PD-L1 and PD-1 Are Associated with Clinical Outcomes and Alveolar Immune Cell Activation in Acute Respiratory Distress Syndrome

Abstract

The relationship between the PD-L1 (Programmed Death-Ligand 1)/PD-1 pathway, lung inflammation, and clinical outcomes in acute respiratory distress syndrome (ARDS) is poorly understood. We sought to determine whether PD-L1/PD-1 in the lung or blood is associated with ARDS and associated severity. We measured soluble PD-L1 (sPD-L1) in plasma and lower respiratory tract samples (ARDS1 [n = 59] and ARDS2 [n = 78]) or plasma samples alone (ARDS3 [n = 149]) collected from subjects with ARDS and tested for associations with mortality using multiple regression. We used mass cytometry to measure PD-L1/PD-1 expression and intracellular cytokine staining in cells isolated from BAL fluid (n = 18) and blood (n = 16) from critically ill subjects with or without ARDS enrolled from a fourth cohort. Higher plasma concentrations of sPD-L1 were associated with mortality in ARDS1, ARDS2, and ARDS3. In contrast, higher concentrations of sPD-L1 in the lung were either not associated with mortality (ARDS2) or were associated with survival (ARDS1). Alveolar PD-1^POS T cells had more intracellular cytokine staining than PD-1^NEG T cells. Subjects without ARDS had a higher ratio of PD-L1^POS alveolar macrophages to PD-1^POS T cells than subjects with ARDS. We conclude that sPD-L1 may have divergent cellular sources and/or functions in the alveolar versus blood compartments, given distinct associations with mortality. Alveolar leukocyte subsets defined by PD-L1 or PD-1 cell-surface expression have distinct cytokine secretion profiles, and the relative proportions of these subsets are associated with ARDS.

Clinical Relevance

We found that higher levels of sPD-L1 (soluble Programmed Death-Ligand 1) in the plasma were robustly associated with mortality in three different cohorts of patients with acute respiratory distress syndrome (ARDS). In contrast, higher levels of sPD-L1 in the lower respiratory tract were either not associated with mortality or were associated with survival. PD-1 was highly upregulated in alveolar T cells with the highest inflammatory cytokine secretion. A low ratio of PD-L1–expressing alveolar macrophages to PD-1–expressing T cells was associated with ARDS. Blockade of the PD-L1/PD-1 interaction in alveolar macrophage/T-cell cocultures was associated with higher levels of inflammatory cytokine secretion. Together, these findings suggest the PD-L1/PD-1 pathway may play different roles in the lung compared with blood compartments, that alveolar leukocyte subsets defined by PD-L1 or PD-1 have distinct functions, and the relative proportions of these subsets are associated with ARDS.

Acute respiratory distress syndrome (ARDS) is present in ∼10% of all patients admitted to an ICU worldwide and is associated with an in-hospital mortality rate ranging from 25% to 45% (1). The pathobiology of ARDS is characterized by dysregulated inflammation leading to increased alveolar barrier permeability and impaired fluid clearance (2, 3). A better understanding of the mechanisms that regulate the inflammatory response in patients with ARDS may lead to therapies that improve clinical outcomes.

The PD-L1 (Programmed Death-Ligand 1)/PD-1 immune checkpoint pathway plays a key role in regulating inflammation; however, its role in ARDS is not well characterized. The binding of PD-L1 on antigen-presenting cells to PD-1 on T cells promotes a self-tolerant state by inhibiting T-cell cytokine production, proliferation, and antigen-specific responses (4). PD-L2 is another ligand for PD-1, although its expression on immune cells is much more restricted than PD-L1. One of the first reported phenotypes of PD-L1 or PD-L2 knockout mice exposed to naive CD4⁺ T cells was fatal pneumonitis (5), and the lung is one of the most common sites for immune-related adverse clinical events associated with PD-L1/PD-1 checkpoint blockade (6–8). In murine models, PD-L1^POS alveolar macrophages (AMs) strongly suppress cytotoxic CD8⁺ T cells (9). We have robustly identified an AM subset in humans that is similarly classified by high PD-L1 cell-surface expression (AM-PD-L1^POS). We have shown AM-PD-L1^POS represents ∼90% of AMs in healthy participants but constitutes only 10–20% of AMs in subjects supported on invasive mechanical ventilation (10, 11). Collectively, these findings suggest a high proportion of the PD-L1 ligand to cells containing the PD-1 receptor might limit injurious inflammation in the lung.

Despite evidence suggesting engagement of PD-L1 to PD-1 in the lung may limit pneumonitis, higher levels of PD-L1 and/or PD-1 in the peripheral blood compartment have mostly been associated with worse outcomes in critically ill patients. In patients with sepsis, increased cell-surface expression of PD-L1 on peripheral blood monocytes is associated with worse outcomes (12, 13). Neutrophils expressing PD-1 have also been shown to predict development of sepsis in patients who present to the hospital with suspected infection (14). There are soluble forms of PD-L1, PD-L2, and PD-1 (sPD-L1, sPD-L2, and sPD-1, respectively) that are generated from splice variants and cleavage products (4). Some studies (15–21), but not all (22, 23), have identified associations between higher levels of plasma sPD-L1 and worse clinical outcomes in critically ill patients. Factors such as timing of sampling, measurement methods, and distinct patient characteristics have likely contributed to these conflicting reports.

We sought to clarify whether lung and blood checkpoint protein expression is associated with ARDS and related severity in multiple, well-characterized cohorts of critically ill patients. We hypothesized checkpoint pathway activation in the lung is associated with decreased inflammation and improved clinical outcomes in ARDS. Some of the results of this study have been reported in an abstract (24).

Methods

Study Overview

Full details of the study population, soluble ligand measurements, cytometric analyses, coculturing experiments, and statistical analyses are included in the data supplement. This study is an analysis of biospecimens and clinical data collected from subjects representing four cohorts (ARDS1, ARDS2, ARDS3, and Mass Cytometry) that were recruited from six ICUs. The University of Washington and Benaroya Research Institute Institutional Review Boards approved all studies.

Soluble Ligand Measurements

Endotracheal aspirate (ETA), BAL fluid (BALF), and plasma levels of sPD-L1, sPD-1, and sPD-L2 were measured using an electrochemiluminescence immunoassay per the manufacturer’s instructions (MSD catalog numbers: F214C; F214D; F214A). We did not measure sPD-1 or sPD-L2 in ARDS1. The limits of detection and intraindividual/interplate coefficients of variation are reported Tables E1 and E2 in the data supplement.

Cytometric Measurements

We analyzed alveolar cells and peripheral blood mononuclear cells (PBMCs) using mass cytometry as previously described (10, 20). Cells were treated with a stimulation condition (PMA-Ionomycin at 50/500 ng/ml, LPS at 1 μg/ml, or media only) and incubated for 4 hours at 37°C. We used the Helios system to generate flow cytometry standard files for analysis. Figure E1 displays representative gating of cell populations.

AM and T-Cell Coculturing Experiments

Leukocytes isolated from BALF were thawed from cryopreserved samples or obtained from a fresh sample the same day as BALF collection. AMs were purified from BALF by isolating the cells that adhered to plastic after culturing in serum-free media for 1 hour. We isolated CD4⁺ T cells from cryopreserved PBMC samples obtained from a single donor using a magnetic untouched CD4⁺ T-cell isolation kit (Easy Sep). These CD4⁺ T cells were cocultured with isolated AMs at an AM:T cell ratio of 1:1 or 2:3 under three experimental conditions incorporating PD-L1 blockade and controls. We collected supernatants after 48 hours of culturing and measured IFN-γ and TNF-α concentrations using an electrochemiluminescence immunoassay per the manufacturer’s instructions (MSD catalog number: K15093K-1).

Statistical Analysis

Soluble ligand concentrations were log₂ transformed to facilitate parametric statistical analyses. We estimated the relative risk of mortality per unit change in soluble ligand levels using Poisson regression. The point estimates are expressed as relative risk of mortality per doubling of soluble ligand level. We used multiple linear regression with robust standard error estimates to test for associations between soluble ligand levels and ventilator-free days (VFDs). The point estimates are expressed as a β, which should be interpreted as the change in VFDs per doubling of soluble ligand level. We used univariate statistics to compare cytometric data between groups.

Results

Study Population

We quantified soluble checkpoint ligand concentrations in biospecimens collected from four separate ARDS cohorts (Table 1). The baseline demographics were similar between the cohorts (Table 2). Almost all subjects in ARDS1 had pneumonia, primarily from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and had higher Acute Physiology and Chronic Health Evaluation III scores and lower arterial oxygen pressure/fraction of inspired oxygen ratios than subjects in ARDS2 or ARDS3. Plasma levels of sPD-L2 were orders of magnitude higher than sPD-L1 or sPD-1 levels (Figure 1). sPD-L1, sPD-L2, and sPD-1 plasma levels were moderately correlated (Figure E2).

Table 1.

Description of Study Cohorts

Characteristic	ARDS1 (n = 59 ETA) (n = 44 Plasma)	ARDS2 (n = 78 BALF) (n = 78 Plasma)	ARDS3 (n = 149 Plasma)	Mass Cytometry (n = 18 BALF) (n = 16 PBMCs)
Study design	Cohort	Cohort	Cohort	Case–Control
Medical centers	HMC, UWMC, VM	HMC, UWMC, VT, BO, TO	HMC	HMC
Population	ARDS (Berlin)	ARDS (Berlin)	ARDS (Berlin)	Cases–ARDS (Berlin) Controls–invasive mechanical ventilation without ARDS
Inclusion criteria	Suspicion for severe COVID-19 (viral syndrome symptoms + ICU admission + supplemental oxygen or radiographic infiltrates)	ARDS from any risk factor	ARDS from medical risk factor (trauma excluded)	Suspicion for ventilator-associated pneumonia
Enrollment era	2020–2021	2006–2008	2006–2010	2017–2019
Sampling	ETA collected during ICU stay; paired ETA and plasma collected for 44 patients	Paired BALF and plasma collected within 48 h of enrollment (ICU admission)	Plasma collected within 48 h of enrollment (ICU admission)	Alveolar leukocytes and PBMCs collected at the time of clinical bronchoscopy. Five of the samples are paired.
Measurements	sPD-L1	sPD-L1, sPD-L2, and sPD-1	sPD-L1, sPD-L2, and sPD-1	Cytometric measurements

Definition of abbreviations: ARDS = acute respiratory distress syndrome; BALF = BAL fluid; Berlin = Berlin definition of ARDS [See Ref. (25)]; BO = St. Alphonsus Medial Center (Boise, ID); COVID-19 = coronavirus disease; ETA = endotracheal aspirate; HMC = Harborview Medical Center (University of Washington); PBMCs = peripheral blood mononuclear cells; sPD-1 = soluble PD-1; sPD-L1 = soluble Programmed Death-Ligand 1; TO = St. Michael’s Hospital (University of Toronto); UWMC = University of Washington Medical Center; VM = Virginia Mason Franciscan Hospital (Seattle, WA); VT = Fletcher Allen Health Care (University of Vermont).

Table 2.

Characteristics of ARDS Cohorts

Characteristic	ARDS1 (n = 59 ETA) (n = 44 Plasma)	ARDS2 (n = 78 BALF) (n = 78 Plasma)	ARDS3 (n = 149 Plasma)
Demographics and comorbidities
Age, yr, mean ± SD	56 ± 14	50 ± 17	54 ± 17
Sex (male)	75	62	71
Race
American Indian	3	3	4
Asian/Pacific Islander	12	5	12
Black/African American	8	3	12
White	63	90	67
Unknown	14	0	6
Coexisting disease
Diabetes	44	22	26
Cirrhosis	2	8	11
Chronic renal insufficiency	10	3	7
Characteristics at sampling
Hospital admission-to-sampling interval, d, median (interquartile range)	7 (3–13)	<2	<2
Sepsis*	100†	63‡	85§
Pneumonia (non–COVID-19)*	5	38	42
Pneumonia (COVID-19)*	93	0	0
Trauma*	0	38	0
Other*	8	12	28
ARDS severity
Mild (P/F, 200–300)	18	29	21
Moderate (P/F, 100–200)	36	58	37
Severe (P/F < 100)	46	13	42
APACHE III score (mean ± SD)	93 ± 26	79 ± 26	65 ± 28

Definition of abbreviations: APACHE III = Acute Physiology and Chronic Health Evaluation; P/F = arterial oxygen pressure/fraction of inspired oxygen ratio; SIRS = systemic inflammatory response syndrome.

Data are presented as percentages unless otherwise noted.

^*Diagnoses are not exclusive in each cohort.

^† Sepsis defined by Sequential Organ Failure Assessment score ⩾ 2 and the presence of probable or suspected infection (SEPSIS-3).

^‡ Sepsis defined by two or more of four SIRS criteria and positive bacterial culture in the 48 hours before study enrollment or documentation of “sepsis” or “septic shock” by ICU providers in the medical record (SEPSIS-2).

^§ Sepsis defined by two or more of four SIRS criteria and the presence of probable or suspected infection as defined by the administration of broad-spectrum antibiotics within 3 days of meeting SIRS criteria and continued for at least 5 days or until death or discharge (SEPSIS-2).

Figure 1.

Lung and plasma concentrations of sPD-L1 (soluble Programmed Death-Ligand 1), sPD-1 (soluble PD-1), and sPD-L2. Shown are the log₂-transformed levels of ETA, BALF, and plasma sPD-L1, sPD-L2, and sPD-1 from the ARDS1, ARDS2, and ARDS3 cohorts. Displayed are the individual values, mean, and standard deviation for each set.

Plasma and Lung sPD-L1 Levels Have Distinct Associations with Mortality

The relationship between sPD-L1 and mortality was distinct, depending on whether it was measured in the plasma versus the lower respiratory tract. Higher plasma levels of sPD-L1 were significantly associated with mortality in ARDS1, ARDS2, and ARDS3 (Figure 2 and Table E3). Higher sPD-L1 plasma levels were also associated with fewer VFDs in ARDS2 and ARDS3, but not in ARDS1 (Table E4). Plasma sPD-L2 was not associated with mortality in ARDS2 or ARDS3 (Table E5); however, higher levels of plasma sPD-1 were associated with mortality in ARDS3 (Table E6).

Figure 2.

Associations between sPD-L1 levels and mortality. We collected lung (endotracheal aspirates in ARDS1 and BALF in ARDS2) and plasma (ARDS1, ARDS2, and ARDS3) samples from subjects with ARDS and measured sPD-L1 levels. Each plot displays the individual values, mean, and SD of log₂-transformed sPD-L1 levels. P values were generated from relative risk regression. The unadjusted and adjusted estimates, as well as the covariates included in each model, are shown in Table E3. Figure E3 displays the results of this analysis for ARDS1 restricted to only subjects with paired lung/blood samples. All samples collected in ARDS2 were paired lung/blood specimens.

In contrast to our findings in plasma samples, higher levels of sPD-L1 measured in the lung were either not associated with mortality or were associated with survival. Higher levels of sPD-L1 measured from the lung were associated with survival in ARDS1 and not associated with mortality in ARDS2 (P = 0.36) (Figure 2 and Table E3). In a sensitivity analysis of ARDS1 limited to subjects with paired lung/blood sampling, we likewise observed distinct associations between sPD-L1 and mortality based on lung versus blood sampling (Figure E3). Lung sPD-L1 levels were associated with more VFDs in ARDS1, but not in ARDS2 (Table E4). Lung levels of sPD-L2 or sPD-1 were not associated with mortality (Tables E5 and E6). Together, our finding of distinct associations between sPD-L1 levels and mortality based on lung versus blood sampling suggests sPD-L1 may have divergent cellular sources and/or functions in these organ systems.

PD-L1/PD-1 Cell-Surface Expression Identifies Alveolar Leukocyte Populations That Associate with ARDS

We measured the percentage of AM-PD-L1^POS, CD4⁺PD-1^POS, and CD8⁺PD-1^POS cells as a proportion of total alveolar leukocytes in a case–control analysis of mechanically ventilated subjects with and without ARDS to determine whether alveolar cell populations based on PD-L1 or PD-1 cell-surface expression were associated with more severe lung injury (i.e., ARDS). Patient characteristics at the time of BALF sampling are shown in Table 3 (Mass Cytometry Cohort). Baseline demographics, severity of illness as measured by Acute Physiology and Chronic Health Evaluation III score, the proportion of patients treated with antibiotics, and the interval of time between intubation and BALF sampling was similar between subjects with and without ARDS.

Table 3.

Subject Characteristics of Mass Cytometry Cohort

Characteristic	No ARDS (n = 8)	ARDS (n = 10)
Demographics and comorbidities
Age, yr, mean ± SD	38 ± 17	41 ± 12
Sex, male	88	70
Race
Asian/Pacific Islander	13	0
Black/African American	25	20
White	63	80
Coexisting disease
Diabetes	0	25
Cirrhosis	0	0
Chronic renal insufficiency	0	10
Admission diagnosis
Neuro/airway protection, n (%)	4 (50)	1 (10)
Trauma, n (%)	2 (25)	5 (50)
Pneumonia/aspiration, n (%)	1 (13)	3 (30)
Other, n (%)	1 (13)	1 (10)
Characteristics at sampling
ARDS, n (%)	0 (0)	10 (100)
IMV-to-sampling, d, median (IQR)	8 (6–12)	7 (6–8)
Bacteria pneumonia,* n (%)	7 (88)	9 (90)
P/F, median (IQR)	282 (251–313)	184 (147–198)
S/F, median (IQR)	281 (192–347)	217 (192–238)
Oxygenation Index, median (IQR)	3.5 (2.9–6.0)	8.7 (6.7–11.6)
APACHE III, median (IQR)	87 (71–98)	83 (50–103)

Definition of abbreviations: APACHE III = Acute Physiology and Chronic Health Evaluation; ARDS = acute respiratory distress syndrome; IMV = invasive mechanical ventilation; IQR = interquartile range; P/F = arterial oxygen pressure/fraction of inspired oxygen ratio; S/F = oxygen saturation as measured by pulse oximetry/fraction of inspired oxygen ratio.

Data are presented as percentages unless otherwise noted.

^*Treated with antibiotics for bacterial pneumonia.

We identified alveolar cell populations by PD-L1 or PD-1 expression through both manual gating (Figure 3A) and unsupervised analyses (Figures 3B, 3C, and E4). Cell-surface expression of PD-L1 was significantly higher on AMs than alveolar monocytes and T cells (Figure 3D). Conversely, PD-1 expression was significantly higher on alveolar T cells than AMs. Almost all AM-PD-L1^POS cells had high expression of human leukocyte antigen - DR isotype (Figure 3C).

Figure 3.

PD-L1/PD-1 cell-surface expression identifies alveolar leukocyte populations that associate with ARDS. We used mass cytometry to analyze alveolar leukocytes isolated from BALF from subjects with (n = 10) or without (n = 8) ARDS. (A) Representative cell population definitions (“gating”) for alveolar leukocytes. Pregating on live, single, CD45^POS cells is shown in Figure E1. We display the individual gating for participants X, Y, and Z, each of whom has different proportions of alveolar macrophage (AM)-PD-L1^POS cells. (B) Uniform manifold approximation and projection (UMAP) plot displaying alveolar leukocyte clustering based on cell-surface protein expression. The cell-surface proteins used for clustering are shown in Figure E4. Color designates assignment of cells to one of the 12 clusters identified by DISCOVE-R. We annotated the clusters based on the normalized expression of each cell-surface marker per cluster (Figure E4). Cells from all patients (n = 18), irrespective of ARDS status, were included in the clustering. (C) UMAPs colored based on marker intensity for each cell-surface protein. (D) The panels show the percentage of PD-L1^POS or PD-1^POS cells as a proportion of their respective parent populations. Shown are individual percentages (dots) and lines connecting individual subjects. Cells from all patients (n = 18), irrespective of ARDS status, were included in these analyses. P values were calculated with paired Wilcoxon tests. (E) The panels show percentages of cell populations as a proportion of their respective parent populations (dots), medians (lines), and interquartile ranges (error bars). P values were calculated with Mann-Whitney tests comparing subjects with or without ARDS. (F) Individual dots show the ratio of the proportion of AM-PD-L1^POS to the proportion of alveolar CD4⁺PD-1^POS or CD8⁺PD-1^POS cells per subject, respectively. Displayed are medians (lines) and interquartile ranges (error bars) for the ratios using either percentage populations or the absolute cell counts for each cell population. The absolute cell numbers for the parent populations are shown in Figure E5. P values were calculated with Mann-Whitney tests comparing subjects with or without ARDS.

The percentage of CD4⁺PD-1^POS alveolar T cells as a proportion of their respective parent population was higher in subjects with ARDS than in critically ill subjects without ARDS (Figure 3E). The median proportion of AM-PD-L1^POS cells in ARDS was 35.8%, compared with 49.8% in subjects without ARDS; however, this difference was not statistically significant (P = 0.21). Subjects with ARDS had a significantly lower ratios of AM-PD-L1^POS to CD4⁺PD-1^POS or CD8⁺PD-1^POS T cells than subjects without ARDS (Figure 3F). The absolute number of cells for each alveolar leukocyte population analyzed was not significantly different between subjects with or without ARDS, suggesting that differences in BALF sample return were not significantly biasing our cell proportion findings (Figure E5). We next determined whether AM-PD-1^POS and/or PD-L1^POS T cells were associated with ARDS, because PD-L1 can be expressed on T cells and PD-1 can be expressed on tumor-associated macrophages (26, 27). The proportions of AM-PD-1^POS, CD4⁺PD-L1^POS, or CD8⁺PD-L1^POS cells were not associated with ARDS (Figure E6). Overall, these findings are consistent with the sPD-L1 data above, suggesting that relatively higher ratios of PD-L1– to PD-1–expressing cells may be associated with less severe lung injury (i.e., non-ARDS vs. ARDS).

PD-L1 and PD-1 Are Associated with Alveolar Immune Cell Cytokine Secretion

To examine the cytokine production profiles of AM-PD-L1^POS, CD4⁺PD-1^POS, and CD8⁺PD-1^POS alveolar cells, we incubated cells with LPS or PMA-ionomycin for 4 hours and measured intracellular cytokine staining for IL-8, TNF-α, and IFN-γ at the single-cell level by mass cytometry. We included subjects with and without ARDS in this analysis because our primary question was whether PD-L1- or PD-1-positive and –negative alveolar cell populations are functionally distinct.

PD-1^POS alveolar T-cell populations had significantly higher proportions of cytokine-producing cells than PD-1^NEG alveolar T cells (Figure 4A). In contrast, AM-PD-L1^POS had a significantly lower percentage of IL-8–producing cells than AM-PD-L1^NEG. We observed that sPD-L1 BALF levels in subjects with ARDS were positively correlated with the proportion of AM-PD-L1^POS cells but inversely correlated with the proportion of CD8⁺ PD-1^POS cells (Figure 4B).

Figure 4.

PD-L1/PD-1 is associated with alveolar immune cell cytokine secretion. We stimulated alveolar immune cells with either LPS or PMA-ionomycin and compared intracellular cytokine staining between AMs, CD4⁺, and CD8⁺ T-cell populations based on PD-L1/PD-1 expression. Representative gating is shown in Figure E1. (A) x-axis displays CD4⁺, CD8⁺, or AMs classified as PD-1^POS/NEG or PD-L1^POS/NEG, respectively. y-axis displays the percentage CD4⁺, CD8⁺, or AMs positively staining for each respective cytokine (percentages are reported as the difference between stimulated and unstimulated [baseline] samples per individual). The lines connect populations from the same individual. Cells from all patients (n = 18), irrespective of ARDS status, were included in these analyses. P values were calculated with paired Wilcoxon tests comparing PD-1^POS versus PD-1^NEG cells or PD-L1^POS versus PD-L1^NEG, respectively, per subject. (B) The panels show the relationship between the proportion of CD4⁺PD-1^POS, CD8⁺PD-1^POS, or AM-PD-L1^POS cells and sPD-L1 levels in the BALF in subjects with ARDS (n = 10). Shown are individual values (dots) linear regression best-fit line (straight lines). r = Pearson’s correlation coefficient. (C) In separate experiments, we cocultured AMs from patients with hypoxemic respiratory failure (n = 3) with peripheral blood T cells from a single healthy donor for 48 hours under three experimental conditions: 1) anti-CD3 antibody (stimulant); 2) anti-CD3 antibody + anti–PD-L1 antibody (treatment); and 3) anti-CD3 antibody + isotype control. We measured IFN-γ and TNF-α levels in the cell culture supernatants at the end of the experiment. Lines connect measurements from an individual subject. AMs from the individual with the highest IFN-γ and TNF-α levels were cultured directly after sample collection. The AMs from the other two individuals were analyzed after cryopreservation. IFN-γ levels were numerically higher in treatment versus control in two of three experiments. TNF-α levels were numerically higher in treatment versus control in three of three experiments.

We performed ex vivo coculture experiments to determine whether blockade of the PD-L1/PD-1 interaction on AMs/T cells would lead to immune cell activation. We cocultured AMs isolated from critically ill mechanically ventilated subjects (n = 3) with CD4⁺ T cells from a healthy donor for 48 hours under three experimental conditions: 1) anti-CD3 antibody (stimulant); 2) anti-CD3 antibody + anti–PD-L1 antibody; and 3) anti-CD3 antibody + isotype control. The levels of INF-γ and TNF-α in the culture supernatants were higher in the anti–PD-L1 condition versus isotype control in two of three and in three of three subjects, respectively (Figure 4C). Future studies are necessary to validate these findings on a larger number of individuals. However, our data suggest that PD-1 is highly upregulated on the most activated alveolar T cells in terms of inflammatory cytokine secretion and that the PD-L1/PD-1 interaction between AMs and T cells plays a functional role in regulating alveolar T-cell activation.

PD-1 Expression Is Higher on Alveolar Compared with Peripheral Blood Leukocytes

We tested whether PD-L1/PD-1 cell-surface expression on PBMCs was different in subjects with and without ARDS, because migration of blood monocytes and lymphocytes from the peripheral circulation to the alveolar space is a key feature of ARDS pathogenesis (2, 3). Subject characteristics are shown in Table E7. The percentages of PD-L1^POS monocytes or PD-1^POS CD4⁺ and CD8⁺ peripheral T cells as a proportion of their respective parent populations were not different between subjects with or without ARDS (Figure E7A). Likewise, the ratio of monocyte–PD-L1^POS to CD4⁺PD-1^POS or CD8⁺PD-1^POS cells was not distinct between subjects with and without ARDS (Figure E7B).

In a subset of patients with paired alveolar and peripheral samples, we found that intracellular cytokine staining was higher in alveolar T cells versus peripheral blood T cells (Figure 5A). The proportion of PD-1–expressing cells was higher in alveolar compared with peripheral blood samples, but there was not a significant difference between blood monocyte and AM-PD-L1 cell-surface expression (Figure 5B). These findings provide further evidence that PD-L1/PD-1 activity in the lung is distinct compared with peripheral blood in patients with ARDS.

Figure 5.

PD-1 expression is higher on alveolar compared with peripheral blood leukocytes. Paired alveolar cells from BALF were simultaneously collected from five subjects who had peripheral blood mononuclear cells analyzed. We stimulated cells with either LPS (1 μg/ml) or PMA-ionomycin (50/500 ng/ml) for 4 hours and then assessed intracellular cytokine staining. (A) Individual dots represent the percentage of a cell population as a proportion of their respective parent population in paired samples. Lines connect samples collected from the same subject. P values were calculated with paired t tests comparing peripheral blood versus alveolar samples for each subject. (B) We compared the percentage of CD4⁺PD-1^POS, CD8⁺PD-1^POS, and AM-PD-L1^POS cells as a proportion of each of their respective parent populations in peripheral blood compared with alveolar samples. Lines connect samples collected from the same subject. P values were calculated with paired t tests comparing peripheral blood versus alveolar samples for each subject. PBM = peripheral blood monocytes.

Discussion

Our findings demonstrate that alveolar leukocyte subsets defined by PD-L1/PD-1 cell-surface expression have distinct cytokine secretion profiles, and the relative proportions of these subsets are associated with ARDS. Blocking PD-L1/PD-1 interactions in AM/T cell cocultures leads to higher inflammatory cytokine secretion. We also determined that higher levels of the soluble ligand sPD-L1 in the plasma were robustly associated with mortality, but higher lung sPD-L1 levels were not associated with mortality and might, in fact, be associated with survival. Taken together, these findings provide evidence for a novel ARDS immunologic model, whereby cytokine-producing alveolar T cells highly express PD-1, but AM-PD-L1^POS or lung sPD-L1 ligands may potentially limit development of ARDS and associated severity (Figure 6).

Figure 6.

A proposed schematic for cell-surface and soluble PD-L1 in ARDS. (A) Observations from association studies of alveolar macrophages and T cells during steady state. (B) Observations from association studies of alveolar macrophages and T cells during ARDS. (C) Potential functional role of PD-L1/PD-1 in the alveolar space. This figure was generated using BioRender.

Our finding that associations between PD-L1/PD-1 and clinical outcomes are distinct in the lung versus systemic compartments has significant implications for our understanding of ARDS pathobiology and future biospecimen collection protocols. We found that higher levels of sPD-L1 in the lung are associated with survival in ARDS1, although this finding did not replicate in ARDS2 (P = 0.36) (Figure 2). Potential reasons for not detecting a significant association in ARDS2 include the dilutional effect inherent in BALF versus ETA sampling (Figure 1), the less severely ill patient population in ARDS2 compared with ARDS1 (Table 2), a more homogenous patient population for ARDS1 (almost all subjects in ARDS1 had severe coronavirus disease [COVID-19] pneumonia), and differences in the duration of hospitalization before sample collection between ARDS1 and ARDS2. The finding that sPD-L1 levels in the lung may be associated with improved ARDS outcomes would be consistent with preclinical studies demonstrating that sPD-L1 reduces T-cell proliferation in vitro (28) and mediates resistance to PD-L1 blockade in cancer immunotherapy (29–31). In a murine model of Pseudomonas aeruginosa pneumonia, Xu and colleagues found that intravenous administration of sPD-L1 improved survival and markers of lung injury (22). They observed very high levels of the labeled sPD-L1 in the alveolar space of the treated mice, suggesting alveolar sPD-L1 may limit inflammation by binding with alveolar PD-1 ligands.

The findings in the lung contrast with what we observed in the systemic circulation. We detected a robust association between higher plasma sPD-L1 levels and mortality in all three cohorts with plasma samples available as well as fewer VFDs in ARDS2 and ARDS3 (Figure 2 and Tables E3 and E4). These associations between higher plasma sPD-L1 levels and worse outcomes are consistent with studies in sepsis (15, 16) and respiratory viral infection (17–19) but are different from the Xu and colleagues study above (which also analyzed human participants [n = 30] with direct ARDS [22]). Xu and colleagues found higher plasma levels of sPD-L1 were associated with survival and correlated with better oxygenation index. The baseline characteristics for subjects in the Xu and colleagues study and our cohorts were relatively similar, although their study measured sPD-L1 through a different method. The reasons for the discrepant findings are not clear; however, we think our relatively large study (n = 271) of well-characterized patients with ARDS pushes the balance of evidence toward the conclusion that plasma sPD-L1 levels are associated with worse outcomes in ARDS. We speculate higher systemic levels of sPD-L1 may contribute to excessive “immune paralysis” that has been shown to drive mortality in critical illness (32). Alternatively, higher plasma concentrations of sPD-L1 in participants who ultimately died might simply reflect inadequate counterregulatory signaling in the most severely ill patients. Possible explanations for the inverse associations between clinical outcomes and sPD-L1 levels in the lung versus systemic circulation include distinct cellular sources of sPD-L1 (33), post-translational processing in different tissues (34), or other mechanisms of alveolar compartmentalization of soluble proteins (35). These findings build on prior work from our group (33, 36, 37) and others (35) that the relationship between lung and systemic immune responses and clinical outcomes can be highly distinct, perhaps most importantly in patients with direct lung injury.

To our knowledge, our study is the first to report sPD-L2 plasma and BALF concentrations in cohorts of patients with ARDS. We identified relatively high concentrations of sPD-L2 in the blood and BALF of patients with ARDS (Figure 1), although levels of sPD-L2 were not associated with mortality (Table E5). PD-L2 is constitutively expressed on endothelial cells (38) as well as dendritic cells and macrophages, although its expression appears to be highly dependent on the local cytokine milieu (4, 39). PD-L2 binds to PD-1 with more affinity than PD-L1 (40, 41). sPD-L2 levels have been reported at high plasma concentrations in patients with malignancies (42, 43), lupus (44), and COVID-19 (45) and healthy donors (42–45), although the primary cellular source of sPD-L2 remains unclear. More investigation into the cellular source and role of sPD-L2 in acute inflammatory conditions of the lung is merited, given the differences in cellular distribution between PD-L2 and PD-L1, their unique as well as overlapping functions, and the very high alveolar and systemic levels of sPD-L2 we observed in patients with ARDS.

We previously identified the AM-PD-L1^POS population in healthy and mechanically ventilated subjects (10, 11). In the present study, we have expanded on this prior work by determining that AM-PD-L1^POS has lower cytokine secretion than AM-PD-L1^NEG, PD-L1 cell-surface expression is significantly higher on AMs than alveolar monocytes, and the primary receptor target for PD-L1 (PD-1^POS T cells) is associated with ARDS (Figure 3). PD-1 has been shown to be upregulated on activated T cells in various tissues (4); however, this is the first study, to our knowledge, that has demonstrated human alveolar PD-1^POS T cells are associated with ARDS and have higher intracellular cytokine staining than alveolar PD-1^NEG T cells. This finding supports the hypothesis that PD-1 may be upregulated on highly activated T cells at sites of inflammation as a local feedback regulatory response mechanism. We observed that a relatively low ratio of AM-PD-L1^POS to PD-1^POS T-cell populations is associated with ARDS (Figure 3F) and identified increased cytokine secretion when AMs were cocultured with T cells in the presence of PD-L1 blockade. These findings are consistent with a study by Sun and colleagues that demonstrated AM-PD-L1^POS strongly suppresses cytotoxic T-cell cytokine secretion by cis- and trans-interacting with CD80 and PD-1, respectively, in coculture experiments using alveolar cells collected from mice (9). Collectively, these findings imply that a relative reduction in the ratio of AM-PD-L1^POS cells to highly inflammatory PD-1^POS alveolar T cells may contribute to the excessive alveolar inflammation that is a hallmark of ARDS pathobiology.

Our mass cytometry data provide additional evidence that PD-L1 cell-surface expression may identify a “homeostatic” AM population and that “recruited” alveolar monocytes are predominantly PD-L1^NEG in humans. PD-L1 gene expression has been shown to be higher in inflammatory recruited monocytes/macrophages than resident AMs in animal studies (46, 47), and PD-L1 is highly upregulated on macrophages exposed to LPS stimuli in vitro (38). However, our prior findings using cellular indexing of transcriptomes and epitopes demonstrated that PD-L1 cell-surface and CD274 (PD-L1) gene expression cannot be used interchangeable to identify mature AMs (11). Our observation that PD-L1 cell-surface expression is significantly higher on AMs versus alveolar monocytes (Figure 3D), our prior work identifying AM-PD-L1^POS on almost all AMs collected from healthy donors (10), and preclinical evidence demonstrating constitutive PD-L1 cell-surface expression on resident AMs (9) provide robust evidence that AM-PD-L1^POS is a “homeostatic” AM population. Future mechanistic work is necessary to better clarify how PD-L1 gene expression regulates differentiation and function in blood monocytes, recruited monocytes/AMs, and resident AMs at different stages of development.

Our study has multiple strengths. First, we analyzed a relatively large number of lower respiratory tract samples collected from critically ill patients with ARDS. Access to these scarce biospecimens allowed us to detect associations between alveolar cellular/molecular signatures and clinical outcomes. Second, we used a standardized protocol to measure sPD-L1 levels in our cohorts, which limited intraassay variability. Previous studies measuring sPD-L1 in subjects with sepsis or other inflammatory conditions have been limited to single cohorts of patients and have reported highly variable concentrations. Finally, we leveraged mass cytometry incorporating cell-surface and intracellular markers in our analysis to allow for both the identification and functional assessment of alveolar leukocyte populations.

Despite our study’s strengths, there are several limitations. Our cell-processing protocols were not designed to capture neutrophils. Several studies have identified neutrophil population marked by PD-L1/PD-1 expression (14, 48); however, we are unable to determine whether these neutrophil populations exist in our cohorts. Although we identified a significant association between higher sPD-L1 ETA levels and decreased mortality in ARDS1, these findings did not replicate in ARDS2. It is possible that biases introduced by the relationship between illness severity and timing of sample collection on the association between lung sPD-L1 and mortality maybe be distinct in ARDS1 and ARDS2. Finally, there were limited numbers of samples collected for the mass cytometry analyses and the coculturing experiments. Inferences based on the mass cytometry and coculturing data should be validated in more definitive future studies.

In conclusion, our finding of distinct associations between sPD-L1 levels in the lung versus plasma and clinical outcomes should inform future mechanistic and clinical study designs. We have provided multiple lines of evidence to support the hypothesis that an increased presence of PD-L1 in the lung may contribute to T-cell tolerance and decreased lung inflammation. The ability to therapeutically manipulate the PD-L1/PD-1 pathway makes it an attractive potential target for drug development in ARDS and other inflammatory conditions of the lung.