We began our study by comparing sampling techniques. To do this, the bacterial species compositions of samples obtained with blind mini-BAL were compared to those obtained with EAs from two patients with hospital-acquired P. aeruginosa by using 16S rRNA clone library sequencing. While there were distinct differences in community compositions between these two patients, the community compositions of the EA and BAL samples from each individual were highly similar (Fig. ). Therefore, in the remainder of this study, EAs were used due to the simplicity and cost-effectiveness of this less-invasive sample collection method.
FIG. 1. (A and B) Comparisons of 16S rRNA clone libraries of EA (white bars) and BAL (black bars) patient samples. (A) A total of 173 clones were sequenced from the EA samples, and 153 clones were sequenced from the BAL samples. (B) A total of 154 clones were (more ...)
Next, as a control for this study, EA samples were collected from three healthy individuals who had been briefly intubated for elective surgery. No 16S rRNA PCR product was detected from these patients (data not shown) under the conditions that readily yielded 16S rRNA amplicons in study patients, confirming that the normal lung is sterile and that our techniques identify organisms present only after colonization of the endotracheal tube and airway has occurred.
We next collected EAs from seven patients colonized by Pseudomonas aeruginosa
. Patient age, sex, timing of EA sampling, periods of antibiotic administration, and antibiotic sensitivity details are presented in Table . All patient samples yielded a 16S rRNA PCR product, and a total of 3,278 nonchimeric 16S rRNA sequences from patient-derived 16S libraries were subjected to phylogenetic analysis. Almost all organisms detected by cloning and sequencing were from five bacterial phyla, the Firmicutes
, and Fusobacteria
(Fig. ). Over half (55%) of the sequences obtained were from Pseudomonas aeruginosa
, followed by Stenotrophomonas maltophilia
spp. (5.8%), Acinetobacter
spp. (5.7%), Serratia marcescens
spp. (3.8%), Neisseria
spp. (3.3%), Mycoplasma
spp. (2.4%), and Streptococcus
spp. (2.3%). An additional18 genera were detected but together represented less than 7% of all clones sequenced. Of these less-abundant species, many are known oral, nasal, and gastrointestinal tract inhabitants, e.g., Porphyromonas
, and Veillonella
). These results support the hypothesis that oral, nasal, and gastrointestinal tract microbiota are the major reservoirs for bacteria that colonize the lower airway in intubated patients (17
Patient information and antimicrobial treatment
We next evaluated the bacterial diversity in five patients (Table , patients 1 to 5) for whom an initial sample was obtained within 24 h of parenteral antibiotic administration and a second sample was obtained 4 to 10 days later. Analysis of microbial diversity in the 16S clone libraries demonstrated a substantial reduction in bacterial diversity during antibiotic administration (Fig. ). The mean number of bacterial species identified fell from 16.2 to 5.6 (P < 0.05), and Shannon's diversity index, a commonly used measure of microbial diversity, fell from 1.48 to 0.59 (P < 0.05). In each case, antibiotic therapy led to selection of a pathogenic species that dominated the community. In four of five patients, P. aeruginosa came to dominate the microbial community at the second time point, despite the administration of antibiotics to which it was susceptible in vitro (Table ). In the fifth patient, the community became dominated by another pathogen, Klebsiella pneumoniae. Even when P. aeruginosa constituted the majority of the population at the initial time point, as in patient 3, who was chronically ventilated, there was initial diversity in the remainder of the community that was lost with antibiotic therapy.
FIG. 2. Temporal changes in bacterial diversity. Each bar represents the color-coded relative abundance of bacteria in a single EA. Numbers above the horizontal bars represent individual patients, and the numbers of clones analyzed for each sample are indicated (more ...)
To investigate how the bacterial community changed over a longer period of antibiotic treatment, we obtained additional samples at later time points for two patients (patients 2 and 5), and during prolonged antibiotic therapy in two additional patients (patients 6 and 7). All of these samples showed reduced diversity compared with the initial time point, and P. aeruginosa was the predominant species in six of seven samples, suggesting that once this organism is established as the dominant species, microbial diversity is slow to recover. Patient 2 did demonstrate a reduction in the relative abundance of P. aeruginosa following adjustment of antipseudomonal therapy 11 days after colonization, and, importantly, this was accompanied by substantial recovery of microbial diversity. This patient improved clinically, and extubation was attempted. Collectively, these data demonstrate that the loss of bacterial diversity and Pseudomonas dominance are highly correlated during antipseudomonal therapy.
Dominance of bacterial communities by one or a few species may result either from overgrowth of the dominant species or from loss of the nondominant species. Due to the limited number of clones that can feasibly be sampled from clone libraries, highly abundant species may mask the presence of less-abundant but clinically significant species. To determine whether the decline in diversity observed in clone libraries was a true reflection of the bacterial community in these patients, bacterial diversity was also analyzed with high-density microarrays (PhyloChips), which have enhanced sensitivity for low-abundance species compared with cloning (7
). For four patients, the same 16S rRNA gene amplicon pools from which clone libraries were prepared were subsequently hybridized to PhyloChips, and the bacterial communities were compared (Fig. ). While the microarray approach detected >30 times more bacterial types than did clone library sequencing (P
< 0.0001), the PhyloChip data clearly mirrored the loss of diversity found in the clone library data, with the mean number of observed phylotypes falling from 517 to 280 during antibiotic treatment (P
< 0.05). These data clearly show that the loss of diversity demonstrated in clone libraries is not an artifact of sampling and confirm that antibiotic treatment leads to lower diversity.
FIG. 3. Comparison of PhyloChip and clone library monitoring of bacterial diversity (phylotype numbers) over time for four patients. Closed symbols show numbers of bacterial phylotypes detected by PhyloChip analysis; open symbols show numbers of bacterial phylotypes (more ...)
Due to the great sensitivity of the PhyloChip, entire bacterial community responses can be monitored by this technique. Figure illustrates temporal changes in the fluorescence intensity of bacteria detected by PhyloChip. Bacteria demonstrating large changes in intensity between time points are labeled, and these correspond well to the dominant bacterial species detected by clone library analysis. The dominance of a few species within a community analyzed by 16S rRNA clone library clearly reflects the limited number of clones sequenced. The PhyloChip readily demonstrated that many bacteria present at the initial sampling point were indeed still present in the subsequent sampling period (Fig. ). Conversely, bacteria such as Klebsiella that became dominant in later clone library samples were detected in the initial sample by PhyloChip but were not detected in the corresponding clone library. This underscores the potential for low-abundance species to eventually dominate bacterial communities during the course of antimicrobial administration.
FIG. 4. PhyloChip analysis of complete bacterial communities over time in EAs. (A) Bacteria are ordered alphabetically from left to right according to taxonomic affiliation. Bars above the zero line represent bacteria that increased in abundance relative to the (more ...)
The microarray analysis also allowed us to correlate changes in phylogenetic groups over time when these groups of bacteria respond in similar manners to antibiotic administration. For example the γ-Proteobacteria (which include Pseudomonas aeruginosa) generally exhibit an inverse relationship in abundance with bacteria in the phylum Actinobacteria and in the class Bacilli (which includes Lactobacillus, Streptococcus, Staphylococcus, and Enterococcus) in all patients examined (Fig. ). Similarly Haemophilus and Pseudomonas also demonstrated an inverse relationship. The reciprocal changes in these subgroups within the bacterial community suggest they may be competing for similar niches in the endotracheal environment or otherwise influencing each other's growth.