Distinct blood and lung proteins drive pulmonary capillary leak in children with severe hypoxemic respiratory failure

Abstract

Objective:

Acute hypoxemic respiratory failure is a leading cause of death and disability in critically ill children. Pulmonary capillary barrier dysfunction, in large part, drives severity. We aimed to define how blood and tracheobronchial lavage (TBAL) fluid differentially affect pulmonary capillary barrier function and, as a second exploratory aim, identify protein mediators.

Design:

We performed a secondary analysis in April and May of 2024 of blood and TBAL samples from a previously completed observational cohort study of mechanically ventilated children collected from October 2018 through February 2020.

Setting:

Single pediatric intensive care unit

Subjects:

Stored blood and TBAL samples from 65 children requiring mechanical ventilation collected at 24 and 48–72 hours after intubation

Interventions:

None

Measurements:.

We quantified inflammatory proteins in plasma and TBAL using Olink Target 48 Cytokine multiplex. We assessed changes in barrier function of cultured human pulmonary microvascular endothelial cells (HPMEC) using electrical cell-substrate impedance to measure trans-endothelial electrical resistance.

Main results:

Plasma samples from critically ill children significantly enhanced HPMEC barrier function and contained lower concentrations of pro-inflammatory cytokines compared to TBAL. Over time, the effects of plasma on barrier integrity diminished, while TBAL samples significantly improved barrier function and had reduced abundance of several pro-inflammatory cytokines. Notably, TBAL samples from children with severe lung injury augmented HPMEC barriers more than those from children without lung injury, highlighting distinct compartmental and disease-specific influences on HPMEC barriers.

Conclusions:

These unexpected findings reveal that blood and lung compartments contain differentially abundant proteins that exert distinct and evolving influences on pulmonary endothelial integrity in critically ill children. These results may inform on the timing and compartment-specific assessment of biomarkers, and eventually, delivery of therapies.

Introduction

Research in Context

• Pulmonary edema is a central mechanism in pediatric acute hypoxemic respiratory failure and is controlled, in large part, by the pulmonary capillaries.

• Proteins in either the blood or lungs may alter pulmonary capillary permeability and differential abundance of key proteins may change over the course of disease.

• This study is the first to integrate paired plasma and tracheobronchial lavage samples with functional endothelial assays to identify distinct and evolving effects on pulmonary capillary barrier integrity.

Acute hypoxemic respiratory failure is associated with substantial morbidity and mortality in critically ill children(1). Despite decades of research, acute patient management is largely limited to intensive supportive care(2). A major barrier to developing targeted therapies is lung injury heterogeneity(3), driven by a complex and dynamic interaction between patient, disease, and treatment-related factors. Clinical phenotypes of severe disease have been proposed, including those categorized by hyper- and hypo-inflammatory pathways in children(4–7). Development of novel treatments has been hindered by two major barriers: 1) disproportionate emphasis on assessing a single compartment (i.e., cytokines in blood); and 2) current biomarkers lack a direct link to distinct pathophysiologic processes.

One distinct pathophysiologic process is the breakdown of the pulmonary capillary barrier(3). Independent of inciting etiologies, inflammation or injury induces loss of pulmonary capillary barrier function and accumulation of interstitial fluid and, when the epithelial barrier is disrupted, pulmonary edema. Inappropriate fluid accumulation in the lung tissue and airspaces has been recognized since the conception of acute respiratory distress syndrome (ARDS) as a core pathophysiologic process(8). This is because interstitial fluid and pulmonary edema exacerbate hypoxemia via three main processes: 1) reduced lung compliance due to fluid-filled lungs; 2) shunt across collapsed and fluid filled alveoli; and 3) increased distance for oxygen diffusion through edematous interstitial space. Loss of pulmonary capillary barrier function is closely tied to clinical severity, with prominent endothelial injury linked to higher mortality (6, 7, 12–19).

Despite its clinical significance, the drivers of pulmonary capillary leak in critically ill children are poorly understood. Prior studies have focused on individual proteins in the blood(5, 7, 9–15), lungs(16–20) and rarely, both(21). Collectively, these studies corroborate a paradigm that pro-inflammatory cytokine levels correlate with disease severity. However, none investigated effects on pulmonary capillary function. Pulmonary capillaries are in intimate contact with both blood and alveolar compartments, and dysfunction may arise from pathology in either location. The objectives of this study are to understand how plasma and tracheobronchial lavage (TBAL) alter pulmonary capillary barrier function in vitro and identify proinflammatory cytokines driving these barrier function changes.

To accomplish this, we analyzed samples from a previously completed cohort study of critically ill children undergoing invasive mechanical ventilation. We collected parallel and sequential blood and TBAL banking alongside pertinent clinical data. We assessed for associations between capillary barrier dysfunction and inflammatory cytokines and evaluated how these variables vary by severity of respiratory failure. Transendothelial electrical resistance (TEER) was used to monitor how patient samples induced barrier function changes in human pulmonary microvascular endothelial cells (HPMEC). We hypothesized that samples from critically ill children with more severe disease would result in increased capillary leak.

Methods

Study Design

We analyzed samples from a previously conducted, but not yet published, single-center observational cohort study that was conducted October 2018 through February 2020. Inclusion criteria included: children <18 years old undergoing invasive ventilation. Both medical and surgical patients were included. Exclusion criteria included: elective intubation, a preexisting tracheostomy, cyanotic heart disease and ventilation duration <48 hours. No patients were excluded based on any demographic factors. Demographics, historical features, admission diagnoses, comorbidities, laboratory data, ventilator parameters, and clinical data during hospitalization were collected. Pediatric ARDS (PARDS) was identified using Pediatric Acute Lung Injury Consensus Conference Group criteria with disease severity established by oxygenation index (OI)(22). The study flowchart and clinical data are presented in Supplemental Figure 1 and Table 1 respectively. The protocol was approved by the Johns Hopkins University Institutional Review Board (IRB#: 00168848, Approved: 4/11/2018) and conducted in accordance with the Code of Ethics of the World Medical Association and Recommendations for the Conduct, Reporting, Editing and Publication of Scholarly Work in Medical Journals. All study procedures were followed in accordance with the ethical standards of the responsible committee on human experimentation (Johns Hopkins University IRB) and with the Helsinki Declaration of 1975.

Table 1.

General patient characteristics, admission diagnoses, clinical outcomes, and the distribution of available biospecimens for pre-specified analyses for the study sample.

Variables / Characteristics	Study sample, N = 65
Patient characteristics and comorbidities
Age (years), median (IQR)	2.5 (0.6–9.8)
Male, n (%)	37 (56.9)
Weight (kg), median (IQR)	14 (7.3–35)
Historical comorbidities, n (%)
Chronic lung disease	9 (13.9)
Congenital heart disease	2 (3.1)
Neuromuscular disease	3 (4.6)
Prematurity (<35 weeks gestation)	12 (18.5)
Admission diagnoses, n (%)
Pediatric Acute respiratory distress syndrome	27 (41.5)
Altered mental status	17 (26.2)
Asthma exacerbation	2 (3.1)
Bacterial pneumonia	24 (36.9)
Status epilepticus	14 (21.5)
Trauma	4 (6.2)
Viral infection	30 (46.2)
Bronchiolitis	17 (26.2)
COVID-19	4 (6.3)
Enrollment ventilator and oxygenation parameters
Ventilator parameters
PIP (cm H2O), mean ± SD	25 ± 6.6
PEEP (cm H2O), mean ± SD	7.5 ± 2.6
MAP (cm H2O), mean ± SD	13.8 ± 5.3
FiO2 (%), mean ± SD	48.7 ± 17.3
Vt (mL/kg), mean ± SD	6.4 ± 1.3
Cdyn (ml/cm H20/kg), mean ± SD	0.4 ± 0.2
Inspired nitric oxide, n (%)	11 (16.9)
OI, median (IQR)	9 (4.9–13.4)
PaO2 (mmHg), median (IQR)	65 (56–91)
PF ratio, median (IQR)	140 (113–208)
Clinical course and hospital outcomes
Duration of ventilation (days), median (IQR)	8 (3–15)
Duration of hospitalization (days), median (IQR)	21 (11–44)
Extracorporeal life support during admission, n (%)	4 (6.2)
Tracheostomy at discharge, n (%)	5 (7.7)
Mortality at discharge, n (%)	4 (6.2)

IQR: interquartile range, SD; standard deviation, PIP; peak inspiratory pressure, PEEP; positive end expiratory pressure, FiO2; fraction of inspired oxygen, MAP; mean airway pressure, V_t; tidal volume, C_dyn; dynamic compliance, P_aO₂; partial pressure of arterial oxygen, PF ratio; partial pressure of arterial oxygen to fraction of inspired oxygen ratio, OI; oxygenation index, TBAL; trachea-bronchiolar lavage; Duration of ventilation for survivors.

Biospecimen collection, processing, and cytokine analysis

Whole blood and TBAL were simultaneously collected within 24-hours of, and 48- to 72-hours after, endotracheal intubation. TBAL was collected via the following protocol(18); 1) a sterile suction catheter gently inserted through the endotracheal tube until resistance was felt (typically 2–4cm beyond the marked distance of the endotracheal tube. 2) 5mL of saline was applied through the endotracheal tube followed by 2–3 mandatory ventilation breaths, 3) On a single pass, suction was applied (80–120 mmHg) and the catheter gradually removed from the endotracheal tube, 4) Samples were aliquoted and cryopreserved at −80°C. Blood was processed to platelet-poor plasma per manufacturer protocols, aliquoted and cryopreserved at −80°C. Cytokines in plasma and TBAL were measured in April 2024 using the Olink^® Target 48 Cytokine multiplex protein affinity assay and quality control information at: https://olink.com/products/olink-target-48) as we have previously(23). To account for sample dilution and variability, the panel included commercially validated Olink internal standards which normalize data across samples by monitoring assay performance independently of antigen concentration(24). 397 out of 400 samples passed Olink quality control; samples that failed were not included in the analysis. All proteomic data, including correlation between plasma and BAL, is presented in Supplemental Data.

Barrier function by electrical cell-substrate impedance sensing (ECIS).

This is a secondary analysis of stored samples. ECIS assays were run in May and June of 2024. We used adult donor HPMECs (Lonza, CC-2527) for TEER assays as previously described(25). Briefly, HPMECs were passaged in EGM-2 MV Microvascular Endothelial Cell Growth Medium (Lonza, CC-3202, CC-3156). HPMECs from passages 4–10 were seeded onto gelatin-coated Applied Biophysics 96W20idf array. Changes in TEER were assessed at 1-Volt, 4,000 Hz using ECIS (ECIS Z-Theta, Applied BioPhysics). Thawed plasma was filtered through Nanosep OMEGA 300k (Cytiva, OD300C34) by centrifugation to prevent coagulation. HPMECs reached a stable TEER of 450 Ohms/mm² and were then treated with 10% v/v patient TBAL or filtered plasma, and TEER was monitored for 10 hours. Plots of TEER changed over time were generated by the manufacturer’s software. TEER data were analyzed as relative values where each data point is divided by the initial TEER for that condition. To quantify the effect of individual patient samples on HPMEC barrier function over experimental time we calculate Area Under the TEER Curve (AUC). Individual AUC curves are used to calculate the Area Between Curves (ABC). ABC is the difference of AUC of two samples and is used to compare effects on TEER between sample types, normalized to the saline control. A positive ABC value indicates gain, and negative value a loss, of HPMEC barrier function over time compared to saline, respectively.

Statistical analysis

Descriptive statistics were used to summarize demographic and clinical data, including proportions, means with standard deviations, and medians with interquartile ranges. Group comparisons were performed using Student’s t-test, Wilcoxon rank-sum test, and Fisher’s exact test, as appropriate. Cytokine differential abundance was analyzed using the Mass Dynamics platform. For in vitro assays, the magnitude of effect over time was analyzed as area under the normalized TEER curve (AUC) using Prism 10.1.0. AUC values were normalized to saline-treated controls by subtracting the saline AUC from the sample AUC (referred to as the Area Between the Curves [ABC]). To quantify changes in effect over clinical time, we calculated the Change in ABC (ΔABC) by subtracting paired ABC values between < 24 and 48–72 hours (Figure 1). Pearson’s correlation was employed to assess relationships between TEER and clinical parameters. For subgroup analysis a Bonferroni correction for statistical significance was applied. Statistical analysis and graph generation were performed using GraphPad Prism version 10 and Stata version 13.1

Figure 1: Plasma and tracheobronchial lavage (TBAL) samples differentially alter human pulmonary microvascular endothelial cell (HPMEC) barrier function.

A) The average effect of all plasma samples (n=63, 27♀, 36♂) is to improve HPMEC barrier function, as indicated by increased trans-endothelial electrical resistance (TEER) above saline control. TEER data is plotted as relative electrical resistance (i.e. normalized to its starting resistance, R(t)/R0). B) Averaging all TBAL samples (n=47, 25♀, 22♂) reveals a short-term increase followed by a longer decrease in barrier function. To quantify the effect of individual patient samples on HPMEC barrier function over experimental time we calculate Area Under the TEER Curve (AUC). Individual AUC curves are used to calculate the Area Between Curves (ABC). ABC is the difference of AUC of two samples and is used to compare effects on TEER between sample types, normalized to the saline control. A positive ABC value indicates gain, and negative value a loss, of HPMEC barrier function over time compared to saline, respectively. C) Plasma samples more significantly improved HPMEC TEER compared to TBAL samples, as indicated by higher ABC values. D) Volcano plot showing significant differential protein abundance in BAL samples compared to plasma, defined as Log2 fold change ≥│1.5│and adjusted p < 0.05.

Results

Patient characteristics.

Sixty-five children were included in the study (Supplemental Figure 1); demographic and clinical parameters along with distribution of available biospecimens are listed in Table 1.

Plasma and TBAL effects on TEER and associated proteomics.

Aggregate analysis of blood and TBAL samples at both timepoint revealed plasma augmented HPMEC barrier function to a greater degree than TBAL, as indicated by median ABC above zero (Figure 1A). TBAL samples exhibited variable effects, with nearly half of the samples decreasing HPMEC TEER (Figure 1B). This difference is significant (p=0.0015, Figure 1C) and was associated with numerous significantly more abundant proteins identified in both TBAL; tumor necrosis factor (TNF), interleukin (IL-)-1β, IL-17c, IL-10, IL-6, IL-33, oncostatin M (OSM), TNF-related apoptosis-inducing ligand (TNFSF10, a.k.a. TRAIL), vascular endothelial growth factor-A (VEGF-A), oxidized low-density lipoprotein receptor 1, epidermal growth factor, C-X-C (CXCL-) and C-C motifs (CCL-) chemokines CXCL8, CXCL9 and CCL3, and plasma; IL27, IL17c, CXCL12, CCL11, −13 and −19, lymphotoxin-α (LTA) and Fms-related tyrosine kinase 3 ligand (FLT3LG) (Figure 1D). Relative abundance of proteins between plasma and TBAL samples did not correlate (Supplemental Figures 2, 3).

TEER effects from paired plasma and TBAL over evolution of lung injury (i.e. clinical time) and associated proteomics.

When analyzed as a group, treatment with 24-hour compared to 48-hour plasma samples induced a non-significant decrease in TEER, as indicated by slightly lower median ABC (Figure 2A). In contrast, 24-hour TBAL samples increase HPMEC TEER as much as later samples, as indicated by higher median ABC (Figure 2B). The difference in change of HPMEC barrier function over clinical time induced by plasma and TBAL samples is significant, as indicated by higher median ABC (Figure 2C). We identified no significantly differentially abundant proteins in the blood over clinical time. We did identify several pro-inflammatory proteins more abundant in TBAL from early time points; TNF, IL-1β, IL-6, IL-10, IL-17c, OSM and matrix metalloproteinase-12, Figure 2D).

Figure 2: Plasma and tracheobronchial lavage (TBAL) samples differentially alter human pulmonary microvascular endothelial cell (HPMEC) barrier function over clinical time.

HPMEC treated with samples from the same patient collected at 24 versus 48 hours and monitored for changes in trans-endothelial electrical resistance (TEER). The magnitude of effect of patient samples on HPMEC TEER are quantified as Area Under the TEER Curve (AUC) and normalized to saline control by calculating the Area Between TEER Curves (ABC). Positive ABC suggests improved TEER over time relative to saline control. A) Non-significant increase in ABC from plasma at 24 hours (left, checkered fill) compared to 48 hours (right, solid fill, 24hrs n=60, 48hrs n=53). B) Significant difference in ABC from TBAL at 24 hours compared to 48 hours (24hrs n=38, 48hrs n=34). C) The change in Area Between Curves (ΔABC) compares changes in HPMEC barrier function induced from samples from the same patient from 24 to 48 hours (i.e. biologic changes over clinical time or evolution of respiratory failure). Plasma samples induced a slightly negative ΔABC (i.e. biologic changes over clinical time). Plasma samples induced a slightly negative ΔABC (i.e. barrier promoting effect of plasma is lost over clinical time), whereas BAL samples exhibit a significantly positive ΔABC. D) Positive TBAL ΔABC is associated with several significantly more abundant cytokines in BAL samples at 24 hours.

Plasma and TBAL TEER by etiology of respiratory failure and associated proteomics.

Patients that met PARDS criteria(22) had worse ventilator parameters and oxygenation, compared to those without primary lung injury (NLI). In paired analyses by patient over time, 24-hour plasma samples from both NLI and PARDS groups similarly augmented HPMEC TEER, as indicated by positive median ABC (Figure 3A, left columns). Somewhat surprisingly, plasma samples from patients without lung disease collected 48-hours after intubation significantly disrupted HPMEC TEER compared to patients with PARDS, as indicated by median ABC less than 0 (Figure 3A, right columns). This was associated with non-statistically significantly increased IL-10 in the later PARDS samples (Figure 3B). Also unexpectedly, both 24- and 48-hour TBAL samples from NLI patients significantly reduced HPMEC TEER compared to samples from PARDS patients, as indicated by lower median ABC values (Figure 3C). The temporal trajectory of barrier function also differed by patient group; plasma from PARDS patients improved HPMEC TEER between 24- and 48-hour samples (as indicated by a ΔABC value above zero), while plasma from NLI reduced HPMEC TEER between 24 and 48 hour samples (as indicated by a ΔABC value below zero) (Figure 3D).

Figure 3: Plasma and tracheobronchial lavage (TBAL) samples differentially alter human pulmonary microvascular endothelial cell (HPMEC) barrier function based on degree of lung injury and these differences change over clinical time.

HPMEC treated with samples from the same patient collected at 24 versus 48 hours and monitored for changes in trans-endothelial electrical resistance (TEER). The magnitude of effect of patient samples on HPMEC TEER are quantified as Area Under the TEER Curve (AUC) and normalized to saline control by calculating the Area Between TEER Curves (ABC). Positive ABC suggests improved TEER over time relative to saline control. Effects of the same sample type from the same patient over clinical time are represented as the change in ABC (ΔABC). Positive ΔABC indicates an increase in HPMEC TEER when treated with samples from same patient collected at 48 to 24 hours. The data show differences in ΔABC from patients with No primary Lung Injury (NLI) compared to patients with Pediatric Acute Respiratory Distress Syndrome (PARDS). A) To compare effects of lung injury severity we identified subgroups of patients with pediatric acute respiratory distress syndrome (PARDS) and no lung injury (NLI). Plasma from NLI patients significantly decreases ABC compared to plasma from patients with PARDS at 48 hours (right, solid fill) but not at 24 hours (left, checkered fill, p=0.0359, NLI 24 hrs n=18, NLI 48 hrs n=15, PARDS 24hrs n=24, PARDS 24hrs n=24). B) This difference is associated with non-significantly upregulated cytokines in PARDS plasma samples at 48h compared to NLI plasma samples at 48h hours. C) BAL samples collected from patients with PARDS significantly increase ABC values compared to NLI patients at both 24 hours (left, checkered fill, p=0.0022) and 48 hours (right, solid fill, p=0.0085, NLI 24 hrs n=14, NLI 48 hrs n=8, PARDS 24hrs n=11, PARDS 24hrs n=15). D) Plasma from PARDS patients induces significantly increased ΔABC compared to NLI patients (p=0.0315, NLI n=14, PARDS n=22). No effect is induced by a limited number of paired TBAL samples (NLI n=8, PARDS n=7).

Changes in TEER in relation to clinical parameters.

Our mechanistic findings correlate to clinical severity of lung injury as measured by OI(26). Change in OI from enrollment through 48–72 hours following intubation (ΔOI) correlated positively with 48-hour TBAL ABC, duration of invasive ventilation and length of stay (Supplemental Table 2). The fraction of inspired oxygen was statistically different between patients with negative ΔABC (i.e., 48-hour sample induced more leak) and positive ΔABC (Supplemental Table 3). Partial pressure of oxygen in the arterial blood was increased in patients with a positive ΔABC induced by TBAL samples (Supplemental Table 4). Correlation analysis of individual protein concentrations to OI and ABC stratified by disease severity and timepoint revealed numerous statistically significant positive and negative monotonic relationships (Supplemental Table 5). Analysis of effects induced by plasma and TBAL over clinical time and by clinical diagnosis (i.e., patient with NLI compared to those with bronchiolitis) revealed small but significant changes in ABC and cytokine concentration (Supplementary Figure 4). All primary clinical, proteomic and ABC values are listed in the Supplementary Data File.

Discussion

What this Study Means

• Plasma and TBAL exert opposing effects on pulmonary endothelial barrier function, and these effects are changing in opposite directions over time.

• TBAL generally contains more pro-inflammatory mediators compared to plasma and this is associated with these samples inducing worse pulmonary capillary barrier function.

• These findings support the development of targeted interventions to reduce pulmonary capillary leak should consider compartmental and temporal changes.

This exploratory study is the first to report the effects of paired plasma and TBAL on pulmonary capillary barrier function correlated with differentially abundant inflammatory proteins. We employed a novel and integrative analytic approach to link the effects of patient samples on vascular barrier function to clinical parameters and differentially abundant protein. TEER provides a real-time, non-invasive, label-free, and quantitative measure of endothelial cell behavior in response to the proteins in patient samples. The inflammatory signatures revealed several molecules associated with known effects on the capillary barrier. Using this approach, we demonstrate that plasma and TBAL samples from critically ill children have differential, time- and disease severity-dependent effects on pulmonary capillary barrier function. Finally, we successfully identified clinical characteristics and protein signatures in plasma and TBAL samples associated with the observed changes in TEER. Our findings support the common-sense notion that there are different biologic processes occurring in the blood and alveolar compartments in critically ill children and demonstrate, for the first time, that these processes have differing effects on pulmonary capillary barrier function.

Loss of pulmonary capillary barrier function is clinically relevant; the severity of acute hypoxemic respiratory failure correlates to microvascular endothelial injury(27). Endothelial injury contributes to ventilation-perfusion mismatch, reduced lung compliance, and increased diffusion distance for gas exchange, ultimately leading to hypoxemia and increased work of breathing(3). The endothelial-alveolar barrier is extremely thin and influenced by the cells and biologic response in both compartments. However, few prior studies have considered both compartments, and none have focused on the ultimate effect of these processes on the capillary barrier.

We demonstrate that TBAL samples induce significantly less increase in TEER in HPMECs compared to plasma samples. These findings correlated with more abundant proinflammatory cytokines in the TBAL samples. Among them, TNF and IL-1β are well-established to disrupt EC junctions(28). This occurs via simultaneous activation of Rho/ROCK/MLKC(29) and nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) signaling pathway(30). These processes ultimately lead to increased actin fiber formation, increased cell tension, disruption of cell junctions allowing paracellular leak(31). Similarly IL-33, which signals through the ST2 receptor (a member of the IL-1 receptor family) and OSM act on endothelial cells to promote inflammation(32). These molecules correlated with leak induced TBAL samples, suggesting they may be mediators of endothelial dysfunction in PARDS.

The cytokines upregulated in plasma have more diverse functions. IL-27 is a pleiotropic cytokine(33) with unclear effects on endothelial cells. LTA may interact with the lymphotoxin-β, TNF-type 1, or TNF-type 2 receptors on endothelial cells(34). IL-17c may contribute to sustained inflammation and drive worse lung injury(35). FLT3LG is a hematopoietic growth factor that may amplify detrimental immune responses in lung injury(36). These findings demonstrate that the alveolar compartment is enriched with immune mediators that promote endothelial permeability and immune cell recruitment. Taken together, increases in the pro-inflammatory and vascular-active proteins in TBAL do, at least partially, explain the loss of barrier function in HPMEC treated with them (as indicated by lower TBAL ABC in Figure 1C).

Several chemokines were also differentially enriched in TBAL and plasma. These findings are striking, but difficult to interpret. In TBAL; CXCL8 (IL-8), CCL3 (MIP-1α), CCL4 (MIP-1β), CCL7 (MCP-3) and CXCL9 (MIG) and plasma; CXCL12, CCL13, CCL11 and CCL19. These chemokines are produced by epithelial, endothelial, and immune cells both in the lung and blood. The signature from TBAL may favor recruitment of neutrophils (CXCL8), NK cells (CXCL9), Th1 polarized T-cells (CCL3, CXCL9), and macrophages and monocytes (CCL3, CCL4 and CCL7). However, there is great redundancy in the action of these chemokines, as well as numerous effects on cells not related to recruitment that limit more definitive conclusions.

More definitively, we found that the different processes occurring in the blood and lungs are changing over time with disparate effects on the pulmonary capillary barrier function. Although there is a slightly negative ΔABC induced by plasma samples, the change is driven by the significant positive ΔABC induced by TBAL samples. This correlated with more abundant pro-inflammatory cytokines in the 24-hour TBAL samples. IL-1β and TNF activate NF-κB signaling and paracellular leak in endothelial cells. OSM and IL-17 may also augment the inflammatory responses in endothelial cells or intimately related pericytes. IL-10, an anti-inflammatory cytokine that has variable, concentration dependent effects in endothelial and immune responses(37), was also more abundant. IL-10 levels have been associated with greater severity of lung injury in children(6) and adults(38). IL-10 is generally thought to be anti-inflammatory(39), and increases may reflect a compensatory response to dysregulated inflammation. The effects of IL-10 may be concentration dependent, and in our hands it does not alter HPMEC TEER (unpublished data).

Our findings support the novel hypothesis that the different processes in the blood and lungs are changing over time, and that these changes have opposite effects on pulmonary capillary permeability. These findings have important implications for the development of future studies on the mechanism of breakdown and repair of the alveolar barrier. Additionally, our results may guide refinement of biomarker-based risk models and therapies in pediatric respiratory failure. Single compartment, single time-point biomarker assessment may miss key, ever-changing, pro- and anti-inflammatory processes. The development of treatments should be targeted to a specific target, in a specific compartment, at a specific time. Our work sets a foundation to explore the best timing and route for therapies. For example, although much more data is needed, our findings suggest targeting TNF or IL-1 may be more effective in the lung rather than in the blood and may be more effective at earlier rather than later timepoints, at least in a way that may ameliorate pulmonary capillary leak.

Finally, the observation that samples from patients without lung injury generally reduced HPMEC TEER, as indicated by lower median ABC comparing NLI to PARDS, is unexpected. This effect is observed in both plasma (Figure 3A, right) and TBAL (Figure 3C, right) at 48 hours. One possible explanation for this finding could be emergence of subgroups of patients that develop more inflammation after 48 hours of critical illness. This hypothesis is supported a clear group of patients whose plasma reduces HPMEC TEER at 48 hours (i.e. the bulge of ABC below zero in NLI plasma at 48 hours, Figure 3A). Unfortunately, our study is underpowered to more definitely explore the evolution of new subgroups of patients.

Limitations

Our study has several limitations. The sample size was small with missing samples (Supplemental Table 1) and considerable variability within our clinical groups. We were also unable to control many factors, including treatments administered or time from disease onset to ICU admission and study enrollment. Due to small sample size, we were unable to compare profiles between direct (pulmonary) and indirect (extrapulmonary) PARDS, which likely have to distinct pathologic mechanisms of great interest. Subgroup analysis is limited by smaller sample sizes heterogenous effects on TEER, as indicated by discrete bulges and tails in the violin plots. In our TEER assay, we observed a downward trend in baseline measurements in some experiments (17% barrier function loss over 10 hours), indicating that endothelial cells were gradually losing barrier integrity over time. Changes in baseline TEER are common due to the use of primary human ECs which are more biologically appropriate and less uniform than cell lines and the TEER assay is sensitive to even small changes in culture conditions. To account for this, each plate was normalized to its internal saline control. Finally, our experimental designed is unable to determine the source of molecules that induce leak. Despite this variability, we were able to make meaningful comparisons between samples. Additionally, we acknowledge that our proteomic analysis is limited by the targeted nature of the panel, which includes only 45 proteins. As such, we are incompletely characterizing the biological response.

Emerging research in PARDS has identified distinct endotypes, or groups of patients defined by unique biological pathways and clinical outcomes(40). We are unable to analyze by previously defined hyper- or hypo-inflammatory endotypes(4, 5) because we did not assay for sTNFR1. The TEER analysis by clinical subgroups is complicated by high heterogeneity, as indicated by clear groupings and values above and below zero in the violin plots for all ABC comparisons. While we were unable to analyze ABC based on previously described endotypes in our study, expanding our current understanding of these endotypes could improve patient characterization and guide the development of targeted therapeutics.

Future Studies

Future studies should explore subgroup analyses of host response (i.e. hyper-inflamed vs immunoparalyzed), longer timepoints to assess repair, multicenter validation given the simplicity of TBAL collection, and broader proteomic profiling to distinguish markers from mediators of pulmonary capillary leak.

Conclusions

This study presents an integrated analysis of paired, serial plasma and TBAL samples to evaluate endothelial barrier function in critically ill children. Using functional and targeted proteomics, we identified compartment-specific and time-dependent effects on endothelial barrier function. Further studies are needed to elucidate the molecular pathways driving capillary leak in PARDS.