Patient characteristics and clinical outcomes
Patient characteristics and clinical outcomes are summarized in Table 1. The median age was 55 years (IQR 47–62.5), 71.4% (n = 15) of the patients were male, and the median ECMO duration was 9 days (IQR 6–14.5). There were no significant differences in patients’ characteristics between the two groups except for APACHE II scores (median 13 vs 10, P = 0.019) and the percentage of pneumonia (66.7% vs 20%, P = 0.040) being higher in the bacteremic group. There were no significant differences in clinical outcomes.
Table 1
Characteristic | Bacteremic (n = 6) | Non-bacteremic (n = 15) | P |
Age | 56.5 (46.3–62.5) | 55 (47–63) | 0.668 |
Male | 4 (66.7) | 11 (73.3) | 0.760 |
BMI | 25.5 (21.7–26.4) | 22.8 (20.4–27.9) | 0.697 |
SOFA | 14 (10.3–16.3) | 12 (10–15) | 0.531 |
APACHE II | 13 (11.8–18) | 10 (8–12) | 0.019 |
Vasopressor | 5 (83.3) | 11 (73.3) | 0.627 |
RRT | 2 (33.3) | 4 (26.7) | 0.760 |
Pre ECMO PF ratio | 62.5 (56.5–68.5) | 66 (57–81) | 0.640 |
ECMO duration (d) | 12 (7.8–23.5) | 9 (4–11) | 0.227 |
ICU day | 21.5 (15.1–48.0) | 42.0 (19.5–70.9) | 0.436 |
Pneumonia | 4 (66.7) | 3 (20) | 0.040 |
Antibiotic use | 6 (100) | 15 (100) | 1.000 |
ECMO complications | | |
Bleeding | 2 (33.3) | 1 (6.7) | 0.115 |
Thrombosis | 0 | 1 (6.7) | 0.517 |
Weaning success from ECMO | 5 (83.3) | 14 (93.3) | 0.481 |
Survival to discharge | 5 (83.3) | 14 (93.3) | 0.481 |
Data are presented as median (interquartile range) or n (%). |
BMI. body mass index; SOFA, sequential organ failure assessment; APACHE II, acute physiology and chronic health evaluation II; RRT, renal replacement therapy; ECMO, extracorporeal membrane oxygenation; PF ratio, PaO2/FiO2; ICU, intensive care unit |
Blood culture isolates
The most common organism isolated by blood culture within 3 days of catheter removal was S. epidermidis (4/6, 66.7%), followed by A. baumanii (1/6, 16.7%), and E.coli (1/6, 16.7%).
Diversity of bacterial community in ECMO
In total, 2,548,172 reads, with an average of 121,341 reads per sample were generated after initial quality filtering and chimera removal (Supplementary Table 1). Initially, we measured alpha diversity of the bacterial community in each group. The total bacterial diversity was estimated by Chao1 index. The evenness of microbiota was estimated by the Shannon index (Fig. 1A). Although it was slightly higher in the non-bacteremic group, there was no statistically significant difference by Wilcoxon-rank-sum-test.
In order to examine the microbial community variability between the two groups, we calculated beta diversity by decomposing microbiome variability onto major components using PCoA on Bray-Curtis dissimilarity. (Fig. 1B) There was no significant difference between the two groups.
Differences of microbiome composition according to the presence of bacteremia
Microbiota play an important role in the pathophysiology of many diseases. In order to identify microbiome profiles of the two groups, we examined the microbiota’s taxonomic composition and the relative abundance of bacteremia at different taxonomic levels. At the phylum level, we found a 47 different phyla, 11 of which were present in all samples. The major phyla were Proteobacteria, Actinobacteria, Bacteroidetes, Cyanobacteria, Verrucomicrobia, and Firmicutes, which were present, on average, in 10% or more of the samples. (Fig, 2A)
We found 367 different genera, 8 of which were present in all samples. Among them, the major genera were Limnohabitans, Flavobacterium, Delftia, Massilia, Bacillus, Candidatus, Xiphinematobacter, and CL0-1 which were present, on average, in 1% or more of the samples. Delftia was more abundant in the non-bacteremic group; however, Bacillus, Flavobacterium, CL0-1, Candidatus, and Xiphinematobacter were more abundance in the bacteremic group (Fig. 2B)
Difference in dominant microbiota between the groups
We used LEfSe to determine the taxa that most likely explain differences between the two groups. In this study, a P value < 0.05 and LDA score > 2.5 were considered to be significant. Differentially abundant taxa and their predominant bacteria are shown in Fig. 3. Significant differences were found between the two groups. LEfSe analysis revealed 10 discriminative genera, all of which were more abundant in the bacteremic group (Fig, 3A): Arthrobacter, SMB53, Neisseria, Candidatus, Ortrobactrum, Candidatus, Rhabdochlamydia, Deefgae, Dyella, Paracoccus, and Pedobacter. (Fig. 3B)
Network analysis of the ECMO microbiome
In order to find a relationship between taxa associated with the bacteremic state, we conducted a network analysis at the genus level. Compared to patients without bacteremia, the bacteremic group was very complex. In non-bacteremic patients, only genera associated with the hub were correlated (Fig. 4A). However, the microbiome of bacteremic patients showed a high correlation with genera that were not associated with the hub (Fig. 4B). Although not a hub, Flavobacterium and CL0.1, which were abundant in the bacteremic group, were considered important genera because they connected different subnetworks.
Metabolic prediction
We explored if metabolic pathways in the bacteremia state were related to the metabolic differences found in the two groups. We performed a comparative prediction analysis of the functional metagenome using PICRUSt. Among the 328 affiliated KEGG pathways, only 1 was shown to be statistically significant with P < 0.05 and LDA > 2. Notably, a significant elevation in the secretion system was found in the non-bacteremic group (Fig. 5)