Aggregative adherence fimbriae contribute to the inflammatory response of epithelial cells infected with enteroaggregative Escherichia coli


  • Susan M. Harrington,

    1. 1 Departments of Microbiology and Immunology, The University of Maryland, Baltimore, MD 21201, USA.
      2 Departamento de Microbiologia, Imunologia e Parasitologia, Universidade Federal de São Paulo, São Paulo SP, Brazil.
      Departments of 3Pediatrics and 4Medicine, and 5Center for Vaccine Development, The University of Maryland School of Medicine, Baltimore, MD 21201, USA.
    Search for more papers by this author
  • 1 Maura C. Strauman,

    1. 1 Departments of Microbiology and Immunology, The University of Maryland, Baltimore, MD 21201, USA.
      2 Departamento de Microbiologia, Imunologia e Parasitologia, Universidade Federal de São Paulo, São Paulo SP, Brazil.
      Departments of 3Pediatrics and 4Medicine, and 5Center for Vaccine Development, The University of Maryland School of Medicine, Baltimore, MD 21201, USA.
    Search for more papers by this author
  • 1 Cecilia M. Abe,

    1. 1 Departments of Microbiology and Immunology, The University of Maryland, Baltimore, MD 21201, USA.
      2 Departamento de Microbiologia, Imunologia e Parasitologia, Universidade Federal de São Paulo, São Paulo SP, Brazil.
      Departments of 3Pediatrics and 4Medicine, and 5Center for Vaccine Development, The University of Maryland School of Medicine, Baltimore, MD 21201, USA.
    Search for more papers by this author
    • Present address: Institute of Child Health, University of Birmingham, B4 6NH, Birmingham, UK.

  • and 2 James P. Nataro 1,3,4, 5

    Corresponding author
    1. 1 Departments of Microbiology and Immunology, The University of Maryland, Baltimore, MD 21201, USA.
      2 Departamento de Microbiologia, Imunologia e Parasitologia, Universidade Federal de São Paulo, São Paulo SP, Brazil.
      Departments of 3Pediatrics and 4Medicine, and 5Center for Vaccine Development, The University of Maryland School of Medicine, Baltimore, MD 21201, USA.
    Search for more papers by this author

E-mail; Tel. (+1) 410 706 7376; Fax (+1) 410 706 6205.


Enteroaggregative Escherichia coli (EAEC) causes watery diarrhoea that is often mildly inflammatory. Previous studies have reported that the flagellin of EAEC induces IL-8 from intestinal epithelial cells (IECs) in culture. To characterize more fully the inflammatory response to EAEC, we infected IECs with EAEC prototype strain 042 and assessed cellular responses by macroarray and reverse transcriptase polymerase chain reaction (RT-PCR). Genes upregulated in 042-infected non-polarized T84 cells included IL-8, IL-6, TNF-α, GRO-α, GRO-γ, ICAM-1, GM-CSF and IL-1α. RT-PCR analyses performed with cDNA from T84 and HT-29 cells infected with an aflagellar mutant (042fliC) suggested that these responses were primarily mediated by flagellin. To better reproduce the conditions of the infection for this non-invasive pathogen, we assessed the responses of polarized IECs to strain 042 infection. As expected, 042 induced IL-8 production from both polarized T84 and HT-29 cells. However, significant IL-8 secretion was induced in polarized T84 cells infected with 042fliC, suggesting that a factor other than flagellin contributes to inflammation in this model. This non-flagellar IL-8 response required expression of the aggregative adherence fimbria (AAF) adhesin, and was related to the presence of the minor fimbria-associated protein AafB. Our data suggest that multiple factors contribute to EAEC-induced inflammation, and further characterization of the nature of EAEC proinflammatory factors will greatly advance our understanding of this emerging pathogen.


Enteroaggregative Escherichia coli (EAEC) is a diarrhoeagenic pathogen initially associated with persistent diarrhoea in children in the developing world (Bhan et al., 1989; Wanke et al., 1991; Germani et al., 1998), but which is now emerging in other clinical settings (Huppertz et al., 1997; Pabst et al., 2003; Huang and Dupont, 2004). Recently, EAEC has been recognized as a significant cause of traveller's diarrhoea, with only enterotoxigenic E. coli being a more common pathogen in this population (Gascon et al., 1998; Vargas et al., 1998). EAEC disease is typically characterized by a mucoid, watery diarrhoea, often accompanied by fever, nausea and vomiting (Cobeljic et al., 1996; Huppertz et al., 1997; Shazberg et al., 2003).

EAEC adherence to the human intestinal mucosa requires expression of aggregative adherence fimbriae (AAFs), which promote the formation of a thick biofilm (Elias et al., 1999). Mucosal secretion may then be induced by the production of one or more bacterial enterotoxins and cytotoxins (Savarino et al., 1991; Nataro et al., 1996; Henderson et al., 1999). At least three distinct AAF alleles have been described (Czeczulin et al., 1997; Nataro et al., 1992; Bernier et al., 2002). All known AAF alleles are encoded on virulence plasmids called collectively pAA. The EAEC prototype strain and proven pathogen, 042, expresses the AAF/II allele, encoded on plasmid pAA2. Cota et al. (Cota et al., 2004) have recently proposed that the AAF adhesins are composite structures composed of a homopolymeric fimbrial shaft (comprising the AafA protein in AAF/II) and a minor fimbria-associated adhesion, represented by AafB in AAF/II. The AafB protein is related to AfaD of uropathogenic E. coli, and mediates binding to β1 integrin and subsequent internalization of the bacteria (Plancon et al., 2003). However, Bernier et al. have found that though the AfaD homologues of EAEC are generally able to confer invasion of epithelial cells, the AafB allele of strain 042 is uniquely non-invasive (Bernier and Le Bouguenec, 2002). This is significant in light of the fact that strain 042 is pathogenic in adult volunteers and that an aafB null mutant of 042 adhered as well as the parent to human colonic tissue sections in vitro (Elias et al., 1999). Thus, despite its universal presence in AAF alleles, the contribution of AafB and of EAEC AfaD homologues remains unknown.

Also encoded on pAA is an AraC homologue, AggR, which is emerging as the master virulence regulator for EAEC (Nataro et al., 1994; Sheikh et al., 2002; Nishi et al., 2003). This transcriptional activator regulates the genes involved in fimbrial biogenesis, formation of the EAEC dispersin protein coat (Sheikh et al., 2002), and secretion of the dispersin protein (Nishi et al., 2003). In addition to these plasmid-encoded genes, AggR has recently been shown to activate a chromosomal operon (E. Dudley and J. Nataro, unpublished).

Several lines of evidence suggest that EAEC infection is mildly inflammatory in nature. Epidemiological reports have documented elevated fecal lactoferrin, IL-8 and IL-1β among infected infants in developing countries and adult travellers to India and Mexico (Steiner et al., 1998; Bouckenooghe et al., 2000; Greenberg et al., 2002). Importantly, EAEC enteritis may be associated with growth impairment in children even in the absence of diarrhoea (Steiner et al., 1998).

Determining the contribution(s) of EAEC virulence factors to this inflammatory response is an area of active research. Steiner et al. have shown that EAEC strain 042 induces IL-8 release from non-polarized Caco-2 intestinal epithelial cells (IECs)(Steiner et al., 1998); this study also reported that the pAA plasmid was required for the full inflammatory effect. In a subsequent publication, these investigators reported that a mutation in the gene encoding flagellin (fliC) abrogated IL-8 release, implicating flagellin as the major proinflammatory stimulus (Steiner et al., 2000). Of note, however, Jiang et al. have reported that significantly more IL-8 was detected in feces of travellers infected with EAEC strains harbouring the plasmid-borne aggR or aafA genes, compared with those infected with virulence factor-negative EAEC (Jiang et al., 2002). Thus, evidence exists to suggest that both chromosomal (i.e. fliC) and pAA plasmid-borne determinants may contribute to cytokine release by EAEC.

Flagellin has been reported to be a mediator of inflammation induced by Salmonella species in vitro, and for other pathogenic E. coli (Berin et al., 2002; Eaves-Pyles et al., 2001; Zhou et al., 2003). Flagellin has been shown to require Toll-like receptor 5 (TLR5) to promote signalling through the NF-κB and MAP kinase pathways, leading to induction of proinflammatory cytokines (Eaves-Pyles et al., 2001; Hayashi et al., 2001; Smith et al., 2003). However, TLR5 has been localized to the basolateral surface of the IEC (Gewirtz et al., 2001a). Thus, access to the receptor is limited to flagellin from enteric bacteria that invade the epithelial cell or breach this barrier in some other manner. Using polarized T84 cells, Gewirtz et al. demonstrated that Salmonella enterica ser. Typhimurium applied to either the apical or basolateral surface elicited an IL-8 response, whereas non-pathogenic E. coli only induced such a response if applied basolaterally, and that a fliC/fljB Salmonella mutant was not proinflammatory (Gewirtz et al., 2001b). Although direct interaction of Salmonella with IECs is required for inflammation, the Salmonella pathogenicity island-1 (SPI-1) type III secretion system, which mediates invasion by Salmonella, appears to be unnecessary (Zeng et al., 2003). Recent reports suggest that Salmonella promotes flagellin transcytosis via SPI-2 mediated vesicular trafficking, allowing interaction with TLR5 on the basolateral cell surface (Gewirtz et al., 2001b; Hobert et al., 2002; Lyons et al., 2004).

In this study, we sought to characterize more fully the inflammatory response to infection with EAEC and in particular, to determine whether factors other than flagellin were proinflammatory. In addition to IL-8 and IL-1β, notably found in the feces of infected patients, other proinflammatory genes were upregulated in response to EAEC infection, with neutrophil chemokines dominating the response. We hypothesized that intestinal cell lines other than Caco-2, either polarized or non-polarized, would respond to pathogen-specific proinflammatory factors in EAEC. For non-polarized T84 and HT-29 cells, the response was primarily flagellin-dependent. Interestingly, however, though polarized cells appeared to respond to flagellin, an additional IL-8 response was attributed to a plasmid-derived EAEC factor(s) only in polarized T84 cells.


A broad inflammatory response in EAEC infected non-polarized cells is flagellin-dependent

To characterize the response of intestinal cells other than Caco-2, we selected two well-studied lines, T84 and HT29-C1. The T84 IEC is a crypt-like cell line that has long been used to study secretion, transcellular trafficking and inflammation (Lencer et al., 1995; Gewirtz et al., 2001b; McCormick, 2003). We have previously described the T84 cell line as a model for EAEC adherence and mucosal toxicity (Nataro et al., 1996). A whole bacterium approach was taken to characterize the inflammatory response of EAEC-infected T84 cells and to determine the role of flagellin in that response. T84 cells grown to confluency, but not under conditions that would permit polarization, were infected for 3 h with EAEC 042 (strains are described in Table 1). The Panorama Human Cytokine Array v1.2™ was used to identify cellular genes whose expression was upregulated following 042 infection. Genes upregulated threefold or more relative to uninfected cells on each of two macroarray experiments are shown in Table 2. Dominating the response were the genes for several neutrophil chemokines, including IL-8, GRO-α, GRO-β, GRO-γ, and fractalkine. Transcripts for IL-1α, IL-6 and TNF-α were also upregulated. Reverse transcriptase polymerase chain reaction (RT-PCR) for specific transcripts was used to confirm the array results. For eight of the 13 genes in Table 2, there was a noticeably greater response in cells infected with 042 compared with uninfected cells (Fig. 1): these included IL-8, IL-6, TNF-α, IL-1α, GRO-α, GRO-γ, GM-CSF and ICAM-1. MIP-3α, reported to stimulate migration of dendritic cells in response to Salmonella flagellin (Sierro et al., 2001), was induced in one of two array replicates, and RT-PCR confirmed induction of this transcript. We also quantified IL-1β mRNA, because this cytokine was reported to be elevated in the feces of patients infected with EAEC; RT-PCR revealed significant induction of IL-1β transcript. Though upregulated on the array, fractalkine, GRO-β, 4-1BB and iNOS mRNA expression were not apparently upregulated when quantified by RT-PCR.

Table 1.  Characteristics and source of host strains used in this study.
StrainRelevant characteristicsReference
  1. Ap, ampicillin resistance; Tc, Tetracycline resistance; Cm, chloramphenicol resistance; Km, kanamycin resistance; Sm streptomycin resistance.

042Prototype EAEC strain (044:H18). Sm, Tc, CmNataro et al. (1995)
042fliC042 harbouring suicide plasmid pJP5603 inserted into the fliC gene. KmSteiner et al. (2000)
042aafA 3.4.14042 harbouring TnphoA inserted into the aafA gene, which encodes the major fimbrial subunit of AAF/II. KmCzeczulin et al. (1997)
042aafA 2.94042 harbouring TnphoA inserted into the aafA gene at a different site than in mutant 3.4.14. KmCzeczulin et al. (1997)
042aggR042 harbouring pJP5603 inserted into aggR. KmNataro et al. (1994)
042aggR(pAggR)042aggR complemented with aggR expressed under arabinose control in plasmid pBAD30. Km ApSheikh et al. (2002)
042aafB042 harbouring an insertion of suicide plasmid pJP5603 into the aafB gene. KmElias et al. (1999)
S. enterica ser. Typhimurium CS401Wild type non-typhoid Salmonella strainRakeman et al. (1999)
HSCommensal E. coli, motile and non-pathogenic in volunteers.Levine et al. (1978)
HS(pAA2)HS transformed with pAA2 harbouring pJP5603 integrated into a silent locusThis work
HS(pAA2aafB)HS transformed with pAA2 carrying pJP5603 inserted in aafB, which encodes a minor protein associated with AAF/IIThis work
GDI20Commensal E. coli; non-motile, not inflammatory on human cells.Steiner et al. (2000)
Table 2.  Genes upregulated following infection of non-polarized T84 cell monolayers.
Gene productFold induction Experiment 1/experiment 2
  1. RNA was extracted following a 3 h infection with EAEC 042 or from uninfected non-polarized T84 cells. cDNA was hybridized to the Panorama Human Cytokine Array v2.1 as described in Experimental procedures. Intensity data were normalized to a set of housekeeping genes. Fold induction of genes upregulated threefold or more by 042 infection relative to uninfected cells from each of two macroarray hybridizations are shown.

IL-1α 7.9/3.7
GM-CSF 4.8/6.2
iNOS 4.8/4.7
Fractalkine 4.2/4.9
Integrin-β2 3.6/3.9
4-1BB 3.6/11.9
Figure 1.

RT-PCR confirmation of cytokines upregulated from non-polarized T84 cells infected with EAEC 042. RNA was extracted from cells 3 h after infection with 042 or 042fliC or from uninfected T84 cells (U). GAPDH is included as an internal control.

We evaluated the contribution of the EAEC flagellin to the profiles of upregulated inflammatory genes by comparing cytokine mRNA induction in response to 042 and 042fliC. As shown in Fig. 1, all of the genes that were induced by 042 infection of T84 cells appeared to be diminished in 042fliC infection. For most of the cytokine genes induced by 042, induction from the aflagellar mutant is visually indistinguishable from that of uninfected cells. For three of the more highly upregulated genes, however, TNF-α, IL-8 and GRO-α, 042fliC appears to induce a cytokine response slightly above that of uninfected T84 cells. These data suggest that flagellin from EAEC 042 is a major mediator of the inflammatory response in infected non-polarized T84 cells, but that perhaps some non-flagellin mediated inflammatory response may exist as well.

The Human Panorama Cytokine Array was used to analyse the response of a different intestinal cell line, HT-29 C1, to infection with 042, 042fliC and the commensal E. coli GDI20, a non-motile strain expressing a low level of flagellin (Steiner et al., 2000). By array analysis, 40 of 847 genes were significantly upregulated by 042 infection of non-polarized HT29 cells. The expression profiles of the two cells lines differed only by induction of COX-2 in the HT29 cells alone. No inflammation-related genes were upregulated by 042fliC or GDI20 compared with uninfected HT29 cells (data not shown). These data suggest that the responses of EAEC-infected non-polarized HT29 and T84 cells are generally similar, and in both cases, the inflammatory response is largely dependent upon flagellin.

Enteroaggregative E. coli induces IL-8 from polarized cells

As receptors on IECs may be differentially distributed to the apical or basolateral domains, polarized cells may provide a more realistic model for studying intestinal infection (Gewirtz et al., 2001a). The response of polarized HT-29 cells to EAEC 042 infection was generally similar to that of non-polarized cells. Three hours after infection, interleukin-1β, MIP-3α, IL-8 and COX-2 were each strongly upregulated by infection of polarized HT-29 cells with 042, but not by 042fliC, GDI20 or uninfected cells (Fig. 2). Significantly, a cytokine response above that of the non-flagellated strains was observed for the flagellated commensal E. coli, HS. To assess the relationship between transcriptional responses and synthesis of proinflammatory cytokines, IL-8 was chosen as a marker, because this transcript was consistently upregulated in all RT-PCR assays. For these experiments, bacteria were applied to the apical surface for 3 h and medium was sampled from the basolateral compartment 24 h later. Infection of polarized HT29 cells by 042 yielded high levels of IL-8 (Fig. 3A). The IL-8 response seen with 042 infection was significantly higher than that induced by 042fliC, GDI20 and the flagellated commensal E. coli strain, HS. IL-8 levels observed after infection with 042fliC were similar to those seen in uninfected cells. To confirm the responsiveness of polarized HT29 cells, we added purified flagellin to the apical and basolateral compartments and measured IL-8 release. Flagellin added apically elicited a weak IL-8 response (1.2 ± 0.2 ng ml−1), compared with that seen when flagellin was applied basolaterally (16.9 ± 0.7 ng ml−1) (Fig. 3B), consistent with previous reports that TLR-5 is located predominantly on the basolateral plasma membrane (Gewirtz et al., 2001a; Bambou et al., 2004).

Figure 2.

RT-PCR analysis of cytokine expression in polarized HT-29 cell infections. RNA was extracted from cells 3 h after infection with 042, 042fliC, GDI20, HS, or Salmonella enterica serotype Typhimurium (CS401), or from uninfected cells (U).

Figure 3.

IL-8 induced by infection of polarized HT-29 cells. Medium was collected from the basolateral compartment 24 h after (A) a 3 h infection with 042, 042fliC, GDI20, HS, or Salmonella enterica serotype Typhimurium (CS401), or from uninfected cells (U); or (B) purified flagellin applied apically or basolaterally. Medium was analysed by ELISA as described in Experimental procedures. Bars represent the means for three experiments, with error bars indicating ± one standard deviation. In A, significantly more IL-8 was released by 042 infection compared with 042fliC, GDI20, HS or medium alone (P < 0.05). Analysis was performed by one-way anova followed by pairwise comparisons with Bonferroni error correction as appropriate.

T84 cells typically form strong epithelial tight junctions, so we were particularly interested in addressing flagellin responsiveness in this cell line when polarized. Figure 4 illustrates IL-8 levels released from the basolateral side of polarized T84 cells at 24 h after a 3 h apical infection with 042, 042fliC, GDI20, HS and S. ser. Typhimurium. Like polarized HT-29 cells, polarized T84 cells released more IL-8 upon incubation with 042 compared with commensal E. coli and uninfected T84 cells, though levels were not as high as with HT29 cells. Importantly, however, the IL-8 response caused by infection with 042fliC was significantly greater than that of either commensal strain, suggesting that an EAEC factor other than flagellin may contribute to the inflammatory response elicited in polarized T84 cells, but not in HT-29 cells.

Figure 4.

IL-8 induced by infection of polarized T84 cells. Medium was collected from the basolateral compartment 24 h after a 3 h infection of polarized cells with 042, 042fliC, GDI20, HS, or S. enterica serotype Typhimurium (CS401) (as positive control) or from uninfected cells (U), and was analysed by ELISA. Bars represent the means of six experiments each performed in triplicate wells, with error bars indicating ± one standard deviation. Significantly more IL-8 was induced by infection with either 042 or 042fliC compared with GDI20, HS or uninfected cells. One-way anova was performed followed by multiple pairwise comparisons using a Bonferroni error correction (P < 0.05).

A time course experiment with polarized T84 cells indicated that IL-8 secretion increased over time up to 12 h after a 3 h infection, and remained high until at least 24 h (Fig. 5). The average flagellin-dependent response of polarized T84 cells is approximately 40% of the total IL-8 response at all three time points. At all time points following the 3 h sample, there was a significant difference between the IL-8 expression elicited by 042fliC infection compared with uninfected or HS-infected cells, revealing consistently the non-flagellar proinflammatory response.

Figure 5.

Time course of IL-8 induced by infection of polarized T84 cells. Medium was collected from the basolateral compartment at the times indicated following a 3 h infection of polarized cells with 042 (black bars), 042fliC (diagonally striped bars), GDI20 (grey bars) or HS (white bars), or from uninfected cells (horizontally striped bars), and was analysed by ELISA. Bars are the mean of quadruplicate wells from one representative experiment, with error bars indicating ± one standard deviation. Mean IL-8 induced by infection with either 042 or 042fliC is significantly greater than that induced by GDI20, HS or uninfected cells at 6, 12 and 24 h. Statistical analysis comprised one-way anova followed by multiple pairwise comparisons using Bonferroni error correction (P < 0.05).

Our initial array and RT-PCR experiments described above in non-polarized cells were performed immediately at the termination of a 3 h infection. To determine whether non-polarized T84 cells would release IL-8 in response to 042fliC at later time points, we evaluated non-polarized T84 cells three and 24 h after infection. As shown in Fig. 6, 042fliC induced little IL-8 production from non-polarized T84 cells at 3 h. At 24 h, however, there was a modest but significant IL-8 response elicited by 042fliC, higher than that caused by infection with GDI20. For the non-polarized T84 cells, the flagellin-dependent response represents approximately 80% of the total IL-8 response at 24 h.

Figure 6.

Time course of IL-8 production after infection of non-polarized T84 cells. Medium was collected from non-polarized T84 cells three and 24 h after the start of a 3 h infection with 042 (black bars), 042fliC (diagonally striped bars), GDI20 (grey bars) and HS (white bars) or uninfected cells (striped bars) and analysed by ELISA. Bars represent the means of triplicate wells from a representative experiment, with error bars indicating ± one standard deviation. Statistical analysis was performed on results at 24 h in one-way anova followed by multiple pairwise comparisons using a Bonferroni error correction as appropriate (P < 0.05). Significantly more IL-8 was induced by 042, 042fliC or HS infection compared with uninfected cells at 24 h.

We next sought to compare flagellin and non-flagellin responses for additional cytokines in polarized T84 cells. RT-PCR was performed for inflammatory factors on RNA extracted at 6 h after infection of T84 cells with 042, 042fliC, GDI20 and HS. Cytokine mRNAs induced in response to the non-flagellar factor(s), as demonstrated with 042fliC infection, include TNF-α, GRO-α, MIP-3α, IL-1α, IL-1β and GM-CSF in addition to IL-8 (Fig. 7). As with non-polarized cells, 042fliC did not upregulate ICAM-1 and GRO-γ. Surprisingly, only a low-level of IL-6 transcript was detected from infected polarized cells. Thus, while the flagellar and non-flagellar factors induce a common set of inflammatory mediators, some components of the response appear to differ.

Figure 7.

RT-PCR analysis of cytokine expression from infected polarized T84 cells. RNA was extracted at the 6 h time point after a 3 h infection with 042, 042fliC, GDI20, HS or uninfected T84 cells (U). Cytokines chosen for analysis were those previously upregulated by EAEC infection of non-polarized T84 cells. GAPDH is included as an internal control.

To determine whether this transcriptional response correlated with detectable cytokine production, tissue culture supernatants collected 24 h after a 3 h infection of both polarized and non-polarized T84 cells were analysed with the BD™ Cytometric Bead Array (CBA) Human Inflammation kit, which measures IL-12p70, TNF-α, IL-10, IL-6, IL-1β and IL-8. Only IL-8 was detected in significant amounts in response to infection of polarized cells with either 042 or 042fliC(Table 3). Similar to ELISA results at the 24 h time point, significant amounts of IL-8 were detected after infection of non-polarized cells with 042 and with HS. Non-polarized cells infected with 042 elicited an IL-6 response that was low, but significantly different from uninfected cells. The IL-6 response to 042fliC infection was not significant, suggesting that the IL-6 response from non-polarized cells is mediated by flagellin and not the non-flagellar factor(s). Although transcripts for TNF-α and IL-1β were clearly observed following infection of either polarized or non-polarized T84 cells, no TNF-α or IL-1β protein expression was determined above the level of detection. Purified flagellin added to the basolateral side of polarized T84 cells elicited only an IL-8 response, suggesting that production of TNF-α, IL-1β and IL-6 from polarized cells may be low under these conditions.

Table 3.  Cytokine expression from experimentally infected T84 cells.
StrainIL-6 (pg ml−1)IL-8 (pg ml−1)
  • a

    . Significantly more cytokine was produced following infection with 042, 042fliC, or HS compared with uninfected cells.

  • b

    . Cytokine production following infection with 042 was significantly greater than following infection with 042fliC. Statistical analysis was performed by one-way anova, followed by pairwise comparisons with Bonferroni error correction (P < 0.05).

  • Cytokine secretion after a 3 h infection of polarized and non-polarized T84 monolayers with 042, 042fliC and HS, was compared with that produced by uninfected cells. Medium was collected 24 h after termination of infection and analysed with the BDCytometric Bead Array. Numbers in the table are mean values in pg ml −1 ± one standard deviation. Secretion of IL-12p70, TNF-α, IL-10 and IL- were all below limits of detection (5 pg ml−1).

042≤ 5.036.6 ± 10.1a254.7 ± 17.7a1618.7 ± 306.7a
042fliC≤ 5.015.2 ± 3.5b177.3 ± 29.7a,b 412.3 ± 39.2b
HS≤ 5.022.0 ± 3.7 19.7 ± 8.2 715.9 ± 43.5a
Uninfected T84≤ 5.0 9.5 ± 0.4 17.1 ± 3.6 173.5 ± 43.5
Flagellin alone≤ 5.0Not done250.6 ± 15.2Not done

Characterization of the non-flagellar proinflammatory factor

Previous reports by Steiner (Steiner et al., 1998) and Huang (Huang et al., 2004) suggested that plasmid-mediated virulence factors may play a role in the inflammatory response to EAEC. However, these studies used non-polarized Caco-2 and Hct-8 cells, respectively, to evaluate the IL-8 response to infection and the latter study did not control for the presence of flagella. Polarized T84 cells were infected with 042, 042aafA, 042aggR and 042aggR(pBADaggR). Two different aafA mutants (3.4.14 and 2.94) were employed. As shown in Fig. 8A, IL-8 production elicited by the 042aggR mutant was significantly less than in response to 042, and was similar to that of uninfected T84 cells. Complementation of the mutation restored IL-8 production to levels greater than the wild-type parent, suggesting that the AggR virulence gene regulator was required for full IL-8 induction. Furthermore, infection with AAF/II mutant 042aafA 3.4.14, which is non-adherent to epithelial cells (Sheikh et al., 2001) elicited a reduced IL-8 response compared with 042 (Fig 8A and B). The second fimbrial mutant, 042aafA 2.94, which synthesizes a non-adhering AAF/II pilus (J. Nataro, unpubl. obs.) also elicited significantly less IL-8 from polarized T84 cells compared with the prototype strain (Fig. 8B).

Figure 8.

IL-8 induced by infection of polarized T84 cells. Medium was collected from the basolateral compartment 24 h after a 3 h infection of polarized cells with (A) 042, 042aafA 3.4.14, 042aggR, 042aagR (pBADaggR), or uninfected cells (U); and (B) 042, 042aafA 3.4.14, 042aafA 2.94, or from uninfected cells (U). Bars show the means of IL-8 levels as measured by ELISA in a representative experiment performed in triplicate, with error bars representing ± one standard deviation. Statistical analysis was performed by one-way anova followed by multiple pairwise comparisons with Bonferroni error correction. In A, P < 0.05 for 042 and for 042aggR(pBADaggR) compared with 3.4.14, 042aggR and uninfected cells. In B, P < 0.05 for 042 compared with 3.4.14, 2.94 and uninfected cells.

The AAF/II adhesin has been suggested to be a composite structure, comprising the AafA major fimbrial adhesin, and a minor adhesin, AafB (Cota et al., 2004). AafB of strain 042 is related to the AfaD protein, which mediates limited cellular invasion of uropathogenic E. coli via binding to β1-integrin (Plancon et al., 2003). We have previously found that AafB does not quantitatively affect binding to intestinal or non-IECs or to the intestinal mucosa in organ culture (Elias et al., 1999). As binding to integrins has been suggested to be a mechanism of IL-8 induction, we tested an aafB mutant of 042 for the ability to elicit IL-8 from polarized epithelial cells. As seen in Fig. 9, the aafB mutant elicited lower levels of IL-8, though this level was above that seen in uninfected cells, or that induced by E. coli HS. Complementation of aafB on a multicopy plasmid impaired bacterial growth and elicited no increased IL-8 induction compared with uninfected cells (not shown), so we pursued an alternative approach to confirming the role of aafB. E. coli HS was transformed with the pAA2 plasmid carrying an antibiotic resistance gene in a silent locus; the ability of this strain to induce IL-8 from polarized T84 cells was compared with that of HS or HS carrying pAA2 with an insertion in aafB. HS(pAA2) and HS(pAA2aafB) adhered to epithelial cells at a similar level, and both at levels higher than the HS parent (not shown). However, HS(pAA2) increased IL-8 production approximately fourfold more than did HS alone or HS(pAA2aafB) (Fig. 9B), suggesting that AafB is the predominant plasmid-encoded IL-8 inducer of strain 042.

Figure 9.

IL-8 induced by infection of polarized T84 cells. Medium was collected from the basolateral compartment 24 h after a 3 h infection. In A, cells were infected with 042 or 042aafB or were uninfected (U). In B, cells were infected with HS, HS(pAA2), HS(pAA2aafB) or were uninfected (U). Bars represent mean IL-8 as measured by ELISA from a representative experiment performed with triplicate wells. Error bars indicate ± one standard deviation. In A, significantly more IL-8 was produced by 042 compared with 042aafB. In B, significantly more IL-8 was induced by infection with HS(pAA2) compared with HS, HS(pAA2aafB) or mock-infected cells. Analysis was performed by one-way anova, followed by multiple pairwise comparisons with Bonferroni error correction (P < 0.05).


Multiple lines of evidence suggest that EAEC induces an inflammatory enteritis, yet the factor(s) responsible for this clinical presentation are not known with certainty. In this study, we have further characterized the cellular response to EAEC infection, have identified a model system (polarized T84 cells) that may reveal additional proinflammatory factors, and have presented data suggesting that the adherence factor AAF/II may contribute to the inflammatory response.

We were initially interested in characterizing the contribution of flagellin to the EAEC response. We infected HT-29 and T84 cells with prototype strain 042 and utilized an array formatted with select cDNAs representing inflammatory mediators. Our array analyses revealed a general acute proinflammatory reaction in non-polarized cells that was largely due to the presence of flagella on the infecting strain. The predominant proinflammatory factors detected were the neutrophil and dendritic cell chemokines, including IL-8, the GRO chemokines and MIP-3α, as well as IL-1β. Notably, both IL-1β and IL-8 were previously observed in stool samples from patients with EAEC diarrhoea (Steiner et al., 1998; Greenberg et al., 2002).

Previous reports implicated flagellin as the principle proinflammatory factor in EAEC (Steiner et al., 1998; 2000; Donnelly and Steiner, 2002; Khan et al., 2004). We reasoned that if this were so, we would observe an effect of flagellin in polarized monolayers. EAEC 042 infection of polarized HT-29 cells induced an IL-8 response significantly above that seen with uninfected cells, or cells infected with commensal E. coli or 042fliC, consistent with results seen with non-polarized cells. We also tested the role of flagellin in polarized T84 cells; polarized T84 cells with resistance > 1000 Ω have been shown to exhibit a substantially higher concentration of flagellar receptor TLR5 on the basolateral surface than on the apical surface (Gewirtz et al., 2001a). In this model system, the contribution of flagella to IL-8 release was diminished. Significantly, both 042 and 042fliC induced IL-8 production from polarized T84 cells, suggesting that a non-flagellar factor contributes substantially to IL-8 production in this model. The partial flagellar response of polarized T84 cells to 042 infection could have been due to monolayer damage as we have previously reported (Nataro et al., 1996), and we did observe a decrease in T84 monolayer resistance during 042 infection (not shown). The response of polarized HT29 cells to flagella is presumably due to accessability of TLR5 in this cell line, which does not maintain as formidable a barrier between apical and basolateral compartments (resistance is maximally between 300 and 400 Ω in our hands), though our polarized HT29 cells were more responsive to flagellin added to the basolateral compartment.

In light of our data, we postulated that T84 cells express a receptor for an EAEC virulence factor, and that this receptor is not as abundantly expressed by non-polarized cells or polarized HT-29 cells. Alternatively, polarized T84 cells may possess a signal transduction capability that is not present in the other cell systems tested. Our data were not consistent with adherence alone accounting for the distinct responses of polarized T84 cells. In fact, EAEC strain 042 adheres more abundantly to HT-29 cells than to any epithelial cell line thus far tested, including T84 (not shown).

We also sought to compare the characteristics of the flagellar and non-flagellar proinflammatory responses. Though the responses were generally similar, and typical of acute inflammatory responses of epithelial cells in general, our data suggested that ICAM-1 and GRO-γ induction were more abundantly upregulated in response to flagellin compared with the non-flagellar factor. The implications of this observation with regard to the signal transduction cascades that could be involved are under investigation.

Interestingly, in spite of apparent transcript production of IL-1β and TNF-α from both polarized and non-polarized T84 cells, no protein production was detected with the CBA utilized in our studies. Furthermore, neither IL-6 transcript nor protein was upregulated in polarized T84 cells. The levels of protein production for these cytokines in response to EAEC infection in this model may be below the limits of detection, or may be produced at time points we did not assess. In spite of our data, such cytokines could conceivably be relevant clinically. Jung et al. did not detect IL-6 transcripts or protein, and detected very low levels of TNF-α protein, in response to infection of T84 cells with an invasive pathogen (Jung et al., 1995). However, these investigators demonstrated IL-6 production by freshly isolated IECs infected with invasive bacteria.

Our studies allowed us to draw additional inferences regarding the nature of the non-flagellar proinflammatory factor in EAEC. Our data clearly implicated plasmid-associated genes in the non-flagellar IL-8 response of polarized T84 cells; moreover, we showed that inactivation of AggR diminishes IL-8 production to levels similar to uninfected cells. Additionally, the non-adherent, flagellated commensal E. coli strain HS did not induce significant levels of IL-8 from polarized T84 cells. However, transformation of HS with the pAA2 plasmid caused a fourfold increase in IL-8 production. Our data also strongly implicate the AAF adhesin, which is under AggR control, as two different AAF mutants did not induce an IL-8 response. Fimbriae of several bacterial species, including E. coli, may cause cytokine responses from epithelial cells (Frendeus et al., 2001; Hedlund et al., 2001). In fact, the Afa/Dr adhesins of diffusely adherent E. coli are the fimbriae most closely related to the AAFs of EAEC (Elias et al., 1999), and strains that harbour Afa/Dr adhesins have been shown to induce transmigration of PMNs and IL-8 production mediated by MAPK signalling pathways (Betis et al., 2003). Earlier studies have suggested that a plasmid-associated gene such as aggR or aafA may be involved in the inflammatory response to EAEC (Huang et al., 2004; Steiner et al., 2000). However, these reports did not utilize polarized IECs.

Our data suggest that the component of the AAF fimbriae which induced IL-8 secretion is the AafB protein. The ability of AafB to induce IL-8 release in our hands is consistent with its reported ability to bind to β1-integrins (Plancon et al., 2003), an event which has been shown to stimulate IL-8 release via multiple pathways (Eitel et al., 2005; Sokolova et al., 2004). Studies to characterize the possible role of AafB and of AAF-dependent cytokine release are underway. We note that β1 integrins are predominantly located at the basolateral surface of epithelial cells (Muza-Moons et al., 2003), so involvement of these receptors in the IL-8 response of polarized cells and intact mucosa could require either opening of the tight junctions and/or transcytosis of AafB. We also note that 042aafB induced levels of IL-8 higher than uninfected cells or HS, suggesting that still further proinflammatory factors may await discovery.

A three-stage model for EAEC pathogenesis has been proposed, comprising (i) adherence to the intestinal mucosa; (ii) increased production and deposition of a mucus biofilm; and (iii) inflammation and cytokine release (Huang and Dupont, 2004). Given that prior studies have implicated AAF adhesins in adherence and biofilm formation (Sheikh et al., 2001), and that inflammation stimulates mucus secretion (Levine et al., 1995), our current data conflate the EAEC pathogenesis model around a central role for the AAF organelle, though other factors clearly contribute as well. Dissecting the contributions of known and still undiscovered EAEC virulence factors to the pathogenesis model is still urgently needed.

Experimental procedures

Bacterial strains and growth conditions

Strains used in epithelial cell infections are listed in Table 1. All bacteria were grown in static conditions overnight in Luria–Bertani (LB) broth with the addition of 50 µg ml−1 kanamycin or 100 µg ml−1 ampicillin to maintain plasmids when appropriate.

Intestinal epithelial cells

T84 cells (ATCC CCL-248) were routinely maintained in DMEM-F12 with 10% fetal bovine serum, 10 U ml−1 penicillin, and 10 µg ml−1 streptomycin. HT29-18-C1 (Huet et al., 1987) cells were kindly provided by Dr Cynthia Sears. These cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) containing 4.5 g l−1 glucose, l-glutamine, 110 mg l−1 sodium pyruvate, and pyridoxine hydrochloride and supplemented with 10% fetal bovine serum, 10 µg ml−1 apotransferrin (Sigma, St. Louis, MO), 10 U ml−1 penicillin and 10 µg ml−1 streptomycin. Tissue culture reagents were obtained from Invitrogen unless otherwise specified. For experiments with non-polarized cells, 6-well tissue culture plates were seeded at a density of 2 × 106 cells/well and used within 48–72 h, at which time confluency was approximately 70–90%. For experiments requiring polarized cell monolayers, 12-well polycarbonate Transwell filters with a 0.4 µm pore (Corning) were coated with 0.5 mg ml−1 rat tail collagen (Sigma) and seeded at 1 × 105 cells/filter. Transepithelial resistance was monitored with an EVOM-G Ohmeter (World Precision Instruments, Sarasota, FL) in an Endhom 12 chamber. Experiments were performed within 10–14 days of seeding when transepithelial electrical resistances (TEER) reached 300–400 Ω cm−2 for HT-29 cells and 1000–2000 Ω cm−2 for T84 cells.

Epithelial cell infections

Non-polarized cells.  Prior to infection, epithelial cells were washed three times in PBS, pH 7.4. For experiments with non-polarized cells PBS was replaced with 2 ml of tissue culture medium containing 1% methyl-α-d-mannopyranoside (Sigma) to block type 1 fimbriae, and lacking fetal bovine serum and antibiotics, followed by incubation at 37°C in 5% CO2 for up to 1 h. For T84 cell macroarray experiments, bacteria from an overnight LB-broth culture were added directly to the cells. To maximize fimbrial expression and biofilm formation (Sheikh et al., 2001) for all other infections, bacteria from overnight culture were pelleted by centrifugation, resuspended in DMEM with 4.5 g l−1 glucose and pyridoxine hydrochloride and appropriate antibiotics and diluted 1:20 in the same broth. Strains were grown to early logarithmic phase in static conditions and standardized to an OD600 of 0.300 ± 0.02 prior to infection. Bacterial inocula were verified by dilution plate counts. Thirty microlitres of bacterial suspension standardized to 2–4 × 108 colony-forming units (cfu) ml−1[multiplicity of infection (moi) ≈ 2] were added to the medium in each of three wells in a 6-well plate containing non-polarized T84 or HT-29 cells and incubated at 37°C in 5% CO2 for 3 h. Following incubation, cells were again washed three times with PBS prior to RNA extraction. For experiments in which culture supernatants were analysed for IL-8 by ELISA, supernatants were either collected at the 3 h time point or cells were washed and treated with medium containing 50 µg ml−1 gentamicin (Sigma). Culture supernatants were collected 24 h from the start of the infection. Supernatants were maintained at −20°C prior to analysis.

Polarized cells.  Prior to inoculation, polarized cell monolayers were washed three times with PBS. One ml of tissue culture medium plus 1% methyl-α-d-mannopyranoside was added to the basolateral chamber of a 12-well Transwell plate and 500 µl was added to the apical side, followed by incubation at 37°C in 5% CO2 for up to 1 h. Medium was aspirated from the apical compartment just prior to addition of 100 µl of 2–4 × 108 cfu ml−1 bacteria (moi ≈ 10) in tissue culture medium plus 1% methyl-α-d-mannopyranoside or medium alone for mock infections. Experiments to evaluate the effects of addition of flagellin to tissue culture cells were also performed. Purified FliC protein was added either apically or basolaterally at 1 µg ml−1. Plates were incubated at 37°C in 5% CO2 for 3 h. Cells were washed three times with PBS prior to RNA extraction. For assays measuring cytokine production in the basolateral compartment, medium was aspirated from the apical surface, the monolayers were washed three times with PBS and 500 µl of tissue culture medium with 1% methyl-α-d-mannopyranoside and 50 µg ml−1 gentamicin was added. Medium in the basolateral compartment was cultured on an LB plate to test sterility before treatment with 50 µg ml−1 gentamicin, but the basolateral medium was not exchanged. The plate was returned to the incubator. Basolateral medium was then collected at various time points for analysis.

Purification of FliC protein

The flagellin gene (fliC) cloned with a histidine tag translationally fused to the N-terminus in the expression vector pCR®T7/NT-TOPO® (Invitrogen) was kindly provided by Dr Ted Steiner (Steiner et al., 2000). This plasmid, termed pTSS8, was transformed into the host strain BL21(DE3)pLysS for flagellin expression and purification. Mid-exponential phase culture was induced with 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG) for 2 h. Following centrifugation, cell pellets were resuspended in Talon buffer (50 mM NaPO4 buffer, pH 7.0; 300 mM NaCl) containing complete mini EDTA-free protease inhibitor cocktail (Roche, Mannheim, Germany) and lysed with three freeze/thaw cycles and French pressure. Talon resin (1.5 ml) (Clontech, Palo Alto, CA), which is a metal affinity column that utilizes cobalt ions for purification of His-tagged proteins, was washed with 50 ml of Talon buffer before addition of the lysate. The lysate-bound resin was again washed with 50 ml of Talon buffer, followed by 50 ml 50 mM imidazole (Sigma) in Talon buffer. Protein was eluted in 1 ml fractions with 200 mM imidazole in Talon buffer. Fractions containing flagellin were identified by SDS-PAGE analysis and pooled. Purified flagellin was quantified with the bicinchoninic acid assay (Sigma). Further purification to remove LPS was not performed, because the IEC lines used in this study are not responsive to LPS in the absence of serum (Cario et al., 2000).

RNA extraction and RT-PCR

T84 and HT-29 cells from two or three wells were pooled into Trizol reagent (Invitrogen) for RNA extraction, which was performed according to manufacturer's instructions. DNA contamination was removed by digestion at 37°C for 30 min with 1 U DNase I per 10 µg of RNA (Ambion, Austin, TX). DNase I was inactivated by incubating at room temperature for 2 min with 0.1 volume of DNase Inactivation reagent (Ambion). Inactivation reagent was subsequently removed by centrifugation and RNA was precipitated. cDNA was generated from 1 µg of purified RNA using 50 ng of random hexamers and the ThermoScript™ RT-PCR System (Invitrogen). Each reaction consisted of 1× cDNA synthesis buffer, 0.05 M DTT, 40 U RNaseOUT, 1 mM each dNTP and 15 units of ThermoScript™ RT. As a negative control for each RNA sample, RT was omitted from a second reaction tube. The RT reaction included a 10 min 25°C primer-annealing period, followed by incubation at 50°C for 50 min with a final inactivation of enzyme at 85°C for 5 min. Finally, each reaction was incubated for 30 min at 37°C with 2 U of RNaseH. One microlitre of the RT reaction was used in a PCR reaction containing 1× PCR buffer (20 mM Tris-HCl, 10 mM KCl, 10 mM(NH4)2SO4, 2 mM MgSO4, 0.1% Triton X-100), 0.2 mM each dATP, dCTP, dGTP, dTTP (Fermentas), 0.2 µM sense and antisense primer (Table 4) and 1 U Taq DNA polymerase (New England Biolabs, Beverly, MA). To minimize amplification of contaminating DNA, primers were chosen in separate exons or spanning intron-exon boundaries with Primer-3 software (Whitehead Institute, Cambridge MA). A typical PCR reaction consisted of a 94°C denaturation for 2 min, followed by 25–35 cycles of 94°C for 30 s, 55 or 58°C for 30 s and 72°C for 30 s with a final 10 min 72°C extension. PCR products were separated on 2% agarose gels and visualized with ethidium bromide staining.

Table 4.  Primers used for RT-PCR of inflammatory mediators.
Gene nameGenBank mRNASequence
Glyceraldehyde phosphateNM0020465′-GAGTCAACGGATTTGGTCGT
Colony-stimulating factor 5′-GCCCTTGAGCTTGGTGAG
Intercellular adhesion molecule-1NM002015′-GGAGCTTCGTGTCCTGTATG
Integrin-beta 2NM0002115′-CTACGAGAAACTCACCGAG
Macrophage inflam. protein-3-αNM0045915′-CAATGCTATCATCTTTCACACA
Tumour necrosis factor-alphaNM0005945′-CCAGGCAGTCAGATCATCTTC

Human cytokine arrays

Cytokine array experiments were performed in duplicate with cDNA from T84 cells hybridized to v1.2 of the Panorama™ Human Cytokine Gene Array (Sigma-Genosys, The Woodlands, TX). This format consisted of 375 different cDNAs spotted in duplicate as PCR products onto charged nylon membranes. Labelling and hybridization were done according to manufacturer's instructions. Briefly, following a 2 min 90°C denaturation step, 2 µg of total RNA was annealed to a pool of array-specific primers (Sigma-Genosys) by gradually lowering the temperature to 42°C over 20 min. Twenty microCuries (α-33P) dCTP (∼2500 Ci mmol−1) (Amersham, Piscataway, NJ) were incorporated into the cDNA in a labelling reaction at 42°C for 3 h. In addition to primed RNA and dCTP-tagged nucleotide, each reaction contained 1× RT buffer, 333 µM each dATP, dTTP and GTP, 1.67 µM CTP, 20 U ribonuclease inhibitor (Sigma) and 50 U AMV RT. Unincorporated-radioactive nucleotides were removed with a Sephadex G-25 spin column. Denatured, labelled cDNA was hybridized to macroarrays overnight at 65°C in 5× SSPE, 2% SDS, 5× Denhardt's reagent (1% Ficoll, 1% polyvinylpyrrolidone, 1% bovine serum albumin) and 100 µg ml−1 salmon testes DNA. Three room temperature washes of 2–3 min each in 0.5× SSPE, 1% SDS, were followed by one 20 min 65°C wash in 0.5× SSPE, 1% SDS and one 20 min wash in 0.1× SSPE, 1% SDS. General-purpose phosphorimaging screens (Kodak, Rochester, NY) were exposed to hybridized membranes and scanned with a Storm Phosphorimaging System (Molecular Dynamics). Images were analysed with ImageQuant™ software (Molecular Dynamics). Spot intensity data were imported into a Microsoft Excel™ spreadsheet downloaded from the Sigma-Genosys website for analysis ( Background was calculated from two Tris-EDTA spots on the membrane and subtracted from each gene-specific signal. The average intensity of a group of housekeeping genes was used to normalize gene-specific signals. For each gene, the difference in intensity between two membranes was calculated and reported as fold difference. Between repeat hybridizations, membranes were stripped by boiling for 20 min in 10 mM Tris-HCl, pH 8.0, 1 mM EDTA, 1% SDS. Autoradiographic film was exposed overnight to stripped membranes to verify that the probe from the previous hybridization was completely removed.


Culture medium from triplicate wells of non-polarized cells or from the basolateral compartment of polarized monolayer dishes were evaluated in duplicate by ELISA for IL-8 as described (Zhou et al., 2003). Microtitre plates (Thermo Labsystems, Franklin, MA) were coated with 1 µg ml−1 anti-human IL-8 in carbonate buffer, pH 9.6 overnight at 4°C. After washing three times in PBS-containing 0.1% Tween-20 (PBST), plates were blocked with 5% skim milk in PBS for 1 h at 37°C. Plates were again washed with PBST and samples and serial dilutions of a recombinant human IL-8 standard were added. Plates were then incubated at room temperature for 2 h. Plates were washed three times in PBST, 0.25 µg ml−1 biotinylated mouse anti-human IL-8 was added, and plates were incubated at room temperature for 1 h. Avidin-horseradish peroxidase conjugate, diluted 1:2000, was added following another PBST wash step. Plates were incubated at room temperature for 30 min. Plates were washed with PBST and IL-8 was detected with TMB substrate solution. Optical densities were read at 450 nm with a 96-well plate reader (Thermo Labsystems). IL-8 concentrations were determined from a best-fit standard curve generated with Microsoft Excel™. All reagents and antibodies were purchased from BD PharMingen (San Diego, CA). The mean and standard deviation of triplicate wells were calculated.

Statistical analysis

IL-8 ELISA measurements were expressed as mean ± one standard deviation of the mean. Statistical significance was calculated using a one-way anova followed by multiple paired comparisons with Bonferroni error correction using the Analyse-it add-in for Microsoft Excel™ database software.


This work was supported by US PHS Grant AI33096 to J.P.N. We thank Ed Dudley for critical reading of the manuscript and helpful suggestions in the course of this work. We also acknowledge the many helpful discussions with Dr Ted Steiner during the course of this work.