After sequencing, a total of 181 sputum samples from 59 LTRs were included for subsequent analysis (Fig. 1). We divided the recipients into three groups according to the presence or absence of pulmonary infection and acute rejection at sampling: clinically stable (or event-free, n=47) recipients, recipients with infection (n=103) and recipients with rejection (n=31). The clinical characteristics of the LTRs are presented in Table 1.
Clinical characteristics, such as laboratory parameters, hospital stay, intensive care unit (ICU) stay, primary graft dysfunction (PGD) grade and pulmonary function, are regarded as factors that influence both clinical diagnosis and the airway microbiota. To assess the contribution of clinical variables to microbial community composition and clinical diagnosis, RDA was performed at the OTU level. The results showed that the event-free, infection and rejection groups could be distinguished from each other, and several clinical characteristics, such as hospital stay, ICU stay and serum PCT, were highly positively associated with the airway microbiota of infection recipients, while T lymphocytes were positively related to rejection recipients (Fig. S1).
The airway microbial community in LTRs with different transplant outcomes
First, we compared the airway microbial diversity among LTRs with different clinical diagnoses. Alpha diversity was significantly different between the event-free and rejection and between the infection and rejection groups (P<0.011, P=0.001, respectively), with the highest Shannon index in the rejection group (Fig. 2A). PCoA based on the unweighted UniFrac distance matrix showed a distinct beta diversity among the 3 transplant groups (P=0.001, R2=0.13, Fig. 2B), as well as between the event-free and rejection groups and between the infection and rejection groups (P=0.004, R2=0.03; P=0.001, R2=0.02; respectively, Fig. 2D, E). However, the event-free and rejection groups were not clearly separated in alpha (Shannon, P=0.136) and beta diversity (unweighted UniFrac, P=0.35, R2=0.01, Fig. 2C).
Additionally, a heat map of the 28 dominant genera (relative abundance >1% in at least one group) showed a different microbial profile among the 3 groups (Fig. 3A). A Venn plot was drawn at the family, genus and OTU levels with relative abundance >1% among the different groups (Fig. 3B, C, D). The results showed that 13 families, 13 genera and 8 OTUs were shared by the event-free, infection and rejection recipients. Several microbial taxa were unique to the 3 groups, including 1 family, 2 genera and 6 OTUs in the event-free group; 2 families, 3 genera and 8 OTUs in the infection group; and 1 family, 1 genus and 5 OTUs in the rejection group. This finding indicated that the 3 groups not only shared a common microbiota but also had their own unique taxa which may associated with the pathogenesis of both diseases.
Changes in the airway microbiota during infection and rejection of LTRs
The top 6 most abundant phyla of the airway microbiota detected in our study were Firmicutes, Proteobacteria, Actinobacteria, Bacteroidetes, Cyanobacteria and Fusobacteria. The 28 dominant genera with an average relative abundance greater than 1% in at least one group (mainly Streptococcus, Rothia, Enterococcus, Haemophillus and Granulicatella) accounted for up to 90% of the total genera. The microbial composition of the airway microbiota at the phylum level and the genus level was distinct among the event-free, infection and rejection recipients (Fig. 4A, B).
Moreover, LEfSe on all bacterial taxa was used to identify the different microbiota constituents among the 3 transplant groups (LDA score >2.0, Fig. S2). To more clearly explore microbial differences among the 3 groups, we selected the above 28 dominant genera to build the LEfSe plot, and a total of 10 distinguished genera were identified (Fig. 4C). Three genera, namely, Corynebacterium, unclassified Enterococcaceae and unclassified Lactobacillales were greatly enriched in the infection group, and 7 genera, namely, Rothia, Granulicatella, Neisseria, Actinomyces, Leptotrichia, Lactobacillus and unclassified Aerococcaceae, were significantly enriched in the rejection group, while no differentially abundant genera were found in clinically stable recipients. Table S1 compares the relative abundance and prevalence of the 10 genera in the 3 groups.
The prediction efficiency of the airway microbiota and clinical features for different clinical diagnoses of LTRs
Finally, we attempted to evaluate whether there were useful adjunctive indicators for the discrimination of different transplant groups. Generally, increases in serum PCT and peripheral blood T-lymphocyte levels are associated with infection and acute rejection after LT, respectively [34,35]. In our study, PCT and T-lymphocyte levels were relatively high in the recipients with infection and rejection, respectively, but the differences were not statistically significant (Fig. 5A, B). Therefore, random forest analysis was performed using individual airway microbiota constituents alone or in combination with clinical variables (PCT and T-lymphocyte levels) to investigate their prediction efficiencies in LTRs with different clinical diagnoses. First, the ROC curve was determined using the above 10 potential indicators identified by LEfSe, with AUC values of 0.697 (95%CI: 68.45-71.03%), 0.854 (95%CI: 84.13-86.63%) and 0.822 (95%CI: 81.08-83.28%) to distinguish between the event-free and infection, event-free and rejection, and infection and rejection groups, respectively (Fig. 5C). In comparison, the model was built based on the combination of the 10 genera and PCT and T lymphocyte levels. The results revealed an improved performance, and the corresponding AUCs reached 0.894 (95% CI: 88.54-90.34%), 0.955 (95% CI: 94.78-96.24%) and 0.913 (95% CI: 90.28-92.31%, Fig. 5D). These results indicated that the airway microbiota, especially combined with PCT and T lymphocyte levels, was a reliable indicator of infection and acute rejection in LTRs.