The Lung Allograft Microbiome Associates with Pepsin, Inflammation, and Primary Graft Dysfunction

Abstract

Rationale

Primary graft dysfunction (PGD) is the principal cause of early morbidity and mortality after lung transplantation. The lung microbiome has been implicated in later transplantation outcomes but has not been investigated in PGD.

Objectives

To define the peritransplant bacterial lung microbiome and relationship to host response and PGD.

Methods

This was a single-center prospective cohort study. Airway lavage samples from donor lungs before organ procurement and recipient allografts immediately after implantation underwent bacterial 16S ribosomal ribonucleic acid gene sequencing. Recipient allograft samples were analyzed for cytokines by multiplex array and pepsin by ELISA.

Measurements and Main Results

We enrolled 139 transplant subjects and obtained donor lung (n = 109) and recipient allograft (n = 136) samples. Severe PGD (persistent grade 3) developed in 15 subjects over the first 72 hours, and 40 remained without PGD (persistent grade 0). The microbiome of donor lungs differed from healthy lungs, and recipient allograft microbiomes differed from donor lungs. Development of severe PGD was associated with enrichment in the immediate postimplantation lung of oropharyngeal anaerobic taxa, particularly Prevotella. Elevated pepsin, a gastric biomarker, and a hyperinflammatory cytokine profile were present in recipient allografts in severe PGD and strongly correlated with microbiome composition. Together, immediate postimplantation allograft Prevotella/Streptococcus ratio, pepsin, and indicator cytokines were associated with development of severe PGD during the 72-hour post-transplantation period (area under the curve = 0.81).

Conclusions

Lung allografts that develop PGD have a microbiome enriched in anaerobic oropharyngeal taxa, elevated gastric pepsin, and hyperinflammatory phenotype. These findings suggest a possible role for peritransplant aspiration in PGD, a potentially actionable mechanism that warrants further investigation.

At a Glance Commentary

Scientific Knowledge on the Subject

Primary graft dysfunction (PGD) after lung transplantation leads to increased morbidity and mortality. The role of the lung microbiome in this disease process is not known.

What This Study Adds to the Field

We found that a lung microbiome enriched in anaerobes, particularly Prevotella, was associated with PGD, elevated pepsin, and increased inflammation. These findings highlight a role of host–microbe interactions and the oral–lung axis in PGD.

Primary graft dysfunction (PGD) is acute lung injury within 3 days of lung transplantation that is not due to infection or other specific injury (1). PGD occurs in approximately 15–20% of patients undergoing transplant, is an important cause of short-term morbidity and mortality (2–4), and is a risk factor for later chronic lung allograft dysfunction (CLAD) (5). Although the pathogenesis of PGD is incompletely understood, it involves ischemia–reperfusion injury leading to endothelial activation, monocyte and neutrophil recruitment, and tissue damage (6). Recent studies have also linked donor nonclassical and alveolar macrophage activation to injury and neutrophil recruitment (7). Although donor, recipient, and operative clinical risk factors have been linked to PGD, it is not understood why certain allografts are primed for injury (3).

We now know that the healthy lung microbiome closely matches the upper respiratory tract and results from balanced microaspiration, mechanical and immune clearance, and limited local replication (8, 9). The lung microbiome composition has been shown to influence immune tone (10, 11), and perturbations of the normal lung microbiome (dysbiosis) have been described in many lung diseases (12). Several studies have characterized the post-transplant lung bacterial microbiome, mainly focused on CLAD, and typically found elevated concentrations of bacterial DNA and outgrowth of taxa relative to the upper respiratory tract (13). Several features of the lung microbiome correlated with CLAD incidence, including bacterial biomass and various taxa (14–17), although findings are not consistent across studies. In contrast, little is known about the lung microbiome in donor lungs or allografts early after transplant or relationship to early outcomes, including PGD.

We conducted a prospective cohort study investigating the bacterial microbiome in lung transplantation, sampling the lung in the donor before procurement and the allograft immediately after implantation. We found a lung microbiome characterized by anaerobic taxa immediately after transplantation was associated with severe PGD. Furthermore, both the microbiome profile and severe PGD outcome correlated with concentrations of pepsin in the lung and dysregulation of host cytokines and chemokines. Our study identifies novel associations between the microbiome, pepsin, inflammation, and PGD and suggests a role for oropharyngeal admixture in risk of acute lung injury after transplantation. Some of the results of these studies have been previously reported in the form of an abstract (18).

Methods

Patients and Sample Collection

Patients enrolled in the Lung Transplant Outcomes Group at Penn Medicine between 2014 and 2017 and listed for transplantation were approached for enrollment in this study. The study was approved by the University of Pennsylvania Institutional Review Board (protocols #817513 and #806468). Donor lung specimens were collected in the operating room immediately before lung procurement, and donor proxies were consented for enrollment. The bronchoscope was wedged in the airway (typically right middle lobe), approximately 20 ml of lavage saline was instilled and withdrawn, and 1 ml was available for these analyses. Recipient allograft samples were collected in the operating room within 60 minutes after implantation and organ reperfusion using a similar technique, and approximately 2 ml were available for analysis. Samples were placed on ice immediately after collection and stored at −80°C until analysis. Demographic and clinical data, including daily PGD scores (19), were collected prospectively. Healthy control lung specimens were collected previously by BAL from nonintubated healthy volunteers as described (8) and were resequenced here. Environmental controls consisted of bronchoscope prewashes, lavage saline, and water as previously described (20).

Microbiome Analysis

Samples were pelleted, total DNA extracted, 16S ribosomal ribonucleic acid (rRNA) gene PCR amplified with V1-V2 primers, and Illumina sequencing performed (20, 21), with details in the online supplement. The QIIME2 pipeline (22) was used for demultiplexing and Deblur (23) used for filtering and creation of amplicon sequence variants (ASVs). Taxonomy assignment, respiration phenotype, other aspects of the analytical pipeline, and code availability are detailed in the online supplement. Sequences are deposited in the National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) under PRJNA788785.

Bacterial biomass was estimated using post-16S PCR amplicon PicoGreen quantity (21), which we empirically validated against 16S quantitative PCR using a subset of 103 samples (rho = 0.77; P < 0.001).

Soluble Factors

Cytokine analysis used the Luminex bead-array platform (Millipore #HCYTMAG60PMX41BK). Pepsin was measured by ELISA (Cloud-Clone Corp.). Cytokine data were log10-transformed for statistical analysis.

Statistical Analysis

Microbiome composition was analyzed by principal coordinate analysis (PCoA) using the weighted UniFrac distances, which compares communities (β-diversity) based on phylogenetic relationship of constituent taxa (24). Permutational multivariate analysis of variance (PERMANOVA) with 999 permutations compared group differences in UniFrac distances, and the first two principal coordinate axes (PC1, PC2) were used for testing associations. Cytokines were analyzed by principal component analysis, and the first two principal component axes were used for testing associations. Statistical testing used nonparametric Wilcoxon rank sum for group comparisons and Spearman’s method for correlations. Hypothesis testing was two-tailed and considered significant at P < 0.05, and multiple testing correction was performed using the Holm method unless otherwise noted. Logistic regression performance was tested with receiver operating characteristic curves and area under the curve (AUC) using the pROC package. Additional statistical tests are detailed in the online supplement.

Results

Subjects, Samples, and Outcomes

Table 1 summarizes the cohort. PGD was assessed at Days 0, 1, 2, and 3 based on severity of hypoxemia (Pa_O2/Fi_O2 ratio) and radiologic infiltrate and graded 0–3 using established criteria (19) (see Figure E1 in the online supplement). Among the 139 recipients, 40 experienced no PGD (persistent PGD-0), 15 experienced severe sustained PGD at all four time points assessed (persistent PGD-3), and the remainder exhibited intermediate grades of PGD over the 72 hours after transplantation.

Table 1.

Clinical Characteristics of the Cohort

	PGD Grade 0 All Days (n = 40)	Intermediate (n = 82)	PGD Grade 3 All Days (n = 15)	P Value
Age, yr, median (IQR)	57 (45.5–64)	61 (55–65)	58 (55.2–64.8)
Sex, female, %	40	49	40	0.608
Bilateral transplant, %	85	62	80	0.022
Diagnosis, %				0.150
ILD	43	52	60
COPD	35	32	13
Suppurative	23	10	20
Other	0	6	7
ECMO bridge, %	5	2	7	0.198
Recipient antibiotics, %	45	43	29	0.612
Recipient immunosuppression, %	3	19	14	0.035
Total ischemic time, min, median (IQR)	583 (468–655)	461 (262–641)	518 (332–630)	0.119
Intraoperative bypass, %	12.5	18	33	0.171
Intraoperative PASP, mm Hg, median (IQR)	41 (33.5–47.2)	41 (35–49)	38 (32–48)	0.761
Alive on D/C, %	97.5	100	87	0.010

Definition of abbreviations: COPD = chronic obstructive pulmonary disease; D/C = discharge; ECMO = extracorporeal membrane oxygenation; ILD = interstitial lung disease; IQR = interquartile range; PASP = pulmonary artery systolic pressure; PGD = primary graft dysfunction.

We analyzed samples taken from the donor lung by bronchoscopy in the operating room immediately before organ procurement (donor lung) and from the same organ immediately after implantation and reperfusion (recipient allograft). Bacterial 16S rRNA gene amplicons were generated from 257 lung samples from 139 transplants (109 donor lung and 136 recipient allograft samples), 12 healthy controls (13), 120 bronchoscope prewashes, and 61 environmental controls. Sequences were clustered into ASVs at 100% identity. After filtering and taxonomic assignment as described in Methods and the online supplement, this yielded 6,169 ASVs representing 436 unique taxa, of which 799 ASVs (109 unique taxa) had >1,000 hits across samples. Figure E2 compares samples to background controls.

The Donor Lung Microbiome

Donors all had no known lung disease, were mechanically ventilated (median, 4 d; interquartile range_25–75, 3–5 d), and experienced brain death. Donor lungs had lower α diversity than healthy individuals (Shannon index; P = 0.000019; Figure 1A). Donor lung diversity varied considerably, but there was no correlation between diversity and donor age, time on ventilator, or PGD outcomes (data not shown).

Figure 1.

Donor lungs are dysbiotic compared with healthy controls. (A) α diversity as measured by the Shannon index is lower in donor lungs than in healthy controls (P = 0.000019; Wilcoxon rank sum). (B) Principal coordinate analysis (PCoA) using weighted UniFrac distances reveals donor lungs have distinct microbiome composition compared with healthy controls (P = 0.014, R² = 0.035; PERMANOVA). All donor lung samples that passed quality filtering are included (n = 75). Vectors indicate the taxa present at >2.5% relative abundance across samples responsible for the ordination of the PCoA plot. ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. CI = confidence interval; PC = principal coordinate; PERMANOVA = permutational multivariate analysis of variance.

Weighted UniFrac, which compares communities based on the phylogenetic relatedness of constituent taxa and their relative abundances (Figure 1B), found donor lung samples differed significantly from healthy subjects (PERMANOVA, R² = 0.035, P = 0.014). Some donor lungs overlapped with healthy control lungs and are characterized by oropharyngeal taxa such as Prevotella, Veillonella, and Rothia, whereas others were enriched with Pseudomonas or Staphylococcus. Accordingly, donor lung samples were more dispersed than healthy samples (ANOVA, P = 0.002).

Microbiome Changes Across the Peritransplant Period

We compared cognate donor lung and recipient allograft samples from 67 pairs in which both samples passed quality filtering (Figure 2). Matched recipient allograft samples had higher α diversity (Figure 2A; P = 0.00006), greater microbial biomass (Figure 2B; P = 0.00009), and different composition (Figures 2C and 2D; weighted UniFrac; R² = 0.043, P = 0.002; PERMANOVA). Despite group difference between donor lung and recipient allograft samples, recipient allograft microbiome composition strongly correlated with cognate donor lung composition (R² = 0.6, P < 0.001; PERMANOVA). Procrustes analysis, which compares the relationship between two ordination plots, confirmed that paired donor lung and recipient allograft microbiomes were closely linked (Figure 2E; M² = 0.77; P = 0.001). Thus, although there is significant change in the lung microbiome across the peritransplant period, the donor lung is an important determinant of the recipient allograft microbiome composition.

Figure 2.

Allografts immediately post-transplant have greater microbial biomass, higher diversity, and distinct composition compared with matched donor lungs immediately before transplant. (A) Recipient allograft specimens have greater Shannon diversity (P = 0.00031; Wilcoxon rank sum), (B) higher microbial biomass (P = 0.000049; Wilcoxon rank sum), and (C and D) distinct microbial composition on weighted UniFrac principal coordinate analysis (PCoA) (R² = 0.041, P = 0.001; PERMANOVA) than their matched donor lung samples. Note that C shows PCoA axis 1 (PC1) and axis 2 (PC2), and D depicts PCoA axis 2 (PC2) and axis 3 (PC3). (E) Procrustes analysis, which compares two ordination plots (donor lung microbiome and recipient allograft microbiome, respectively), demonstrates that donor lung and recipient allograft microbiomes are highly correlated (m12 squared = 0.77, P = 0.003). (F) Relative abundances of the most abundant oropharyngeal taxa in peritransplant subjects are increased in lung samples after transplant (P < 0.05; Wilcoxon rank sum). All pairs for which both donor lung and recipient allograft samples were available and passed quality filtering are included (n = 67 pairs). ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. CI = confidence interval; PERMANOVA = permutational multivariate analysis of variance.

The pre-/post-transplant difference was driven largely by PCoA axes 2 and 3, which substantially associated with typical oropharyngeal taxa. Therefore, we specifically queried the top five most abundant taxa that we previously reported comprised the oropharyngeal microbiome of patients with end-stage lung disease awaiting transplantation (Streptococcus, Prevotella, Veillonella, Rothia, Neisseria) (25). All five of these genera were enriched in recipient allograft samples compared with donor lungs (Figure 2F), suggesting oropharyngeal admixture across the transplant period.

Severe PGD Is Associated with a Distinct Post-transplant Microbiome Profile

To investigate dichotomous outcome groups that are likely to have more homogeneous biology, we performed a nested case–control analysis of PGD extreme phenotypes drawn from all 139 subjects by comparing those with severe PGD (PGD-3 at each of four time points assessed; n = 15) versus no PGD (PGD-0 at all four time points; n = 40). There was no difference between the groups in donor lung microbiome microbial biomass (P = 0.59), diversity (P = 0.18), or composition (PERMANOVA, R² = 0.049, P = 0.23).

In contrast, the recipient allograft microbiome in persistent PGD-3 differed significantly in composition from persistent PGD-0 (Figure 3A; PERMANOVA, R² = 0.07, P = 0.003). At the taxonomic level, differences were driven mainly by higher Prevotella and lower Streptococcus relative abundances (Figure 3B). PGD-3 and PGD-0 recipient allografts did not differ in microbial biomass (P = 0.66) or diversity (P = 0.97).

Figure 3.

Severe primary graft dysfunction (PGD) is associated with distinct recipient allograft microbiome composition. (A) Recipient allograft samples from subjects with persistent PGD-3 (n = 15) have distinct microbial composition compared with persistent PGD-0 (n = 40) on weighted UniFrac PCoA, with an upward shift in axis 2 (PC2) (PERMANOVA; P = 0.005, R² = 0.064). (B) Relative abundance of Prevotella is significantly increased (P = 0.017), and Streptococcus is decreased (P = 0.0035) in PGD-3 compared with PGD-0 (Wilcoxon rank sum). ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. CI = confidence interval; PCoA = principal coordinate analysis; PERMANOVA = permutational multivariate analysis of variance.

Prevotella and Streptococcus each had eight ASV taxonomic assignments (Table E1). The most abundant Prevotella ASV was assigned at the species level to Prevotella melaninogenica. The second-most abundant ASV was assigned only at the genus level, but by BLASTn aligned best with Prevotella oris (100% sequence coverage). Within Streptococcus, one ASV accounted for the majority of reads but was not assigned to a species level; BLASTn indicated 100% alignment with Streptococcus salivarius, with other Streptococcus spp. showing lower coverage. Thus, taxa in the Prevotella genus, which are typically anaerobic, associate with PGD, whereas taxa in the Streptococcus genus, which are typically facultative, inversely associate with PGD.

Lung Microbiome, PGD, and Clinical Covariates

We investigated the relationship between clinical covariates, PGD, and microbiome composition given differences seen in Table 1 and what is known about clinical factors and PGD (Table 2). We tested recipient factors (primary pulmonary diagnosis, exposure to antibiotics, and immunosuppression before transplantation), intraoperative factors (use of cardiopulmonary bypass [3], total cold ischemic time), and donor factors (cause of death, donor smoking). PGD remained significantly associated with recipient allograft microbiome in PERMANOVA testing after individually accounting for the variance of these clinical variables. There were no significant interaction terms to suggest that they mediated the relationship between the microbiome and PGD, though intraoperative bypass and donor cause of death nearly reached significance. Recipient pulmonary diagnosis was significantly associated with the lung microbiome (as expected, due largely to subjects with cystic fibrosis), but, as above, the interaction term with PGD was not significant. In a sensitivity analysis, removing the subjects with cystic fibrosis did not change the relationship between PGD and the recipient allograft microbiome (R² = 0.074, P = 0.004). Thus, the relationship between recipient allograft microbiome and PGD is not mediated by these clinical covariates.

Table 2.

Impact of Clinical Covariates on Recipient Allograft Microbiome and on the Relationship between Recipient Allograft Microbiome and Primary Graft Dysfunction

	Variance of Recipient Allograft Microbiome Explained by Clinical Covariate		Variance of Recipient Allograft Microbiome Explained by PGD Extreme Phenotype, Adjusted for Clinical Covariate		Interaction Term
Clinical Covariate	R2 (%)	P Value	R2 (%)	P Value	R2 (%)	P Value
None	na	na	7.00	0.003	na	na
Pulmonary diagnosis*	12.41	0.003	5.63	0.007	2.91	0.579
Antibiotics pretransplant	2.17	0.307	7.21	0.004	0.80	0.915
Immunosuppression pretransplant	1.89	0.404	6.20	0.005	1.40	0.604
Intraoperative bypass	2.07	0.311	7.62	0.001	3.85	0.060
Total ischemic time	0.30	0.995	7.04	0.007	2.3	0.303
Donor smoking	1.72	0.521	7.21	0.014	0.82	0.910
Donor death mechanism	2.80	0.702	7.04	0.004	6.95	0.058

Definition of abbreviations: na = not applicable; PERMANOVA = permutational multivariate analysis of variance; PGD = primary graft dysfunction.

PERMANOVA was used to test the relationship between clinical covariates, PGD, and microbiome composition (based on weighted UniFrac). Each row reports the output of a PERMANOVA model testing what proportion of the microbiome variance could be explained by the indicated clinical variable, the variance of the microbiome that was explained by PGD phenotype (persistent PGD-0 vs. PGD-3) after accounting for the variance explained by the clinical variable, and interaction term of the clinical variable and PGD in explaining microbiome variance. Data reported are the R² and P value for each relationship.

^*The impact of recipient pulmonary diagnosis on recipient allograft microbiome was driven by transplant recipients with cystic fibrosis versus transplant recipients without cystic fibrosis.

Lung Pepsin Is Elevated in Severe PGD and Correlates with Microbiome Composition

Severe PGD associated with recipient allograft enrichment in Prevotella and other members of the Bacteroidetes phylum. This finding led us to hypothesize that aspiration may be involved, so we measured lung concentrations of pepsin, often used as a biochemical marker of gastric aspiration (26). Pepsin was measured in recipient allograft samples with residual lavage available (n = 116) and healthy control subjects (n = 7). Pepsin was significantly higher in subjects with persistent PGD-3 compared with PGD-0 (P = 0.005; Figure 4A). Furthermore, pepsin was measurable in nearly all PGD-3 specimens (11/12) and only one-third of PGD-0 specimens (12/32) and was undetectable in all healthy subjects (0/7) (Fisher exact test; odds ratio, 17.2; P = 0.004 for PGD-3 vs. PGD-0).

Figure 4.

Pepsin is elevated in the lungs of subjects with persistent primary graft dysfunction-3 (PGD-3) and correlates with microbiome composition. (A) Pepsin concentrations are increased in post-transplant lung specimens of patients with persistent PGD-3 compared with PGD-0 (P = 0.0022; Wilcoxon rank sum). (B) Weighted UniFrac PCoA plot of all recipient allograft microbiomes, with pepsin concentrations indicated by color. Microbiome composition as represented by the principal coordinate (PC) axis is associated with pepsin concentrations in recipient allografts (pepsin vs. PC2, rho = −0.25, P = 0.007). ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. PC1 was not statistically significantly correlated. PCoA = principal coordinate analysis.

Furthermore, recipient allograft pepsin significantly correlated with microbiome composition (PCoA axis 2, P = 0.012; Figure 4B), with highest concentrations mainly in specimens enriched in anaerobic taxa (bottom left region of the PCoA plot). Pepsin concentration was inversely correlated to Streptococcus (rho = −0.26, P = 0.006) and showed a positive trend with Prevotella (rho = 0.17, P = 0.082). There was also a positive correlation between pepsin concentrations and richness (number of taxa present) (rho = 0.26, P = 0.006).

Persistent PGD-3 Has a Distinct Lung Inflammatory Profile

We measured 41 inflammatory markers in immediate post-transplant allografts by bead array. Compared with healthy lung specimens, recipient allografts had higher concentration of most of the measured molecules (Figure E3). This result indicated broad immune activation within the lung allograft immediately after implantation.

Lung cytokine concentrations in recipient allografts that developed persistent PGD-3 were significantly different from persistent PGD-0 in a principal component analysis (Figure 5A), driven by coordinated elevations in multiple mediators (PC1, P = 0.021). By univariate analysis, significant elevations were found for 10 of the 41 mediators in PGD-3, including Eotaxin C-C motif chemokine ligand 11 (CCL11), monocyte chemoattractant protein-1 (MCP-1) (CCL2), RANTES (CCL5), granulocyte colony-stimulating factor (G-CSF), granulocyte-macrophage colony-stimulating factor (GM-CSF), IL-8, interferon gamma-induced protein 10 (IP-10), fms like tyrosine kinase 3 ligand (Flt-3L), IL-6, and TNF-α (tumor necrosis factor-α) (all P < 0.05; Kruskal-Wallis) (Figure 5B). Thus, patients who develop severe persistent PGD-3 in the 72 hours after transplantation exhibit a broad inflammatory lung profile immediately after implantation.

Figure 5.

Elevated cytokines in recipient allograft of subjects who develop severe primary graft dysfunction-3 (PGD-3) compared with those without PGD. (A) Principal component analysis (PCA) of cytokines in recipient allografts shows an association between PGD status (persistent PGD-0 vs. PGD-3) and inflammatory markers, mainly driven by principal component axis 1 (PC1) (PGD vs. PC1, P = 0.05; Wilcoxon rank sum). Vectors show the top five cytokines responsible for PCA ordination along PC1 and PC2, respectively. (B) Ten of 41 measured cytokines independently associate with persistent PGD-3 in univariate analysis (P < 0.05; Kruskal-Wallis). ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05. CI = confidence interval; Flt-3L = fms like tyrosine kinase 3 ligand; G-CSF = granulocyte colony-stimulating factor; GM-CSF = granulocyte-macrophage colony-stimulating factor; IL = interleukin; IP-10 = interferon gamma-induced protein 10; MCP-1 = monocyte chemoattractant protein-1; RANTES = regulated on activation, normal T cell expressed and secreted; TNFα = tumor necrosis factor alpha; TNFβ = tumor necrosis factor beta.

Inflammatory and Noninflammatory Microbiome Types in Post-transplant Lung

We then investigated microbiome communities, inflammatory mediators, and pepsin across the entire cohort of recipient allograft samples, including those with intermediate PGD phenotypes. As shown in Figure 6A, there were multiple significant correlations between lung microbiome composition and host response mediators. Hierarchical clustering revealed two broad microbiome types: one characterized by Prevotella, Bacteriodales, and Veillonella that was positively correlated with multiple inflammatory mediators, and one negatively correlated with inflammation characterized by Streptococcus, Gemellaceae, and Stenotrophomonas.

Figure 6.

The recipient allograft microbiome is correlated with lung cytokines. (A) Heatmap shows correlation between cytokines and microbial taxa present at >1% abundance across samples. The color intensity corresponds to the magnitude of the Spearman correlation coefficient (red, positive association; blue, negative association). Rows and columns are clustered based on hierarchical distance. (B) Principal component analysis plot of cytokines in recipient allograft samples, with pepsin concentrations indicated by color. Pepsin concentrations correlated with cytokines as represented by principal component (PC) axes (pepsin vs. PC1, rho = 0.35, P < 0.0001; pepsin vs. PC2, rho = −0.21; P = 0.016). All recipient allograft microbiome samples analyzed for Luminex (n = 117) and pepsin (n = 116) are included. Flt3L = fms like tyrosine kinase 3 ligand; GCSF = granulocyte colony-stimulating factor; IFNgamma = interferon gamma; IL = interleukin; IL-17α = interleukin-17alpha; MIP1β = macrophage inhibitory factor 1 beta; TNFβ = tumor necrosis factor beta. Vectors are as described in Figure 5.

We next asked which mediators were associated with global microbiome composition by applying the Microbiome Regression–based Kernel Association Test (MiRKAT) (27), based on Bray-Curtis dissimilarity of taxa with ⩾20% prevalence. MiRKAT identified 17 mediators that met significance at an false discovery rate < 0.2 and 9 associated with microbiome type with false discovery rate < 0.05 (G-CSF, IL-10, IL-1RA, IL-1b, IL-6, IL-8, Macrophage inflammatory protein-1 beta (MIP-1b), RANTES, and TNF-α). Thus, allograft cytokine concentrations differed by microbiome composition.

There was also a strong association between pepsin concentration and analytes in the Luminex panel (Figure 6B; PC1, P = 9.2 × 10⁻⁷; PC2, P = 0.0013). The strongest two correlations were between pepsin and IL-6 (rho = 0.67, P = 3.21 × 10⁻¹⁶) and MCP-1 (rho = 0.62, P = 2.9 × 10⁻¹³), whereas IL-9 and IL-3 had modest negative correlations with pepsin (both rho = −0.26, P = 0.005); overall 38 of 41 analytes were significantly correlated, of which 36 had a positive correlation (Figure E4).

Prevotella/Streptococcus Ratio Associates with Pepsin, Inflammation, and PGD Risk

We sought to identify a taxonomic metric that associated with PGD. Given their distinct positive and negative associations, respectively, we calculated a Prevotella/Streptococcus ratio for all recipient allograft samples (Figure 7). The Prevotella/Streptococcus ratio was significantly associated with PGD (Figure 7A; P = 0.0015 for trend), with pepsin concentrations (Figure 7B; P = 0.0048), and with host response mediators (Figure 7C; PC1, P = 0.017).

Figure 7.

Prevotella/Streptococcus ratio associates with primary graft dysfunction (PGD) outcome and host cytokines. (A) The ratio between Prevotella and Streptococcus within a sample increases stepwise between persistent PGD-0, intermediate PGD, and persistent PGD-3 (P = 0.0021 for trend; Jonckheere trend test). (B) The Prevotella/Streptococcus ratio positively correlates with pepsin (rho = 0.26, P = 0.0058). (C) Prevotella/Streptococcus ratio indicated by color overlaid on PCA plot of cytokines in recipient allograft samples. The ratio correlates with lung cytokine profiles as represented by principal component (PC) axes (Prevotella/Streptococcus ratio vs. cytokine PC1, rho = 0.23, P = 0.015). All recipient allograft samples with microbiome analysis are included in A (n = 129), with microbiome and pepsin analysis in B (n = 116), and with microbiome and Luminex in C (n = 117). Vectors are as described in Figure 5. Flt3L = fms like tyrosine kinase 3 ligand; GCSF = granulocyte colony-stimulating factor; IFNgamma = interferon gamma; IL = interleukin; IL17α = interleukin-17alpha; MIP1β = macrophage inhibitory factor 1 beta; PCA = principal component analysis; TNFβ = tumor necrosis factor beta.

Because Prevotella and Streptococcus are typically anaerobic and facultative, respectively, we applied a bacterial respiration phenotype classifier that we previously reported (25). Classification of taxa with >1% mean abundance across post-transplant samples (Figure E5) revealed significant associations, with increasing anaerobes and decreasing facultative bacteria across PGD-0, intermediate, and PGD-3 (P = 0.0052 for increasing trend and P = 0.0011 for decreasing trend, anaerobes, and facultative bacteria, respectively; Jonckheere test). This result is concordant with differences in Prevotella and Streptococcus, which were the two most abundant taxa across samples. We compared the ability of these variables to predict PGD in a logistic regression model (Table E2) and found that the Prevotella/Streptococcus ratio generally performed best, with lowest Akaike information criteria and Bayesian information criteria with comparable AUC. This suggests that these most abundant taxa best explain the microbiome association with PGD and that cumulative effects from rare taxa do not contribute substantially.

Integrated Microbiome, Immune, and Biochemical Features in PGD

Last, we queried the discriminatory value of immediate postimplantation lung microbiome, inflammatory, and pepsin data to delineate PGD risk (Figure 8). We used the extreme phenotype subjects (n = 55) to perform feature selection with LASSO, which identified Prevotella/Streptococcus ratio, pepsin, RANTES, and TNF-α as key variables. A logistic regression model incorporating these factors had excellent discriminatory capacity for the extreme phenotype PGD outcome (AUC = 0.81). Interestingly, a model including only Prevotella/Streptococcus ratio without pepsin or mediators had only slightly reduced discriminatory power (AUC = 0.76). We then used this model on our full dataset, applying an alternative outcome definition previously used in the clinical literature (2) (PGD grade 3 on Day 2 or 3, which is remote from Day 0 sampling). Using this PGD outcome measure, the model still demonstrated discriminatory capacity, albeit modestly reduced (AUC = 0.66).

Figure 8.

A composite model of microbiome, pepsin, and cytokines classifies subsequent primary graft dysfunction (PGD). LASSO was used for variable selection, and then a multivariable logistic regression model was created to measure performance characteristics in differentiating different PGD outcome classifications for 129 transplant recipients. Receiver operating characteristic curve analysis of microbiome, pepsin, and cytokines in immediate postimplantation recipient allograft lung and development of PGD over 72 hours after transplantation. The Prevotella/Streptococcus ratio alone had a good ability (area under the curve [AUC] = 0.76) for prediction of subsequent development of severe PGD. This was improved with addition of Pepsin, RANTES, and TNFα to the model (AUC = 0.81). Testing on the outcome of PGD grade 3 on either Day 2 or 3 showed AUC = 0.66. TNFα = tumor necrosis factor α.

Discussion

We show that the development of PGD is associated with recipient allografts that immediately postimplantation exhibit enrichment of Prevotella and decreased Streptococcus, elevated pepsin, and a hyperinflammatory host response. In contrast, PGD did not correlate with the donor lung microbiome. This is the first analysis of the microbiota and inflammatory composition within the lung at the time of transplant and provides microbial and biochemical evidence implicating orogastric admixture in the allograft in the peritransplant period in the development of acute lung injury immediately post-transplantation.

Recipient allografts with greater Prevotella and decreased Streptococcus were at higher risk of incident PGD. Notably, the donor lung microbiome differed significantly from healthy and strongly influenced the recipient allograft microbiome, yet it did not correlate with PGD development. Concordant findings showed lung inflammatory mediators and pepsin correlated positively with Prevotella and negatively with Streptococcus. Together these data suggest that the peritransplant microbiome and aberrant host response contribute to early allograft injury and raise two main possibilities for this relationship: that the lung bacteria drive inflammation and/or injury, or that the lung bacteria serve as a marker for another event, such as aspiration, which drives PGD.

Previous work on PGD pathogenesis has implicated IL-8 signaling and neutrophil immigration (28, 29), anti-collagen type-V immunity (30), elevated IL-1, IL-6, and CCL4, as well as innate immune signaling pathways, particularly related to TLR (toll-like receptor) signaling (4, 31, 32). We showed here that recipient allografts had elevations of multiple mediators compared with healthy subjects, and severe PGD had a hyperinflammatory profile that correlated with the microbiome and pepsin. This suggests that for severe persistent PGD the die is likely already cast by the time of organ reperfusion. Indeed, most measured mediators were elevated in patients with PGD, suggesting a broad host response. Although we did not have sufficient material to assay mediators and pepsin in donor lungs, our microbiome data implicate factors developing during the time between organ procurement and postreperfusion sampling.

Relative enrichment of Prevotella in severe PGD is intriguing because this genus has been noted to be proinflammatory in lung, suggesting that it could be a proximal driver of inflammation and injury (33). Prevotella interacts with epithelial and dendritic cells to produce first-order inflammatory cytokines that recruit and amplify Th17 cells. Prevotella and other oropharyngeal taxa have been associated with increased Th17 immune tone in healthy subjects (10) and animal models (34). However, IL-17 was not elevated in our subjects with PGD, suggesting alternative inflammatory pathways were activated or that we sampled before a later increase in IL-17. Although multiple cytokines and chemokines were elevated in PGD, innate immune pathways (Eotaxin, RANTES, IL-6, TNF-α) and those more specific to granulocytes (IL-8, MCP-1, G-CSF) were particularly enriched. This may reflect first-order innate immune responses to proinflammatory bacterial taxa. Alternatively, the increase in oral taxa might be cooccurring with additional biological factors, including pepsin, responsible for priming the lung for early injury.

Prevotella, which is generally anaerobic, and Streptococcus, which is generally facultative, are both typical oral taxa. Combined with the finding of increased pepsin in PGD, these data implicate oropharyngeal and gastric admixture in the allograft. However, why Prevotella but not Streptococcus associates with lung pepsin is unclear. Interestingly, a recent study reported that gastroesophageal reflux disease after lung transplant was associated with Prevotella and Veillonella in the lung and that after Nissen fundoplication the Prevotella-dominated microbiome type decreased (35). Although that study reported that a Prevotella-type microbiome was less inflammatory than a pathogen-enriched microbiome, the different contexts between this study and that (acute perioperative versus post-transplant longitudinal) may account for host response differences. It is also plausible that pretransplant treatment or intraoperative factors could favor expansion of some oropharyngeal-derived taxa over others in our study, although we did not find an association between the microbiome and organ ischemic time and do not have information on donor antibiotic treatment. Alternatively, Streptococcus may compete with Prevotella in the respiratory tract, particularly S. salivarius, which best aligns with our data’s most abundant Streptococcus ASV (36). Thus, a future question is whether this or other beneficial taxa might hold therapeutic potential as a “respiratory probiotic” (36).

Several clinical risk factors have been associated with PGD, including pulmonary hypertension, cardiopulmonary bypass use, and obesity (3). We focused on lung allograft microbiome profiles and developed a model with good discriminatory capacity for severe persistent PGD, although it had less power for determining PGD grade 3 on Days 2 or 3. This might reflect heterogenous mechanisms leading to PGD, with some cases of PGD-3 on Days 2 or 3 reflecting later events not captured on Day 0. However, recipients in the intermediate group who started with high PGD scores and then decreased did not differ in allograft microbiome composition, inflammation, or pepsin compared with recipients who started with low PGD scores and increased (data not shown). Therefore, the clinical utility of our model as a biomarker will require work with additional cohorts for validation in predicting PGD. Future studies with sequential sampling after transplant could also answer whether subjects who develop intermediate levels of PGD acquire similar allograft microbiome, biochemical, and inflammatory features in the days after immediate postimplantation sampling.

Although our findings of elevated oropharyngeal anaerobes and the gastric biomarker pepsin in recipient allografts who go on to develop severe PGD suggest the possibility of aspiration, when or how this might occur is not clear, because recipients were still intubated when the recipient allograft sample was taken. Possibilities include donor or recipient microaspiration (despite intubation) during the procurement or implantation procedures, or recipient microaspiration before the procedure that impacted the larger airways before secondarily involving the lower airways. Although we did not have sufficient sample to measure pepsin or cytokines in the donor lung sample, the fact that PGD correlated with the microbiome in recipient allografts, but not donor lungs, would argue against donor aspiration before sampling. Thus, further study is warranted to investigate aspiration as a possible driver and, if so, clarify the timing involved.

Strengths of our study include a relatively large number of subjects, rigorously defined PGD, a unique set of specimens from both the donor lung before procurement and allograft after implantation, and the integrated analysis of microbial, biochemical, and inflammatory datasets. Also, this large sample set enabled us to take an extreme phenotype approach to minimize the impact of clinical heterogeneity and subjects with intermediate phenotypes. Finally, the highly correlated nature of the three types of data (microbiome, inflammatory, biochemical) and multiple parameters associating with PGD suggests a robust relationship.

Our study has several limitations. Despite the overall large sample, the numbers of subjects in the extreme phenotype analysis were modest (40 persistent PGD-0 and 15 persistent PGD-3, of whom 14 had recipient allograft samples), and therefore PGD correlations with the lung microbiome, pepsin, and inflammatory markers should be interpreted accordingly. We did not have sufficient sample to measure pepsin and mediators in donor lung. The lavage volume used in this study was less than the typical ∼100 ml BAL, although it was consistent across sample types. Although microbiome profiles were also correlated with intermediate PGD phenotypes, the discriminatory power among all subjects was not as strong and warrants validation in independent cohorts.

In summary, we found links between the microbiome, inflammation, pepsin, and PGD. Based on anaerobic oral-type taxa involved and elevations in pepsin, our data implicate the possibility of aspiration during the peritransplant period as a potential priming mechanism for acute lung injury. Given this is potentially actionable, future studies should further investigate this putative mechanism, timing and mechanism, possible interventions, and the ability for risk stratification based on microbiome and related host factors.