All patients admitted to the ICU during a 26-week period who needed intubation and mechanical ventilation were included. Patients were evaluated at admission by the Acute Physiology and Chronic Health Evaluation II score (20). At admission and every week thereafter, temperature, leukocytes and platelet counts, hepatic enzymes, and presence of SIRS and ventilator-associated pneumonia (VAP) were recorded. Serum samples were obtained from patients at admission, every 7 days afterwards, and at discharge. BAL was obtained at admission immediately after intubation by using protected mini-bronchoalveolar lavage (Combicath, Plastimed, Le-Plessy-Bouchard, France), then performed when a lung infiltrate suggestive of pneumonia appeared, and repeated every week until pneumonia resolved. BALs are part of the routine diagnosis and follow-up of pneumonia in the ICU and were not performed specifically for the study. Informed consent was obtained from the patient’s family, according to French legislation. Data about bacteria isolated from blood cultures, lung secretions, and urine by conventional procedures were recorded. When isolated from urine and lung secretions, bacteria were only considered as pathogenic when concentration in the specimen was >10⁵ and >10⁶ CFU/mL, respectively (21).

Definitions of VAP and SIRS were based on previously published criteria (22–25), but to increase specificity, we limited our study to severe cases of VAP by adding strict criteria (Table 1). Taps and ice machine water were sampled every other week, as previously reported (11). The procedure for isolating bacteria from water and lung secretions by using cocultivation with Acanthamoeba polyphagia followed by subculture onto BCYE agar plates has been detailed elsewhere (11,26). Bacteria were identified by using 16S rRNA gene sequence comparisons as previously described (27). Legionella species were identified by using mip gene amplification and sequencing (28).

Twenty bacterial antigens were tested by microimmunofluorescence as previously reported (14,29). Bacterial species isolated from the ICU water during the studied period and 10 other species previously isolated in the same conditions from other sites (Bosea eneae, B. vestrisii, B. thiooxydans, Mesorhizobium amorphae, Azorhizobium caulinodans, Afipia felis, A. felis genospecies A, A. clevelandensis, A. birgiae, and A. massiliae [11–14]) were tested. The serologic tests were performed on the first serum samples from all patients admitted to the ICU; these patients were available for a second sample (30 patients). The control group comprised 10 patients in the same ICU. These patients had shorter stays and had samples taken at admission but was not available (100 patients) and 114 patients with other diseases, including Q fever (10 samples), trench fever (5 samples), tularemia (8 samples), Mediterranean spotted fever (10 samples), epidemic typhus (5 samples), syphilis (10 samples), cat scratch disease (5 samples), pneumonia caused by Chlamydia pneumoniae (5 samples), C. psittaci (5 samples), Mycoplasma pneumoniae (10 samples), L. pneumophila (10 samples), hepatitis C virus (5 samples), infections caused by cytomegalovirus (5 samples), Epstein Barr virus (11 samples), and HIV (10 samples). Serum samples were diluted at 1:25, 1:50, and 1:100 for immunoglobulin (Ig) G and IgM determination. The cutoff titer for a positive detection was determined as the lowest titer for which all first serum samples and control serum samples were negative. Then, all patients’ serum samples were tested and serial twofold dilutions from 1:50 to 1:1600 were made on these samples with titers at least equal to the cutoff titer. Tests by Western blot were performed as previously described (29) for any patient with a seroconversion.

DNA was extracted from BAL samples by using QIAMP Tissue kit (QIAGEN, Hilden, Germany), according to the manufacturer’s instructions. PCR detection was attempted on all BAL samples for bacteria against which at least one seroconversion was generated. DNA extraction from serum samples was performed in a laboratory other than the one in which the isolates were identified to avoid vertical contamination from previous amplifications, and no positive control was used to avoid horizontal contamination from the same experiment (16). We used a nested PCR that incorporated two primer pairs used only once (Table 2) followed by sequencing and comparison to the targeted sequence, as previously described for suicide-PCR (15,16). All samples were tested the same day in the same assay. Sample testing was blinded, and positive amplicons were sequenced. Negative controls consisted of BAL samples from 10 patients of the same ICU obtained at admission, BAL samples from 200 patients with nosocomial pneumonia hospitalized in other medical centers of the city, water samples, and a suspension of A. polyphagia. At least one negative control was used for every two serum samples.

Comparisons of demographic, clinical, and laboratory data between patients with evidence of amoeba-associated bacteria contact (seroconversion or positive detection in BAL samples) were performed by using chi square and Mann-Whitney tests, respectively. The tested variables were age; sex; an underlying disease; severity score; intubation duration; hospitalization duration; SIRS; death; increase of hepatic enzymes, platelet count, and leukocytes; VAP; and fever. We also compared demographic, clinical, and laboratory data between patients with and without unexplained VAP, fever, and SIRS. Multivariate analysis adjusted for age, sex, prolonged intubation, and an underlying disease was performed to confirm observed associations. STATA software (v. 7.0, Stata Corporation, College Station, TX) was used for analysis.

Ten species (310 isolates) were identified from 864 water samples (Table 3). B. massiliensis and L. anisa were the two most commonly isolated species (62.3% of the 310 isolates). In the ICU admission rooms , isolation of L. anisa ranged from 75% to 100% of the tested taps during the first 20 weeks; whereas all were negative during the last 6 weeks, after taps were changed. No amoeba-associated bacteria were isolated from BAL.

Ninety serum samples from the 30 patients and the 214 control serum samples were tested by an immunofluorescence assay for IgG and IgM on the 20 amoeba-associated bacteria antigens (12,160 tests). The 30 first serum samples and the samples from blood donors did not have an IgG titer of >1:50 or IgM titer >1:25. Cutoff titers for positive serologic tests with 100% specificity are shown in Table 4. Twelve (40%) patients seroconverted from 10 to 35 days after admission to at least one antigen; 10 showed IgM antibodies (Table 5). Five patients seroconverted to L. anisa, six to B. massiliensis, including one to both, one to L. quinlivanii, and one to A. broomeae. Western blots confirmed the seroconversions, with the appearance of several reacting bands on convalescent-phase serum samples (Figure). Patients also seroconverted to amoeba-associated bacteria not detected in the water in this study: two to A. clevelandensis (patient 2, IgG=1:100 and IgM=1:100; patient 8, IgG=1:50 and IgM=1:200;) and two to A. felis (patient 2, IgG=1:100 and IgM=1:100; patient 9, IgG=1:1600 and IgM=1:25). Seroconversions were significantly more frequent against amoeba-associated bacteria obtained in this ICU than against amoeba-associated bacteria isolated in previous studies: 13 of 300 tests versus 4 of 300 tests and 12 of 30 patients versus 4 of 30 patients (p=0.046 and p=0.039, respectively). Patients also seroconverted more frequently to the most commonly isolated bacteria, L. anisa and B. massiliensis (>50 isolates, p=0.021).

Table 6 shows the clinical characteristics of patients with serologic evidence of exposure to amoeba-associated bacteria isolated in the ICU. Analysis for risk and potential confounding factors did not indicate differences between patients with or without seroconversion. However, seroconversion was statistically associated with VAP (p=0.026), unexplained VAP (p=0.034), SIRS (p=0.024), and unexplained SIRS (p=0.045). Multivariable logistic regression demonstrated that seroconversion was independently associated with VAP even after we adjusted for intubation duration, hospitalization duration, number of serum samples, and underlying disease (p=0.014 to 0.030).

The DNA of L. anisa and B. massiliensis were each detected once in the 66 BAL samples from 30 patients. For two samples, seroconversion to the identified bacteria, L. anisa and B. massiliensis, respectively, was observed 4 and 2 weeks after the PCR was positive in BAL. None of the 210 control patients was positive for these bacteria in BAL samples compared to 2 of 30 patients in the ICU (p<0.01).

The death rate was 33.3%; disease was mainly associated with fever (96.6%), VAP (56.6%), and SIRS (76.6%). No microbial etiologic agent was found in 18 (62%) of 29 patients with fever, 10 (43%) of 23 patients with SIRS, and in 8 (47%) of 17 patients with VAP. VAP was significantly associated with duration of hospitalization (median hospitalization days/interquartile range of 23/18–41 with VAP versus 14/10–21 without VAP, p=0.04), SIRS (16/17 with VAP versus 7/13 without VAP, p=0.025) and seroconversion to amoeba-associated bacteria (10/17 with VAP versus 2/13 without VAP, p=0.026). No other statistical difference was observed between patients with VAP and without VAP in terms of demographic, clinical, and paraclinical data, risks, and potential confounding factors. Most patients received cephalothin for the first 3 days of hospitalization as an antibiotic prophylaxis. The most commonly used antibiotics in patients with VAP were third-generation cephalosporins. Patients with VAP received antibiotics more frequently and for more than 1 week (16/17 with VAP versus 7/13 without VAP, p=0.025).

We confirmed that several amoeba-associated bacteria are common in the water in ICU; we recovered 310 amoeba-associated bacteria isolates from 10 species from the water of the ICU. As exposure to a microorganism is a prerequisite to infectious disease, we first evaluated contact of patients hospitalized in the ICU with amoeba-associated bacteria in water. We found that 12 (40%) of 30 of patients seroconverted to amoeba-associated bacteria and that these seroconversions were significantly more common against local isolates than to amoeba-associated bacteria from other ICUs. Moreover, antibody response parallels that of water contamination with B. massiliensis and L. anisa, which cause 83% of seroconversions identified in 62% of isolates (p=0.021). The cutoff titers, chosen to have 100% specificity, and the detection of several reactive antigens in the Western blots (Figure) suggest that these seroconversions represent specific serologic reactions. Detection of antibodies reacting to amoeba antigens would also be important because infections could occur through inhalation (perhaps after colonization) of infected amoebae acting as a “Trojan horse” (6). In future studies, isolation of amoebae in the aquatic environment as in the BAL samples of patients will be performed for further use as antigens.

Patient exposure to amoeba-associated bacteria from the ICU was also evident in results of amoeba-associated bacteria DNA detection in BAL samples from 2 of 30 patients as compared with 0 of 210 control patients (p<0.01). Moreover, for these patients, bacterial DNA was detected in the BAL sample before the seroconversion to the same bacteria 2 to 4 weeks after admission. This rate is compatible with an acute infection occurring during hospitalization rather than colonization of the respiratory tracts of patients. Contamination is unlikely when using a suicide-PCR procedure (15,16). Isolation of these agents was probably hampered by the antibiotic prophylaxis instituted at admission in this ICU for trauma patients and likely explains the lack of amoeba-associated bacteria isolation in BAL samples.

Among the 30 patients, VAP occurred frequently (56.6%) and was associated with hospitalization duration, as previously reported (30,31). Of the patients with VAP, 58.8% seroconverted to amoeba-associated bacteria, as compared to 15.4% of the remaining 13 patients (p=0.026). In patients with seroconversion, SIRS was also more prevalent (p=0.024). The percentages of unexplained VAP and unexplained SIRS were four times more common in amoeba-associated bacteria seroconverters than in nonseroconverters and remained statistically significant in spite of the small population studied (Table 6). Thus, amoeba-associated bacteria may be a common cause of unexplained VAP and SIRS.

Finally, we identified a cryptic outbreak in the ICU caused by L. anisa, a pathogen commonly encountered in the environment (32,33), only implicated in a few epidemics of Pontiac fever (34,35) and four cases of legionellosis (26,36–38). Five (16.7%) of 30 patients were infected with this bacterium, considered a relatively rare pathogen. Serologic tests are not currently used for L. anisa, and no urinary antigen test is available. Therefore, diagnosis based only on isolation may explain why L. anisa is so rarely reported. However, its absence in the BAL samples from the 200 patients from other ICUs shows that local epidemiology plays a major role. This study confirms that L. anisa is common in the environment (32,33).

Members of the Bosea genus are gram-negative, oxidase-positive, catalase-positive rods belonging to the α-2 subgroup of Proteobacteria. All are motile. They grow well on BCYE agar from 25°C to 37°C but do not grow or grow weakly on Columbia agar with 5% sheep blood. Colonies are smooth, mucoid, round, and cream colored and are urease positive and α-hemolytic on Columbia agar with 5% sheep blood and 0.2% yeast extract. B. massilensis are negative in assays for arginine dihydrolase activity, esculin and gelatin hydrolysis, β-galactosidase activity, maltose assimilation, and acid production by fermentation or oxidation of substrates tested in API 50 CH (Biomérieux, Marcy l’étoile, France), especially D-glucose, D-fructose, D-mannose, and sucrose. The species of the Bosea genus have high MICs to penicillin and amoxicillin and low MICs to doxycycline (13). In co-culture with A. polyphagia, B. vestrisii, B. eneae, and B. massiliensis are phagocyted and form progressively large vacuoles that lead to amoebal lysis; however, they have never been reported as pathogenic agents before. As B. massiliensis has not been tentatively isolated elsewhere, whether our findings reflect a local phenomenon or whether the bacterium is widely encountered is not known. We think that our data support the role of B. massiliensis in severe VAP, but confirmation is needed to definitely establish a role.

Our study indicates that most patients with VAP received β-lactam agents, mainly amoxicillin-clavulanic acid and third-generation cephalosporin. These antibiotics may have inhibited bacterial culture, which explains why no amoeba-associated bacteria were isolated from BAL samples. B. massiliensis has low MICs (<0.5 mg/L) to ceftriaxone, doxycycline, rifampin, and erythromycin (13). However, β-lactam agents are also active in vitro on Legionella spp., but animal models and clinical studies have demonstrated their inefficacy in the treatment of Legionellosis (39).

The results of our study confirm that the bacteriologic tests of hospital water supplies is largely ignored. Our work demonstrates that patients are exposed specifically to the most common water amoeba–associated bacteria in their environment, as evidenced by seroconversion against these bacteria and DNA of these bacteria in BAL samples. The route of infection, even if caused by aerosols generated in the ICU, remains unclear. Patients of this ICU sometime receive water through nasogastric tubes but only bottled, sterile water. However, we cannot exclude mistakes caused by not following recommended procedures. Patients for whom exposure to these bacteria is supported by seroconversion or DNA detection in BAL samples have unexplained VAP and SIRS more commonly. We speculate that amoeba-associated bacteria in the environment of intubated patients may concurrently cause unexplained infections and cryptic outbreaks. Research of new etiologic agents of pneumonia in ICUs should be based on environmental study of each ICU since ecologic findings of amoeba-associated bacteria in water points in hospital vary.