Using our predefined cutoff points for cluster analysis, we identified four clusters of VSEF (Figure 1). VSEF clustering seemed to be source-related. Clusters 1 (n=4) and 2 (n=12) contained surveillance isolates from community sources predominantly. All but one of the clinical infections isolates belonged to clusters 3 (n=66) and 4 (n=10), respectively.

On the basis of our calculations of the identification factors of VSEF and representative isolates of different VREF genogroups, we found that isolates of cluster 1 (n=4) resembled those of genogroup A, previously allocated to nonhospitalized patients and pig-derived VREF (Table 2). Similarly, isolates of cluster 2 (n=12) fitted best in genogroup B. Isolates from cluster 3 (n=66) showed almost equal resemblance to isolates from genogroups B and C, and isolates from cluster 4 (n=10) were also most identical to isolates from genogroup C (Table 2). The relationship between AFLP clusters of VSEF and sources was congruent with the previously described clustering of VREF. Almost all VSEF isolates (99%) derived from clinical infections clustered in clusters 3 and 4, clearly distinct from most VSEF isolates from community surveys (93%), which were found in clusters 1 and 2. A similar distribution was found previously among VREF with most (60%) isolates from clinical infections in genogroup C and most (89%) isolates from community surveys in genogroup A (11).

The presence of the variant esp gene in VREF and VSEF was strongly associated with a specific epidemiologic source because the presence of esp is higher in clinical infections and epidemic-associated isolates than in surveillance isolates (Figure 2). VREF isolates associated with nosocomial outbreaks were esp-positive, except for one. Prevalences of the variant esp gene in clinical infectious isolates were 57% and 40% for VSEF and VREF, respectively (p=ns). Prevalence of the variant esp gene in clinical and community survey isolates was low among VREF (6% and 3%, respectively) and completely absent among the 19 VSEF isolates tested.

Associations similar to the variant esp gene were found between ampicillin resistance and epidemiologic source for enterococcal isolates (Figure 2). All isolated associated with nosocomial VREF-outbreaks but one were resistant to ampicillin, as were 81% and 65% of infectious isolates of VSEF and VREF, respectively. Thirty-one percent of 36 nosocomial surveillance isolates of VREF were ampicillin resistant, as compared to none of five VSEF isolates obtained by clinical surveillance (p=ns). Finally, all but one isolate of VREF (n=36) and all VSEF isolates (n=14) obtained by surveillance of healthy persons were susceptible to ampicillin. When we combined these data, we found strong associations between the presence of the variant esp-gene and ampicillin resistance, both in VREF and VSEF: 98% of esp-positive VSEF and 92% of esp-positive VREF were resistant to ampicillin, as compared to 37% esp-negative VSEF and 20% esp-negative VREF isolates (p<0.0001).

The purK housekeeping gene was sequenced in 103 isolates: 64 VREF and 39 VSEF. The previously described type 1 allele was found in 39 isolates: 23 VREF and 16 VSEF. This specific allele type was associated with the presence of the variant esp-gene and ampicillin resistance but not with vancomycin resistance. The variant esp-gene was found in 25 (64%) of 39 isolates containing the purK type 1 allele and in only 1 (2%) of 64 isolates carrying other purK alleles (p<0.0001). Similarly, ampicillin resistance was detected in 36 (92%) of 39 isolates with purK type 1 allele and in only 5 (8%) of 64 isolates with other allele types (p<0.0001). In contrast, the vanA transposon was present in 23 (59%) of 39 isolates with purK type 1 allele and in 41 (64%) of 64 isolates with other allele types (p=ns).

Our study demonstrates the genetic relatedness of clusters of isolates of vancomycin-resistant and -susceptible E. faecium strains from different epidemiologic sources and provides evidence for selection of an E. faecium subtype associated with hospital outbreaks. This subtype is characterized by the presence of both ampicillin resistance and the variant esp gene. Furthermore, our findings suggest random horizontal spread of the vanA transposon to multiple genogroups of E. faecium. We hypothesize that the rise in infections caused by VREF resulted from nosocomial selection of a specific ampicillin-resistant E. faecium genotype harboring the variant esp gene and subsequent horizontal transfer of the vanA transposon.

Our study confirms that the previously demonstrated dichotomy between VREF isolated from healthy persons and patients (11) also exists for vancomcyin-susceptible E. faecium isolates. In VREF isolates, we could identify four genogroups, which were associated with particular hosts and environments and in which most isolates from healthy persons clustered distinctly from patient isolates. We showed that vancomycin-susceptible isolates clustered into three of these groups and that VSEF isolates from healthy persons also clustered distinctly from patient isolates. The genetic relationship between isolates and the genetic distinction between the four genogroups were based on AFLP analysis. We recently confirmed these findings with multilocus sequence typing (MLST) (13). Other researchers have also demonstrated host specificity of E. faecium. Quednau et al. suggested host specificity of isolates from chicken, pork, and humans by comparing restriction endonuclease profiles (26). In contrast, a recent study by Vancanneyt et al. also used AFLP; they did not confirm host-specific clustering of E. faecium (27). They described two main genomic groups in a population of 78 E. faecium strains isolated from seven European countries, and both groups comprised strains from (healthy) humans, animals, and food. All human clinical strains clustered in the largest genogroup, as did all strains (n=16) containing the vanA gene. Our findings that the vanA transposon was present in isolates of all genogroups proves that acquisition of this transposon was not influenced by and did not affect the preexisting relationship between the bacterium and host, which is a result of a long-term coevolution and mutual adaptation.

The presence of four genogroups in E. faecium seems to parallel the phylogenetic structure in E. coli, in which four ancestral groups (A, B1, B2, and D) have been described with some level of host specificity (28). However, in contrast to our finding of a single genetic lineage in E. faecium related to clinical symptoms and carrying the esp virulence gene, clinical isolates of E. coli are more widely distributed among the different ancestral groups (29). Furthermore, MLST of pathogenic E. coli strains showed that different ancestral lineages have acquired the same virulence factors (30), indicating that pathogenic potential in E. coli is not confined to a single ancestral lineage, which is suggested by our findings for E. faecium. Human and animal pathogenic E. coli strains share closely related genotypes and carry similar virulence factor profiles, suggesting that certain E. coli strains are pathogenic for both animals and humans (31). Whether this holds true for pathogenic E. faecium strains is unknown.

We recently found that the presence of the variant esp-gene is associated with nosocomial outbreaks of VREF in three continents, although this gene was not found in VREF strains isolated from healthy persons or animals (12). The outbreak strains were also characterized by a specific allele type of the purK gene, one of the housekeeping genes sequenced in the MLST method (13). Recently, other investigators reported the presence of the variant esp gene in clinical isolates of VSEF, demonstrating that this gene is not linked specifically to the vanA transposon (15–18). The findings of our study confirm the strong association between the presence of the esp gene and the relation with hospital outbreaks and clinical infections among patients with VREF as well as VSEF. Although the esp gene is virtually absent among community isolates, the presence of esp among VSEF and VREF from clinical infectious sites apparently unrelated to hospital outbreaks implies that this gene is not exclusively related to epidemic strains. Excluding the outbreak potential of the esp-positive VSEF strains in this and other studies is difficult. Only few outbreaks with VSEF have been documented (32–34). We have investigated and could not demonstrate the presence of the variant esp gene in isolates from one hospital outbreak of ampicillin-resistant E. faecium in Norway (data not shown).

Little is known about the function of the variant esp-gene, although epidemiologic findings support its role as a virulence factor. In E. faecalis, the homologue of this gene encodes for the enterococcal surface protein, and the presence of this gene has been associated with enhanced adherence capacities to uroepithelial surfaces, but not with increased virulence, in a mice model (35). Moreover, in E. faecalis, the esp-gene was highly associated with biofilm-formation capacity (36). Increased adherence capacities and biofilm formation of esp-positive E. faecium strains might explain its association with hospital outbreaks.

Like the variant esp gene, ampicillin resistance was found more frequently in isolates associated with infections and nosocomial outbreaks, both in VSEF and VREF. This source-relationship is probably caused by selective pressure of β-lactam antibiotics that are used extensively in hospitals. Emergence of ampicillin resistance in E. faecium was already demonstrated in the early 1980s and seemed to precede the emergence of vancomycin resistance by 10 years (2). A correlation between high prevalences of the esp gene and antibiotic resistance among E. faecium isolates from hospitalized patients was also reported recently by Coque et al. (17).

Considering the sources of isolates, presence of ampicillin and vancomycin resistance, the presence of the variant esp gene, and the type 1 allele of the purK gene, we propose an evolutionary scheme for the specific genogroup of E. faecium associated with nosocomial outbreaks (Figure 3). However, our findings might have been biased by the composition of our collection of isolates and the fact that the purK was sequenced in a subset of all isolates. The esp gene and ampicillin resistance can obviously co-occur in a distinct genetic lineage of E. faecium characterized by the type 1 allele of the purK gene. We propose that E. faecium strains containing the type 1 allele of the purK gene have acquired the esp virulence gene and that this E. faecium genotype (purK­-1, esp-positive) is prominently present among clinical relevant strains and virtually absent among survey isolates. Yet another substantial part of the clinical relevant strains with genotype purK-1 do not carry the esp gene, which emphasizes that other virulence genes of E. faecium apart from esp are involved in the development of infections. Ampicillin resistance was predominantly found among the purK-1 genotype and is almost absent among other E. faecium genotypes. This occurrence of resistance is, presumably, the result of selective antibiotic pressure. Chromosomal linkage of the purK-1 allele, the variant esp gene, and ampicillin resistance could have promoted this selection. Finally, glycopeptide usage in and outside hospitals, both in humans and animals, resulted in the selection of vancomycin-resistant strains in both the purK-1 genotype and the other genotypes. The presence of similar proportions of vancomycin resistance in all genotypes probably reflects horizontal transfer of the vancomycin-resistance transposon. This hypothesis implies the development of a hospital-adapted genogroup of E. faecium, characterized by the type-1 allele of purK, the variant esp-gene and ampicillin resistance, which has spread unnoticed, thereby creating a pool of strains with epidemic potential. Only after becoming vancomycin-resistant has this genogroup become recognized as clinically relevant.