

Concordance analysis (40) of the phenotypic traits using the strains isolated from Mexico revealed that the presence or absence of these traits is not independent (Table 7). Three major forms of association were detected: (i) the concurrence of the ability to utilize the less frequently fermented sugars, such as dulcitol and inositol; (ii) the concurrence of the inability to exploit sugars utilized by most strains, such as xylose and trehalose; and (iii) the negative association of two traits, such as where the ability to exploit one sugar (e.g., salicin) results in a lower-than-expected frequency of strains able to exploit a second sugar (e.g., sucrose). Resistance to various antibiotics was also found to be concurrent in a strain more often than would be expected by chance. Resistance to ampicillin was associated with the ability to utilize arabitol, adonitol, dulcitol, and salicin. Also the ability to exploit salicin was found to be more frequently associated with streptomycin resistance than expected. Similar kinds of association were seen in the strains from Australia (results not shown).

We divided the samples by the distance of the host to the human environment (Table 8). We observed in the Mexican isolates and in the ECOR isolates a significantly higher amount of plasmids among isolates collected in cities and places close to humans than in those collected at isolated sites (χ² = 28.98, P < 0.001) (Table 8). The effect of possible human contact on the number of plasmids was also observed on a larger geographical scale. In Mexican strains, we observed an average of 1.7 plasmids per strain, while in the less-populated Australia, the strains presented a significantly lower average (0.45 plasmids per strain; χ² = 322.8, P < 0.001).

Country of origin of isolate	Proximity to human environmenta	Source or type of isolate	No. of isolates	Avg no. of plasmids (SE)	Avg no. of antibiotic-resistant isolates (SE)
Australia	More than 100 miles from city	Wild mammals	46	0.45(0.09)	0.19(0.01)
Mexico	More than 100 miles from city	Wild mammals	86	1.20(0.12)	0.58(0.06)
Mexico	Less than 100 miles from city	Wild mammals	14	1.55(0.42)	0.42(0.09)
Mexico	Zoo	Wild mammals	9	1.86(0.45)	0.33(0.17)
Mexico	Cities	Domestic mammals	23	2.09(0.42)	0.47(0.08)
Mexico and United States	Cities	ECOR and nonpathogenic	14	2.62(0.65)	0.66(0.09)

On the other hand, E. coli associated with wild animals in isolated places in Mexico is, on average, resistant to as many antibiotics (average ± SE = 0.58 ± 0.06; Table 8) as E. coli in the other human-related sources from Mexico (weighted average ± SE of all the other sources = 0.43 ± 0.01). Our sample of the ECOR collection had the highest antibiotic resistance (average ± SE = 0.68 ± 0.02; data not shown in table), while Australian isolates had the lowest antibiotic resistance (average ± SE = 0.19 ± 0.01). In our sample, we did not observe that the antibiotic resistance is correlated with the mean number of plasmids. Most of the multidrug-resistant strains from wild mammals did not have any plasmids, suggesting that the genes for antibiotic resistance are chromosomal in some of the wild animal strains.

The estimated genetic diversity (H = 0.682) from our collection of E. coli is higher than the diversity reported for any collection of E. coli (5, 26, 33, 34, 38, 42, 46–48) and higher than most studied bacteria (7, 9, 10, 14, 28, 29, 39, 43, 45). This high diversity could be due in part to the fact that we intentionally selected a wide range of “good and informative characters” based on the analysis done by Selander and collaborators (40, 42). On the other hand, the observed genetic diversity would change if other loci were selected or if MLEE were performed with starch and a discontinuous buffer with a different pH (17). For example, based on 11 loci, the H was originally determined to be 0.42 for the ECOR collection (32). Subsequently, the number of loci examined was increased to 35, resulting in an H of 0.34 (41). However, the high genetic diversity we observed in our study could also be the result of the high diversity of hosts represented in our collection; the sample we studied includes isolates from all of the major mammalian orders (98.3% of the species are within those orders [31]). Even though isolates from some minor orders are lacking, when we add 10 isolates from 10 different birds the genetic diversity increased only 0.2%. The geographic scope of the sample is also patchy, with only a few strains from Africa or Asia and none from either polar region. The results suggest, however, that filling these gaps may not significantly increase estimates of the genetic diversity of E. coli. For example, in the Mexican mammalian samples alone, the genetic diversity (H) was estimated as 0.698 ± 0.048 (data not shown). But inclusion of the Australian samples, which represent an isolated continent with a unique mammalian fauna, lowered the estimate of the genetic diversity by 1%.

Previous studies of E. coli suggest that much of the observed allelic variation is selectively neutral and this, coupled with a large effective population size, can explain much of this species’ genetic diversity (42, 50). Population structure has been thought to account for little of the observed diversity (5, 11, 20, 42). However, genetic diversity in our collection of E. coli is probably ecologically (ecotypically) structured and as such, adaptation to the host plays an important role in population structure. We observed that host order and geographic origin are the most important factors to differentiate E. coli. Overall, 7.5% of the diversity can be explained by the order of the host, while nearly 10% of the genetic diversity of strains from rodents (Australia versus America) can be explained by geography alone. This contrasts with the observation that spatial structure accounts for only 2% of the genetic diversity among strains isolated from humans living in North America and Europe (27). That spatial structure accounts for so little of the genetic diversity of E. coli from humans probably reflects the fact that the samples are derived from a single, highly mobile host species. The larger geographic component of diversity observed in the strains from rodents can be attributed both to Australia’s long isolation and to the distinct rodent communities of the two continents. In Mexico, the diet of the host accounts for 4.4% of the diversity of E. coli. However, in Australia or in the total collection, the diet of the host was not an important factor differentiating populations.

The observed differentiation by host could be due in part to the presence of different ecological niches in the different orders. We observed that the type and range of sugars that can be exploited by E. coli strains are associated with the taxonomic group of the host from which the strains were isolated. Significant heterogeneity in sugar utilization also resulted when strains were grouped on the basis of host diet. Nevertheless, these results are somewhat ambiguous. This may reflect the joint problem of sample size for the diet grouping (for example, four dietary modes among the 14 bat hosts) and the inevitable interactions between host taxonomy and diet. Overall, E. coli from the less-represented hosts with specialized diets (monotremes, dolphins, Xenarthra, and Sirenia) exploited the fewest sugars. The strains from hoofed mammals also used few sugars, while bacteria from birds, marsupials, and rodents could utilize a larger number of sugars. The frequency with which strains can exploit some sugars was also found to vary with the geographic origin of the strains. For example, a greater proportion of strains from Mexican rodents can utilize raffinose and mannitol than strains from Australian rodents.

In previous studies, the majority of strains (69%) in the ECOR collection that were isolated from strictly herbivorous mammals could exploit raffinose, whereas in this study, less than half the strains recovered from the herbivorous groups could utilize raffinose (27, 32, 42). Only 30% of ECOR strains can ferment sorbose, compared to the 82% frequency of utilization observed in this study (27, 32, 42). No ECOR strains could use salicin, while in this study, 9% of the strains did so, in contrast with 41% in a population of strains from yellow baboons (35). Host digestive physiology and anatomy clearly vary among mammal species and reflect the interaction of phylogeny and diet. Host diet will in turn influence the kinds of growth substrates available to E. coli. As a result, E. coli strains will be confronted with very different biotic and abiotic environments in different host species. Our study suggests that these different environments result in E. coli that exhibit some degree of host specificity. Further work is required to determine the mechanisms responsible for this specificity and the role that such specificity plays in E. coli’s evolution.

The dendrogram depicted in Fig. 1 provides additional support for the observation that host and geographic effects contribute to the genetic structure of E. coli populations. For example, strains isolated from birds, rodents, and carnivores seem to cluster more often than other hosts, as do Australian strains. Strains from ungulates are in many clusters. In our tree, the ancestral cluster corresponds to some of the Australian strains along with the ECOR strains from groups B1 and C, while the most differentiated cluster presents strains from the ECOR group A along with carnivores from Mexico and other humans. Using the sequence data of 13 gene phylogenies, Lecointre et al. (23) observed that the ECOR strains had a different organization than previously reported. In their study the most pathogenic strains from group B2 were the ancestral group, while A and B1 were the most evolved sister groups. In their scenario, the capability of E. coli to invade other niches (extraintestinal virulence) is an ancestral trait that has evolved to the more benign intestinal strains in the other groups (23). We do not have strains from group B2 in our study, but preliminary data from our laboratory (data not shown) suggest a similar scenario for the evolution of diarrheic E. coli. We observed that four of the Mexican strains from our ancestral group present genes from the pathogenic island LEE (eae and espB) associated with an intestinal lesion in the EPEC and EHEC strains. We also observed in this ancestral clade six strains with ETEC serotypes. Preliminary analysis (data not shown) indicates that pathogenic chromosomal genes are common and diverse along the tree, suggesting an ancestral origin.

We also observed both that the VT marker is widespread in the tree and that VTs are more frequent in strains from rodents from both continents. The latter observation is consistent with the idea that rodents act as a reservoir of many diseases (1). On the other hand, artiodactyls have strains with the lower percentage of β-galactosidase production, contrary to the belief that domestic cattle naturally harbor β-galactosidase-producing E. coli (i.e., O157:H7) (36, 44). However, none of the levels of VT production per host are significantly different from the average for their country of origin. Another characteristic that is widespread in the collection is antibiotic resistance. With the exception of strains from Australian mammals, most of the E. coli strains present some antibiotic resistance, and multidrug resistance is frequent in Mexico. This could be in part the result of the widespread use of antibiotics and to the lack of real isolation from human environment in a country like Mexico. However, since antibiotic resistance is not randomly distributed in the wild E. coli strains (bats and rodents host strains with higher multidrug resistance), we believe that at least some of the antibiotic resistance may be related to additional nonhuman environmental pressures.

On the other hand, the number and size of plasmids are highly variable in E. coli, ranging from hundreds of base pairs to several hundred kilobase pairs. Their G+C contents can vary widely and are often different from that of their usual bacterial host, indicating a variety of sources from which plasmids (or part of them) are derived (3, 18, 21, 22, 24). No pattern has been observed in the number of plasmids in these bacteria, and this may be due in part to their possible heterogeneous origin (3, 21, 22). Nevertheless, we observed that with human proximity, the number of plasmids per strain increases. This could be a reflection of the acquisition of accessory elements due to increased densities of both hosts and bacteria in cities. Increased density could facilitate the movement of plasmids among strains of E. coli as well as other related bacteria. Such an exchange was observed by Boyd et al. (3), who analyzed the structure of F group-related plasmids in the ECOR collection. They found different phylogenetic relationships between plasmids and bacterial strains, suggesting that horizontal transfer of plasmids occurs at high rates within the ECOR collection.

We thank the numerous colleagues who assisted us in obtaining samples. David Gordon collected all the Australian samples, helped with some statistical analysis, and reviewed many previous versions of the manuscript. Blanca Hernández, Jordan Goluvov, Meli Mandrujano, Rodrigo Medellin, Saul Aguilar, Osiris Gaona, Angeles Mendoza, Gabriel Pérez, Luis Medrano, Adolfo Navarro, Juan Castillo, Alejandro Zavala, Benjamin Morales, Ricardo Frias, Alejandro Velázques, Hector Arita, Carlos Alvarez, José Charles, Hilary Charles, Alvaro Miranda, David Valenzuela, Luisa Sandner, René Cerritos, and Jorge Ortega collected the Mexican samples. We also thank Antonio Cruz for valuable technical assistance and Armando Navarro, who performed the serotype analysis in the Faculty of Medicine, Universidad Nacional Autónoma de México. We thank Pilar Beltrán, Robert K. Selander, Brandon Gaut, Andrew Peek, and Juan Nuñez, who read the manuscript and gave useful comments. Richard Goldstein kindly provided the MG1655 strain of E. coli K-12, and Dan Dykhuizen sent us the yellow baboon strains.

This research was supported by DGAPA-UNAM IN208995 grant to V.S. and CONACyT 3675-N grant to L.E.E. and V.S., and M.R. was supported by a graduate student scholarship from CONACyT.