Molecular Epidemiology and Phylogenetic Distribution of the Escherichia coli pks Genomic Island

Abstract

Epidemiological and phylogenetic associations of the pks genomic island of extraintestinal pathogenic Escherichia coli (ExPEC), which encodes the genotoxin colibactin, are incompletely defined. clbB and clbN (as markers for the 5′ and 3′ regions of the pks island, respectively), clbA and clbQ (as supplemental pks island markers), and 12 other putative ExPEC virulence genes were newly sought by PCR among 131 published E. coli isolates from hospitalized veterans (62 blood isolates and 69 fecal isolates). Blood and fecal isolates and clbB-positive and -negative isolates were compared for 66 newly and previously assessed traits. Among the 14 newly sought traits, clbB and clbN (colibactin polyketide synthesis system), hra (heat-resistant agglutinin), and vat (vacuolating toxin) were significantly associated with bacteremia. clbB and clbN identified a subset within phylogenetic group B2 with extremely high virulence scores and a high proportion of blood isolates. However, by multivariable analysis, other traits were more predictive of blood source than clbB and clbN were; indeed, among the newly sought traits, only pic significantly predicted bacteremia (negative association). By correspondence analysis, clbB and clbN were closely associated with group B2 and multiple B2-associated traits; by principal coordinate analysis, clbB and clbN partitioned the data set better than did blood versus fecal source. Thus, the pks island was significantly associated with bacteremia, multiple ExPEC-associated virulence genes, and group B2, and within group B2, it identified an especially high-virulence subset. This extends previous work regarding the pks island and supports investigation of the colibactin system as a potential therapeutic target.

Escherichia coli, an important extraintestinal pathogen in humans and certain animals, is a common cause of urinary tract infection, sepsis, neonatal meningitis, and colibacillosis (12, 21). The specialized E. coli strains that cause most extraintestinal infections, often termed extraintestinal pathogenic E. coli (ExPEC), exhibit multiple accessory traits, such as adhesins, toxins, polysaccharide coatings, invasins, and siderophores, that are absent or uncommon among nonpathogenic commensal strains, such as E. coli K-12 (8, 12). These traits, commonly referred to as virulence factors (VFs), contribute to the distinctive ability of ExPEC to resist host defense mechanisms and injure or invade the host. A better understanding of ExPEC VFs could guide the development of preventive measures, such as vaccines against extraintestinal E. coli infections, and could be used in surveillance systems for identifying reservoirs and transmission of ExPEC strains (4, 9, 20).

We recently identified in Escherichia coli a hybrid nonribosomal peptide-polyketide compound called colibactin that induces DNA double-strand breaks in eukaryotic cells (16). Such DNA damage activates host cellular signaling pathways that lead to cell cycle arrest, the progressive enlargement of cell bodies, and eventually cell death. We identified a 54-kb genomic island that is responsible for this bacterial trait. This genomic island, which is hereafter referred to as the pks island, encodes the machinery for the synthesis of colibactin. This machinery consists of nonribosomal peptide megasynthases (NRPS), polyketide megasynthases (PKS), hybrid NRPS/PKS enzymes, different tailoring or editing enzymes, and a critical phosphopantetheinyl transferase that mediates NRPS and PKS activation (16). The pks island was first characterized in sequenced ExPEC prototype strains and up to now was found exclusively in phylogenetic group B2 strains (16). Many strains belonging to this phylogenetic group cause extraintestinal infections, such as urinary tract infections and septicemia. Colibactin can thus be regarded as a VF of ExPEC, but little is known about its prevalence among isolates from defined clinical syndromes versus commensal strains or its associations with other VFs (13, 16, 25). In addition, certain commensal strains from healthy people also possess the pks island (13). Of note, one such commensal, the probiotic strain Nissle 1917, contains the pks island and produces a functional genotoxin (16).

Accordingly, in the present study, we sought to better define the clinical correlates of this polyketide synthesis system and its associations with other ExPEC VFs and phylogenetic group. For this, we took advantage of a previously published collection of 62 bloodstream isolates from U.S. veterans with diverse-source E. coli bacteremia and 69 concurrent fecal isolates from uninfected veterans (22), for which the major E. coli phylogenetic group (A, B1, B2, and D) and 48 virulence genes had already been analyzed (22). We newly characterized these isolates for clbB and clbN (as markers for the 5′ and 3′ regions of the pks island, respectively), clbA and clbQ (as supplemental pks island markers), and 12 other putative ExPEC virulence genes. We then compared clbB and clbN with the 60 other virulence genes and phylogenetic group as predictors of blood versus fecal source, examined the phylogenetic distribution of clbB and clbN, and defined the associations of clbB and clbN with other virulence genes.

MATERIALS AND METHODS

Patients and strains.

The bacteremia and fecal isolates studied have been described previously (22). Briefly, all blood isolates of E. coli recovered by the Veterans Administration (VA) Medical Center clinical microbiology laboratory between 1 December 1996 and 30 September 1999 were obtained from the laboratory. For fecal controls, during the same time interval, newly admitted VA Medical Center inpatients who lacked evidence of acute infection were recruited as donors of rectal swabs. Patients identified to study personnel by their primary physicians or nurses as potential study subjects were approached regarding study participation. A rectal swab was collected from each consenting subject and processed as described elsewhere (3) to recover an isolated colony of presumptive E. coli, i.e., lactose- and indole-positive, citrate-negative, gram-negative bacilli with characteristic colonial morphology. Previous data show that an arbitrarily selected fecal E. coli colony has an 86% probability of representing the quantitatively predominant clone in the sample (14). Blood and fecal isolates were stored at −80 C in 15% glycerol pending molecular analysis.

Virulence genotyping.

The 12 newly detected markers besides clbB and clbN (colibactin polyketide synthesis system) included afaE8 (afimbrial adhesin variant), astA (enteroaggregative E. coli heat-stable toxin), clpG (K-88-related CS31A adhesin), the F17 gene (F17c fimbriae), hlyF (hemolysin variant), hra (heat-resistant agglutinin), kfiC (K5 capsule), kpsM K15 (K15 capsule), pic (protein involved in intestinal colonization; serine protease), tsh (temperature-sensitive hemagglutinin; serine protease), usp (uropathogen-specific protein), and vat (vacuolating autotransporter; serine protease). Primers for the F17, sat, usp, afaE8, clpG, astA, kfiC, and pic genes were as published previously (7, 17, 24). Primers for clbB and clbN were clbBr (r for reverse orientation) (5′-CCA TTT CCC GTT TGA GCA CAC-3′), clbBf (f for forward orientation) (5′-GAT TTG GAT ACT GGC GAT AAC CG-3′), clbNr (5′-CAG TTC GGG TAT GTG TGG AAG G-3′), and clbNf (5′-GTT TTG CTC GCC AGA TAG TCA TTC-3′). Primers for two internal loci within the pks island, i.e., clbA (phosphopantetheinyl transferase) and clbQ (thioesterase), were IHAPJPN42 (clbAr) (5′-CAG ATA CAC AGA TAC CAT TCA-3′), IHAPJPN46 (clbAf) (5′-CTA GAT TAT CCG TGG CGA TTC3′), IHAPJPN55 (clbQr) (5′-TTA TCC TGT TAG CTT TCG TTC-3′), and IHAPJPN56 (clbQf) (5′-CTT GTA TAG TTA CAC AAC TAT TTC-3′). Primers for the K15 kpsM variant and tsh, as designed based on published sequence data (15, 23), were k15f (5′-ACG GAT TCA CGA CAA AGC TC-3′), k15r (5′-GGC AAA TAT CGC TTG GGT TA-3′), tshf (5′-GGC GCA TAT CTG GAT AAG GA-3′), and tshr (5′-GCA GGT TAT CGC CCT TAA CA-3′). Testing was done in duplicate using separately prepared boiled lysates as template DNA, with appropriate positive and negative controls, and published amplification conditions (7, 10, 11).

Isolates were operationally regarded as ExPEC if they were positive for two or more of the following genes: papA and/or papC, sfa and/or foc (S and F1C fimbriae), afa and/or dra (Dr-binding adhesins), iutA (aerobactin system), and kpsM II (group 2 capsule). The virulence score was the number of virulence genes detected, adjusted for multiple detection of the pap, sfa or foc, and kpsM II operons.

Phylotyping.

Major E. coli phylogenetic groups (A, B1, B2, and D) were determined by multiplex PCR. For several isolates, seeming discrepancies between newly determined virulence genotype and previously determined phylogenetic group (22) prompted repeat phylotyping, leading to their reclassification into a different phylogenetic group.

Statistical methods.

Comparisons of proportions were tested using Fisher's exact test (two-tailed). Virulence scores were compared by using the Mann-Whitney U test. Odds ratios and 95% confidence intervals were calculated by using univariable logistic regression and stepwise multivariable logistic regression.

Similarity relationships among the individual isolates with respect to VF profiles and phylogenetic group were assessed by using principal coordinate analysis (PCoA), which is a multivariate technique related to correspondence analysis (2) that allows one to plot the major patterns within a multivariate data set, e.g., multiple loci and multiple samples (18). Using GenAlEx6 (18), PCoA was applied to the VF data set in such a way so as to collapse the multiple VFs for simplified among-group comparisons. Values for each isolate from the first three PCoA axes, which are the axes that capture most of the variance within the data set (18), were used in a one-way multivariate analysis of variance (MANOVA) to test for differences between resistant and susceptible isolates.

RESULTS

Virulence genotypes.

The colibactin system markers clbB and clbN were detected in 58 (44%) of the 131 total isolates, always jointly. Additionally, in a subset of the isolates (40 isolates; 10 each clbB and clbN positive and 10 clbB and clbN negative for blood and fecal isolates alike), PCR-based detection of two additional pks island markers (clbA and clbQ) yielded results precisely concordant with those for clbB and clbN, implying the presence of a complete pks island in all clbB- or clbN-positive strains. Accordingly, clbB alone was used in subsequent statistical analyses.

clbB was significantly associated with bacteremia, occurring in over half of the bacteremia isolates but in fewer than one-third of fecal isolates (odds ratio, 2.96; P = 0.003), although it accounted for only 9% of the total bacteremia-versus-fecal isolate variance (Table 1). Many other bacterial traits were also associated with bacteremia; they exhibited odds ratios for bacteremia ranging from 2.33 (for iroN [siderophore receptor]) to 16.69 (ompT [outer membrane protease]) (median, 3.96) and individually accounted for 5% to 43% (median, 9%) of the bacteremia-versus-fecal variance (Table 1). Aside from clbB, among the significantly bacteremia-associated markers were two other of the newly studied genes, i.e., hra (heat-resistant agglutinin) and vat (vacuolating toxin), which exhibited odds ratios and r² values similar to those of clbB (Table 1). Thus, clbB was more closely associated with bacteremia than were most of the other virulence genes studied, but not as closely as some, exhibiting a mid-range odds ratio and r² value compared with other significantly bacteremia-associated genes. Consistent with the greater prevalence of many individual virulence genes (including clbB and clbN) in bacteremia isolates, they exhibited significantly higher aggregate virulence scores than did fecal isolates (median score of 13.5 [range, 0 to 21] versus 7.0 [range 0 to 18]; P < 0.001).

TABLE 1.

Bacterial traits significantly associated with blood source among 131 fecal or blood isolates of Escherichia coli from veterans^a

Category and traitb	Prevalence of trait (no. [%]) in isolates:			P value (fecal vs blood isolates)	Odds ratio (fecal vs blood isolates)	95% Confidence interval	Negelkerke r2d
	Total (n = 131)	Fecal (n = 69)	Blood (n = 62)
Adhesins
papA, papC, papEF, or papG	64 (49)	21 (30)	43 (69)	<0.001	5.17	2.46-10.89	0.19
F10 papA allele	26 (20)	6 (9)	20 (32)	0.001	5.0	1.85-13.49	0.12
papG allele II	42 (32)	12 (17)	30 (48)	<0.001	4.45	2.01-9.88	0.14
sfa or focDE	35 (27)	11 (16)	24 (39)	0.005	3.33	1.46-7.48	0.09
focG	17 (13)	4 (6)	13 (21)	0.017	4.31	1.32-14.04	0.07
iha	35 (27)	10 (14)	25 (40)	0.001	3.99	1.72-9.24	0.11
hra	42 (32)	14 (20)	28 (45)	0.005	3.12	1.44-6.75	0.09
Toxins
hlyD	39 (30)	11 (16)	28 (45)	<0.001	4.34	1.92-9.82	0.13
cnf1	32 (24)	9 (13)	23 (37)	0.002	3.93	1.65-9.38	0.10
sat	37 (28)	13 (19)	24 (39)	0.019	2.67	1.21-5.90	0.06
clbB and clbN	58 (44)	22 (32)	36 (58)	0.003	2.96	1.45-6.05	0.09
vat	78 (60)	34 (49)	44 (71)	0.013	2.52	1.22-5.19	0.06
Siderophores
iroN	46 (35)	18 (26)	28 (45)	0.028	2.33	1.12-4.86	0.05
fyuA	91 (69)	40 (56)	51 (82)	0.004	3.36	1.50-7.54	0.09
ireA	25 (25)	5 (7)	20 (32)	<0.001	6.10	2.12-17.50	0.13
iutA	48 (37)	14 (20)	34 (55)	<0.001	4.77	2.21-10.31	0.16
Capsule kpsM II	84 (64)	34 (49)	50 (81)	<0.001	4.29	1.95-9.42	0.14
Miscellaneous
traT	70 (53)	30 (43)	40 (65)	0.02	2.36	1.17-4.78	0.06
ompT	66 (50)	15 (22)	51 (82)	<0.001	16.69	7.71-39.72	0.43
malX	80 (61)	36 (52)	44 (71)	0.02	2.24	1.09-4.62	0.05
ExPEC (combination of above)c	82 (63)	31 (45)	51 (82)	<0.001	5.68	2.54-12.72	0.19
Phylogenetic group B1	15 (11)	13 (19)	2 (3)	0.005	0.14	0.03-0.67	0.09

^aA total of 36 genes or traits were detected in at least one isolate each but not yielding P < 0.05 (in comparisons of the percentage of positive isolates; newly detected traits shown in bold type): papA alleles F7-1 (5%), F7-2 (1.5%), F8 (<1%), F9 (<1%), F11 (14%), F12 (2%), F13 (5%), F14 (7%), F16 (3%); papG allele III, P-adhesin variant (15%); sfaS, S fimbriae (10%); afa and/or draBC, Dr-binding adhesins (5%); afaE8, afimbrial adhesin variant (1.5%); bmaE, M fimbriae (1.5%); F17 fimbriae, mannose-resistant adhesin (1.5%); clpG, mannose-resistant adhesin (<1%); tsh, temperature-sensitive hemagglutinin (4%); fimH, type 1 fimbriae (97%); hlyF, variant hemolysin (7%); cdtB, cytolethal distending toxin (5%); astA, enteroaggregative E. coli toxin (3%); pic, autotransporter protease (9%); K1, group 2 capsule variant (26%); K2, group 2 capsule variant (6%); kfiC, K5 group 2 capsule variant (11%); kpsM III, group 3 capsule (1.5%); rfc, O4 lipopolysaccharide (4%); cvaC, microcin V (6%); usp, uropathogen-specific protein (62%); ibeA, invasion of brain endothelium (14%); iss, increased serum survival (8%); H7 fliC, flagellin variant (18%); and phylogenetic group A (11%), phylogenetic group B2 (61%), and phylogenetic group D (16%). Five traits were sought but not detected (newly sought traits shown in bold type): F15 and F536 papA allele, papG allele I, gafD (G fimbriae), and K15 (group 2 capsule variant).

^bThe 25 individual traits shown are those that yielded P < 0.05 when the P values for fecal versus blood isolates were compared. Newly detected traits are shown in bold type. Definitions: papA, papC, papEF, and papG, P-fimbria structural subunit, assembly, tip pilins, and adhesin, respectively (all gave highly similar results; values shown are for papA); sfa and/or focDE, S and/or F1C fimbriae; focG, F1C fimbriae; iha, adhesin siderophore; hra, heat-resistant agglutinin; hlyD, alpha-hemolysin; cnf1, cytotoxic necrotizing factor 1; sat, secreted autotransporter toxin; clbB, colibactin (clbN gave identical results); vat, vacuolating toxin; iroN, salmochelin receptor; fyuA, yersiniabactin receptor; ireA, siderophore receptor; iutA, aerobactin receptor; kpsM II, group 2 capsule; traT, serum resistance associated; ompT, outer membrane protease; malX, pathogenicity island marker.

^cExPEC, extraintestinal pathogenic E. coli, which was operationally defined as the presence of at least two of the following genes: papA and/or papC, sfa and/or focDE, afa and/or draBC, iutA, and kpsM II.

^dNegelkerke r², fraction of variance (blood versus fecal) accounted for by the indicated trait.

Multivariable analysis.

Because of known associations among virulence genes and phylogenetic background, multivariable logistic regression analysis was used to identify independent predictors of bacteremia. All the virulence markers studied, plus the phylogenetic groups (A, B1, B2, and D), were used as candidate predictor variables in a forward stepwise model. The resulting model, which accounted for 65% of total bacteremia-associated variance, identified seven significant predictor variables. Of these, ompT, iutA, papA, and sfa and/or foc were positive predictors of bacteremia, whereas pic, group B2, and the F16 papA allele were negative predictors, after adjustment for the other variables in the model (Table 2). Thus, in this model, clbB was not a significant independent predictor, whereas the only newly identified predictor was pic, which exhibited a negative (adjusted) association with bacteremia.

TABLE 2.

Stepwise multivariable logistic regression analysis for predicting blood source among 131 fecal or blood isolates of Escherichia coli from veterans

Variablea	Step of entry in model	P value	Odds ratio	95% Confidence interval
ompT	1	<0.001	59.9	11.34-315.8
pic	2	0.004	0.04	0.005-0.36
iutA	3	0.008	6.9	1.7-28.6
Group B2	4	0.001	0.03	0.004-0.2
F16 papA allele	5	0.007	0.007	0.000-0.2
papA	6	0.017	5.1	1.3-19.6
sfa and/or foc	7	0.039	5.4	1.9-26.9

^aVariables shown are those that yielded P < 0.05 when entered into the stepwise model along with previously entered traits. Definitions are as in Table 1, footnotes a and b. The newly detected trait (pic) is shown in bold type.

ExPEC isolates.

A greater proportion of bacteremia isolates than fecal isolates satisfied molecular criteria for ExPEC (82% versus 45%; P < 0.001). Among the 82 ExPEC isolates, five virulence genes differed significantly in prevalence according to clinical source. Of these, three were significantly more prevalent (papEF [P-fimbria tip pilins], ireA [siderophore receptor], and ompT) and two significantly less prevalent (pic and the H7 fliC [flagellin] variant) among bacteremia isolates than fecal isolates (Table 3), whereas clbB was similarly prevalent in each group (71%). Accordingly, ExPEC blood isolates exhibited significantly higher aggregate virulence scores than did ExPEC fecal isolates (median score of 14 [range, 5 to 21] versus 12 [range, 2.25 to 18]; P = 0.01).

TABLE 3.

Bacterial traits significantly associated with blood source among 82 fecal or blood isolates of extraintestinal pathogenic Escherichia coli from veterans

Traita	Prevalence of trait (no. [%]) in isolates:			P value (fecal vs blood isolates)
	Total (n = 82)	Fecal (n = 31)	Blood (n = 51)
papEF	58 (71)	17 (55)	41 (80)	0.02
pic	10 (12)	7 (23)	3 (6)	0.037
ireA	24 (29)	5 (16)	19 (37)	0.048
ompT	58 (68)	9 (29)	47 (92)	<0.001
H7 fliC allele	24 (29)	14 (45)	10 (20)	0.02

^aNote that traits shown are those that yielded P < 0.05. Definitions are as in Table 1, footnotes a and b. The newly detected trait (pic) is shown in bold type.

Associations of clbB with other traits in relation to bacteremia.

To identify possible explanations for the univariable but not multivariable association of clbB with bacteremia, associations of clbB with other virulence genes and phylogenetic group were examined. clbB was significantly associated with many of the virulence genes studied, including, among the newly sought markers, hra, pic, vat, and kfiC (Table 4). clbB was also confined to ExPEC isolates and to phylogenetic group B2, occurring in approximately 75% of either subset. Accordingly, aggregate virulence scores among clbB-positive isolates were significantly higher than the scores among clbB-negative isolates. This was true both in the total population overall and within relevant subgroups, including fecal isolates and blood isolates) and in group B2 (overall, fecal isolates, and blood isolates) (Table 5). Likewise, according to multiple linear regression, group B2 and clbB were each significantly associated with increased virulence scores when present together in the model (P < 0.001 for each). Additionally, within group B2, clbB-positive isolates occurred nearly twice as often among blood isolates (36/58 [62%]) as among clbB-negative isolates (8/24 [33%]) (P = 0.006). These findings suggested that clbB identified a particularly VF-rich and bacteremia-capable subset within group B2, with multiple traits other than (or in addition to) colibactin possibly contributing to the enhanced virulence.

TABLE 4.

Association of clbB or clbN with other bacterial traits among 131 blood and fecal isolates of Escherichia coli from veterans

Category and traita	Prevalence of trait according to clbB or clbN status (no. [%])		P value
	clbB and clbN negative (n = 73)	clbB and clbN positive (n = 58)
Adhesins
papA, papC, papEF, or papG	16 (22)	48 (83)	<0.001
F7-2 papA allele	0 (0)	7 (12)	0.003
F10 papA allele	3 (4)	23 (40)	<0.001
F11 papA allele	4 (5)	12 (24)	0.004
F14 papA allele	1 (1)	8 (14)	0.01
papG allele II	12 (16)	30 (52)	<0.001
papG allele III	2 (3)	18 (31)	<0.001
sfa and/or focDE	4 (5)	31 (53)	<0.001
sfaS	2 (2)	11 (19)	0.003
focG	3 (4)	14 (24)	0.001
iha	13 (18)	22 (38)	0.016
hra	13 (18)	29 (50)	<0.001
Toxins
hlyD	7 (10)	32 (55)	<0.001
cnf1	3 (4)	29 (50)	<0.001
cdtB	0 (0)	7 (12)	0.003
sat	13 (18)	24 (41)	0.004
pic	3 (4)	9 (16)	0.03
vat	20 (27)	58 (100)	<0.001
Siderophores
iroN	13 (18)	33 (57)	<0.001
fyuA	33 (45)	58 (100)	<0.001
ireA	3 (4)	22 (38)	<0.001
iutA	19 (26)	29 (50)	0.006
Capsule
kpsM II	29 (40)	55 (95)	<0.001
K1 kpsM variant	11 (15)	23 (40)	0.002
kfiC(K5 capsule)	4 (5)	10 (17)	0.045
Miscellaneous
usp	23 (32)	58 (100)	<0.001
traT	33 (45)	37 (64)	0.037
ompT	23 (32)	43 (74)	<0.001
H7 fliC variant	3 (4)	21 (36)	<0.001
malX	23 (32)	57 (98)	<0.001
ExPEC (combination of above)	24 (33)	58 (100)	<0.001

^aNote that the traits shown are those that yielded P < 0.05. Definitions are as in Table 1, footnotes a and b. The newly detected traits are shown in bold type.

TABLE 5.

Virulence scores according to clbB or clbN status among blood and fecal isolates of Escherichia coli from veterans

Population (total no. of isolates)	clbB- and clbN-positive isolates		clbB- and clbN-negative isolates		P valuea
	No. of isolates in subgroup	Virulence score (median [range])	No. of isolates in subgroup	Virulence score (median [range])
Total population
Overall (131)	58	14.9 (8-21)	73	4 (0-15)	<0.001
Fecal (69)	22	13.9 (8-18)	47	2 (0-13.8)	<0.001
Blood (62)	36	15 (11-21)	26	8.5 (0-15)	<0.001
Group B2
Overall (80)	58	14.9 (8-21)	22	10 (1-15)	<0.001
Fecal (38)	22	13.9 (8-18)	16	8.3 (1-13.8)	<0.001
Blood (42)	36	15 (11-21)	6	14 (12-15)	0.003

^aP values for comparison of virulence scores were determined by the Mann-Whitney U test.

PCoA.

For simplified two-group comparisons and to avoid possible multiple-comparison artifacts resulting from collinearity among virulence genes and phylogenetic group, PCoA was used to assess the distribution of bacteremia and fecal isolates and, separately, the clbB-positive and clbB-negative isolates, according to all the bacterial traits studied. In a PCoA that included all virulence markers studied and the four phylogenetic groups, the initial MANOVA indicated that the first three PCoA axes, when considered jointly, significantly differentiated the bacteremia and fecal isolates (P < 0.001). When the three axes were assessed individually, axis 1 was found to account for only 16% (P < 0.001) and axis 3 for 5% (P = 0.005) of the total bacteremia versus fecal variance, whereas axis 2 was noncontributory. In contrast, in a PCoA that included the same variables except clbB, the initial MANOVA again indicated an overall differentiation of bacteremia from fecal isolates by the first three PCoA axes (P < 0.001). Individual assessment of each axis showed that axis 1 accounted for fully 59% (P < 0.001) of clbB-specific variance, whereas axes 2 and 3 were noncontributory. Thus, compared with clinical source, clbB was much better able to resolve distinct bacterial subpopulations according to the assessed characteristics.

DISCUSSION

In this molecular epidemiological analysis of blood and fecal E. coli isolates, we found that two of the recently discovered colibactin synthesis genes, clbB and clbN, were significantly associated with bacteremia, being present in 58% of blood isolates but only 32% of fecal isolates (P = 0.003). Additionally, they were significantly associated with multiple other virulence genes, were confined to phylogenetic group B2, and within group B2 occurred within an especially high-virulence subset that exhibited extremely elevated virulence scores and a significantly increased likelihood of causing bacteremia.

Notably, however, in univariable analyses, clbB and clbN were less closely associated with bacteremia than were certain other traditionally recognized virulence genes. Likewise, according to multivariable logistic regression analysis, these markers were not a significant independent predictor of bacteremia, likely because they were overshadowed by more potent group B2-associated traits (i.e., ompT, sfa and/or foc, papA) (6) that in turn are associated with clbB or clbN. This leaves uncertainty as to whether the colibactin system per se contributes directly to virulence or instead is a marker for other traits that play a more direct role in pathogenesis. Experimental assessment of clbB or clbN mutants in an animal infection model are needed to answer this question.

In the PCoA, clbB explained a substantially greater proportion of total variance (according to bacterial traits) than did bacteremia versus fecal source, i.e., 59% (clbB) versus 16% (clinical source). This is probably because clbB is an excellent marker for group B2 and related virulence genes, which in turn were responsible for much of the variance in the molecular data set. In contrast, clinical source is significantly associated with, but somewhat removed from, these traits, with its effect being reduced by host status (which allows low-virulence strains to cause bacteremia in certain individuals) and the presence of ExPEC strains in the fecal flora of many uninfected individuals.

A similar statistical association between the pks island and sequences involved in adherence, iron acquisition, and lipopolysaccharide synthesis, as suggested by correspondence analysis, was noted by Le Gall et al. among group B2 isolates from diverse clinical sources and time periods (13). In that study, these same traits were associated with experimental mouse lethality, although primary associations and causality were not assessed. Similarly, in a molecular analysis of asymptomatic bacteriuria (ABU) E. coli isolates, Zdziarski et al. found clb sequences in five of six group B2 strains (including probiotic strain 83972), despite the absence of other typical B2-associated virulence traits, but in none of four non-B2 strains (25). This may indicate that the pks island contributes to sustained commensalism within the urinary tract in such group B2 strains. However, the small numbers, absence of a non-ABU comparison group, and uncertain functional status of the pks island in the strains leaves uncertainty regarding the implications of these findings.

In the present study, beside clbB and clbN, only 2 other markers among the 14 newly studied virulence genes were significantly associated with bacteremia in univariable analyses, and in the multivariable model, only 1 (pic) of the 14 genes was a significant predictor—and a negative one at that. This suggests that diminishing returns with respect to explaining bacteremia-versus-fecal source variance can be expected from adding more and more new virulence genes into molecular epidemiological surveys, unless markers that account for some of the currently unexplained variance can be discovered. The latter might apply particularly to non-B2-associated virulence traits, few of which are currently recognized, despite the not-inconsiderable representation of non-B2 isolates in clinical collections.

Our findings have several practical implications. First, on epidemiological grounds, i.e., the high prevalence of clbB and clbN among bacteremia isolates and the strong association of these genes with bacteremia, clbB and clbN appear to represent potential targets for preventive interventions, even if not the most strongly statistically supported (20). Experimental assessment is warranted. Second, other newly studied virulence genes (e.g., hra and vat) may also be potential candidate targets and hence also may deserve experimental assessment. Finally, no new “blockbuster” virulence markers were identified; moreover, no single virulence gene accounted for more than 45% of blood-fecal variance and only one accounted for more than 19%. This supports the notion that extraintestinal virulence in E. coli is multiply determined, i.e., likely involves not only a constellation of accessory traits (i.e., virulence factors) but also possibly metabolic functions and other phenotypes not traditionally considered virulence related (1, 5, 19).

The present study confirms previously demonstrated epidemiological association of clbB and clbN with clinical (as opposed to fecal) isolates and with group B2 (13, 16). Furthermore, it adds significantly to previous work by studying a well-defined clinical population, i.e., bacteremia isolates, by determining the associations of clbB and clbN with multiple other traditional and recently recognized virulence genes and by using multiple complementary analytical approaches to establish the relationships among these bacterial traits, including phylogenetic group and clinical source. Limitations include the single study locale, predominantly elderly male veteran study population, the modest sample size (with its risk of type 2 errors, although the many observed significant associations reduce this concern), use of multiple comparisons (with its risk of type 1 errors, which were guarded against by using PCoA and virulence scores), and reliance on presence/absence testing. Future study of isolates from different locales, syndromes, host groups, and time periods is needed to round out our understanding of the epidemiology of clbB and clbN.

In summary, we found that the recently discovered E. coli colibactin synthesis genes, clbB and clbN, (i) were significantly associated with bacteremia and with multiple other virulence genes, (ii) were confined to phylogenetic group B2, and (iii) within group B2 identified an especially high-virulence subset that exhibited extremely elevated virulence scores and an increased likelihood of causing bacteremia. These findings confirm and extend previous work regarding clbB and clbN and support investigation of the colibactin system as a potential target for preventive or therapeutic measures.