Plasma Soluble Intercellular Adhesion Molecule-1 Has a Central Role in Biomarker Network Analysis and Is Associated With Poor Outcomes in Two Distinct Pediatric Cohorts of Acute Respiratory Distress Syndrome and Acute Respiratory Failure

Abstract

OBJECTIVES:

Intercellular adhesion molecule-1 (ICAM-1) is a glycoprotein expressed on immune, endothelial, and epithelial cells. In the setting of inflammation, it becomes upregulated and spliced into a soluble form (soluble ICAM-1 [sICAM-1]). This study examined the association of sICAM-1 with clinical outcomes in two large pediatric cohorts with acute respiratory distress syndrome (ARDS) and acute respiratory failure (ARF) and examined the relationships between sICAM-1 and other protein biomarkers utilizing network analysis to contextualize its role in ARDS pathophysiology.

DESIGN:

Secondary analysis of prospective cohort studies.

SETTING:

Multicenter PICUs.

PATIENTS OR SUBJECTS:

Critically ill children with ARDS (Pediatric Acute Lung Injury [PALI], 2008–2014) and ARF (Coagulation and Fibrinolysis in Pediatric Insulin Titration Trial [CAF-PINT], 2012–2016).

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

sICAM-1 levels were measured from plasma collected within 72 hours of diagnosis. The primary outcome was in-hospital mortality, and secondary outcomes included multiple organ dysfunction and ventilator-free days. We constructed a biomarker correlation-based network that included sICAM-1 and 32 plasma biomarkers reflective of inflammation, endothelial and epithelial injury, and extracellular matrix degradation. Key biomarkers with centrality metrics in the top 10% (≥ 90th percentile) were defined as critical hubs within the network. The study included 214 children from PALI and 251 from CAF-PINT. In-hospital mortality was 18% and 14%, respectively. Baseline median oxygenation index ratios were 10 (interquartile range [IQR], 5.6–19.7) and 8.5 (IQR, 3.5–17.7). Higher plasma sICAM-1 was associated with in-hospital mortality, multiple organ dysfunction, and fewer ventilator-free days in each of the two cohorts (all p < 0.05). Tissue inhibitor of metalloproteinase-1 (composite centrality, 0.99), tumor necrosis factor receptor-1 (0.83), sICAM-1 (0.74), and interleukin-8 (0.74) were identified as network hubs.

CONCLUSIONS:

Elevated sICAM-1 levels were associated with poor outcomes in two separate cohorts of ARDS and ARF patients. Network analysis revealed sICAM-1 as a central hub, characterized by high centrality metrics. These findings underscore the multifaceted role of sICAM-1 in leukocyte transmigration, inflammation, and endothelial dysfunction and highlight its critical role in ARDS pathophysiology.

Acute respiratory distress syndrome (ARDS) is a heterogeneous syndrome associated with significant morbidity and mortality among critically ill children, accounting for 16–30% of PICU deaths (1–3). ARDS lacks targeted therapies, and prior treatment trials have been negative in part due to its heterogeneous pathobiology (4, 5). Plasma biomarkers have been useful for risk stratification, identifying disease subphenotypes (6–9), and deepening our understanding of disease pathogenesis and potential therapeutic targets (4, 10).

Intercellular adhesion molecule-1 (ICAM-1) is expressed on the vascular endothelium, immune cells, and alveolar epithelium (11–13) forming integral barriers at cellular junctions and mediating key cellular transduction signaling pathways (14–16). Exposure to pro-inflammatory cytokines upregulates ICAM-1 expression. Further, ICAM activation recruits circulating leukocytes to sites of inflammation, while concurrently increasing proteolytic cleavage of soluble ICAM-1 (sICAM-1) from cellular surfaces (17) and releasing it into circulation. sICAM-1 is also sheared directly from the surfaces of the alveolar epithelium and endothelium in ARDS, and sICAM itself activates alveolar macrophages and propagates a pro-inflammatory cytokine cascade leading to tissue injury (18, 19).

High levels of sICAM-1 discriminated patients with ARDS from those with hydrostatic edema and have been associated with higher mortality, prolonged length of mechanical ventilation, and severe multiple organ failure in adults with ARDS (20). Small pediatric studies have also reported association between high circulating sICAM-1 levels and poor clinical outcomes among children with acute lung injury (ALI) (21–23). However, like other single biomarker studies in ARDS, these studies cannot contextualize the role of sICAM within the large landscape of previously reported prognostic ARDS biomarkers. Network analysis has emerged as a useful quantitative tool to identify systems level molecular interactions and reveal molecules that may have high potential for therapeutic impact.

The primary objective of this study was to determine the association of early plasma sICAM-1 with clinical outcomes, including in-hospital mortality, multiple organ dysfunction, and ventilator-free days (VFDs) in a large multicenter pediatric ARDS cohort and replicate findings in an external multicenter large cohort of children with acute respiratory failure (ARF) requiring invasive mechanical ventilation (IMV).

Our secondary objective was to contextualize ICAM’s role as a putative factor in pediatric ARDS pathophysiology and examine sICAM-1 within a correlation-based network analysis of ARDS protein biomarkers of known pathobiological significance. Given the multifaceted pathobiology in ARDS (inflammation, endotheliopathy, and transcellular migration), we hypothesized that among children, with ARDS, sICAM-1 would be associated with worse clinical outcomes and also represent a key biomarker critical to the connectivity of the network.

METHODS

Patient Population

Pediatric Acute Lung Injury Cohort.

Children 30 days to 18 years old born greater than 36 weeks gestation, with pediatric ARDS were enrolled across five PICUs (e-Appendix, https://links.lww.com/CCM/H728). Plasma samples for sICAM-1 measurement were collected within 24 hours of meeting ARDS criteria.

Coagulation and Fibrinolysis in Pediatric Insulin Titration Trial Cohort.

Children 2 weeks to 18 years old with respiratory failure requiring IMV and hyperglycemia (two consecutive blood glucose measurements ≥ 150 mg/dL) were enrolled in a multicenter randomized controlled trial of tight glycemic control (e-Appendix, https://links.lww.com/CCM/H728). Plasma for sICAM-1 measurement was obtained within 72 hours of IMV initiation.

Data presented as a secondary analysis of preaccrued data. Both original studies were approved by an institutional review board (e-Appendix, https://links.lww.com/CCM/H728). Informed consent for enrollment was obtained from each patient or authorized surrogate. Procedures were followed in accordance with the ethical standards of the responsible committee and with the Helsinki Declaration of 1975.

Definitions

Categories of severity of oxygenation failure ranged from mild to severe according to consensus definitions (24). Oxygenation index (OI) was calculated using published definitions (e-Appendix, https://links.lww.com/CCM/H728). The development of severe organ dysfunction was assessed by using the validated Pediatric Logistic Organ Dysfunction-2 (PELOD-2 [25]) score, scored after sICAM-1 measurement and calculated as the worst individual score evaluated at serial time points (study days 1–7, days 14, 21, and 28) and censored at 28 days (26).

sICAM-1 Measurement

sICAM-1 levels were quantified from plasma in the Pediatric Acute Lung Injury (PALI) cohort (ng/mL) using Human InflammationMAP 1.0 multiplex immunoassays (Myriad RBM, Austin, TX) following the manufacturer’s instructions. sICAM-1 in the Coagulation and Fibrinolysis in Pediatric Insulin Titration Trial (CAF-PINT) cohort was measured using a custom multiplex magnetic bead assay (Luminex, R&D Systems, Minneapolis, MD) at the UCLA Immune Assessment Core. Other biomarkers were measured using a combination of single and multiplex enzyme-linked immunosorbent assay (e-Appendix, https://links.lww.com/CCM/H728) (27).

Outcomes

The primary outcome was all-cause in-hospital mortality. Secondary outcomes included: multiple organ dysfunction (as assessed by the PELOD-2 score), 28-day VFDs, and ICU length of stay (LOS; in days).

Statistical Analyses

Categorical variables were summarized with frequency (%). Continuous variables were summarized using median and interquartile range (IQR). sICAM-1 distribution was assessed using Shapiro-Wilk and, due to non-normal distribution, values were log₁₀-transformed for inclusion in the regression models.

Primary Outcome.

The association between sICAM-1 and mortality was evaluated using logistic regression. When univariate models were significant, multivariable models adjusted for age, sex, race, and severity of oxygenation failure (OI) as covariates, which were chosen because of their effect on biomarker levels (28). Odds ratios and 95% CIs were calculated per 1 sd increase in log biomarker levels. Detail on area under the receiver operating characteristic curve (AUROC) analysis, including use of principal component analysis (PCA) with the addition of other endothelial markers, and statistical test of interaction are described in the e-Appendix (https://links.lww.com/CCM/H728).

Secondary Outcomes.

Similar analyses were performed for clinical outcomes, including VFDs, ICU LOS, and PELOD score, using linear regression models with bootstrap SEs.

Coefficients with 95% CIs were reported, and p value of less than 0.05 was considered statistically significant. All analyses were conducted using Stata, Version SE17 (StataCorp, College Station, TX).

Network Analysis

We leveraged the rich biomarker dataset from the PALI cohort and selected a large panel of biomarkers (33 plasma biomarkers) for their prognostic significance and their relevance to signaling pathways involved in dysregulated inflammation, microvascular endothelial dysfunction, extracellular matrix destruction, thrombosis, and epithelial disruption, all of which are key pathobiological pathways of ARDS (e-Appendix, https://links.lww.com/CCM/H728).

Correlation heat maps illustrated the relationships between biomarkers (and in part previously published) (27). There was varying degree of missingness in the biomarker data set and missing data were imputed using multiple imputation by chained equations (e-Appendix, https://links.lww.com/CCM/H728). We expanded upon established biomarker network analysis methodologies (29–31) to construct a biomarker correlation network based on pairwise Spearman correlations. The network was generated using a Spearman coefficient threshold of greater than or equal to 0.2 and a false discovery rate p value threshold of less than 0.05. Each protein biomarker was represented as a node, with pairwise correlations shown as undirected weighted edges. Network topology—the arrangement of nodes and edges—was visualized as a force-directed graph using Cytoscape (32). Centrality metrics (degree, strength, betweenness, closeness, and eigenvector) were computed to identify key nodes or “network hubs.” These measures were normalized to a common scale and combined into a composite metric, ranging from 0 to 1. Network hubs were defined as those with composite centrality values in the top 10% (≥ 90th percentile). The fast-greedy optimization method was applied to identify densely connected nodes (i.e., communities) within the broader network (e-Appendix, https://links.lww.com/CCM/H728).

RESULTS

Cohort Baseline Characteristics

PALI (ARDS) Cohort.

Of the 334 patients enrolled, 269 patients had blood drawn for sICAM measurement within 24 hours of ARDS diagnosis. Of the 269 patients, 55 patients were further excluded due to insufficient plasma volume for sICAM quantification, and 214 patients were included in the analysis. Baseline demographics of these patients are shown in Table 1; and e-Table 1 (https://links.lww.com/CCM/H728). Among the included patients, 62.6% were diagnosed with direct and 37.4% with indirect pediatric ARDS. OI at the time of ARDS diagnosis was 10 (IQR, 5.6–19.7), Pao₂/FIO₂ (P/F) ratio was 142 mm Hg (IQR, 88–228 mmHg), and Pediatric Risk of Mortality (PRISM) score was 12 (IQR, 5–19). In-hospital mortality was 18%. Nonsurvivors had higher OI and PRISM scores and were more likely to have cancer/bone marrow transplantation diagnoses compared to survivors (all p < 0.05).

TABLE 1.

Baseline Characteristics

	Pediatric Acute Lung Injury Cohort			Coagulation and Fibrinolysis in Pediatric Insulin Titration Trial Cohort
Characteristics	Survivors (n = 175)	Nonsurvivors (n = 39)	p	Survivors (n = 217)	Nonsurvivors (n = 34)	p
Age (mo), median (IQR)	47 (10–136)	94 (30–164)	0.15	75 (23–145)	70 (20–144)	0.74
Sex, n (%)
Male	91 (52.0)	26 (66.7)	0.1	116 (53.5)	15 (44.1)	0.31
Female	84 (48.0)	13 (33.3)		101 (46.5)	19 (55.9)
Race, n (%)
Caucasian	119 (68.0)	26 (66.7)	0.35	134 (61.8)	23 (67.7)	0.72
African American	15 (8.6)	2 (5.1)		58 (26.7)	8 (23.5)
Asian/Pacific Islander	13 (7.4)	1 (2.6)		9 (4.2)	2 (5.9)
Other	28 (16.0)	10 (25.6)		16 (7.4)	1 (2.9)
Significant comorbidities, n (%)
Cancer/bone marrow transplantation	16 (9.1)	19 (48.7)	< 0.001	9 (4.2)	6 (17.6)	0.002
Vasopressor use	69 (39.7)	15 (38.5)	0.89	94 (43.0)	22 (64.7)	0.02
Illness severity, median (IQR)
Baseline OIa	9.5 (5.2–18.9)	16.5 (8.1–28.0)	0.028	8.4 (3.5–16.2)	13.3 (3.8–36.7)	0.12
Baseline OIb	9.7 (5.4–18.9)	18.3 (9.1–34.3)	0.02	7.8 (3.5–15.8)	14.4 (5.3–47.7)	0.045
Pediatric Risk of Mortality score	11 (5–18)	17 (9–21)	0.006	11 (5–17)	15 (12–28)	0.0007
Baseline Pao2/Fio2 ratio	150 (89–244)	107 (76–170)	0.04	177 (116–284)	153 (60–281)	0.11
Day 1 respiratory support, n (%)
Mechanical ventilation	158 (90.3)	34 (87.2)	0.85	217 (100)	34 (100)
Oscillator	7 (4.0)	2 (5.1)		NA	NA
Continuous positive airway pressure or bilevel positive airway pressure	10 (5.7)	3 (7.7)		0	0

IQR = interquartile range, NA = not applicable, OI = oxygenation index.

^aBaseline OI with nonmissing data.

^bBaseline OI including missing data by imputation.

CAF-PINT (ARF) Cohort.

Of 303 patients enrolled in CAF-PINT, 251 had blood drawn for sICAM-1 measurement within 72 hours of IMV. Forty-two were excluded because of having had blood drawn more than 72 hours after IMV initiation and four because of no blood drawn. Baseline demographics of these patients are also shown in Table 1. The cause of ARF requiring IMV was determined to be of a direct etiology in 71% and indirect etiology in 29% of the total cohort. PRISM-3, OI, and P/F ratio were similar to the PALI cohort (Table 1; and e-Table 1, https://links.lww.com/CCM/H728). In-hospital mortality was 14% and also similar to the PALI cohort.

sICAM-1 and Mortality

PALI (ARDS) Cohort.

Median plasma log sICAM-1 levels were higher in nonsurvivors than survivors (p = 0.029) (e-Fig. 1, https://links.lww.com/CCM/H728). In unadjusted analysis, higher sICAM-1 levels were associated with increased odds of in-hospital mortality (1.88 per sd; 95% CI, 1.24–2.84; Table 2). After adjusting for age, sex, race, and OI, the odds of mortality were 1.71 times higher (95% CI, 1.12–2.62; p = 0.014) per sd of log-increase in sICAM-1. We also tested whether baseline sICAM levels varied by race, cancer status, and vasoactive use. We found no significant difference in sICAM levels when stratified by race or vasoactive use (e-Figs. 2 and 3, https://links.lww.com/CCM/H728), but higher sICAM levels were seen in cancer patients compared to noncancer patients (p < 0.01; e-Fig. 4, https://links.lww.com/CCM/H728). However, no significant interaction was observed between sICAM levels and cancer status on the outcome of mortality (e-Table 2, https://links.lww.com/CCM/H728).

TABLE 2.

Clinical Outcomes

Clinical Outcomes	Derivation Cohort (Pediatric Acute Lung Injury Cohort)			Validation Cohort (Coagulation and Fibrinolysis in Pediatric Insulin Titration Trial Cohort)
Mortality	n	OR (95% CI)	p	n	OR (95% CI)	p
Univariate	214	1.88 (1.24–2.84)	0.003	251	1.49 (1.02–2.19)	0.04
Multivariable (age, sex, race, OI)	214	1.71 (1.12–2.62)	0.014	251	1.47 (0.97–2.23)	0.06
Multiple organ failure/Pediatric Logistic Organ Dysfunction	n	Coefficient (95% CI)	p	n	Coefficient (95% CI)	p
Univariate	214	2.78 (1.16–4.4)	0.001	248	1.62 (0.25–3.0)	0.02
Multivariable (age, sex, race, OI)	214	2.45 (0.78–4.13)	0.004	248	1.32 (0.05–2.6)	0.04
Ventilator-free days	n	Coefficient (95% CI)	p	n	Coefficient (95% CI)	p
Univariate	201	−1.63 (−3.1 to −0.17)	0.03	251	−1.77 (−2.95 to −0.59)	0.003
Multivariable (age, sex, race, OI)	201	−1.25 (−2.74 to 0.25)	0.1	251	−1.32 (−2.72 to −0.074)	0.06
ICU length of stay	n	Coefficient (95% CI)	p	n	Coefficient (95% CI)	p
Univariate	214	−0.96 (−4.3 to 2.38)	0.57	217	4.32 (2.0–6.65)	< 0.001
Multivariable (age, sex, race, OI)	214	−1.08 (−5.37 to 3.2)	0.62	217	4.23 (2.1–6.37)	< 0.0001

OI = oxygenation index, OR = odds ratio.

The AUROC (0.68) (95% CI, 0.6–0.8; e-Fig. 5, https://links.lww.com/CCM/H728) for mortality was similar to the AUROC of PRISM alone (AUROC, 0.7; 95% CI, 0.6–0.8) and did not further significantly increase when PRISM was added to the logistic regression model (AUROC, 0.7; 95% CI, 0.54–0.73). We also performed additional AUROC analysis incorporating other endothelial markers (thrombomodulin, angiopoietin-2 [Ang-2], vascular cell adhesion molecule-1, and von Willebrand factor) with PCA. These endothelial markers together showed a similar AUROC (0.68) to sICAM-1 alone. After the addition of sICAM-1 to the model containing endothelial markers, the AUROC increased from 0.68 to 0.71 (p = 0.07), suggestive of sICAM-1 providing additional predictive value beyond that of the other endothelial markers alone.

CAF-PINT (ARF) Cohort.

In the CAF-PINT cohort, median plasma log sICAM-1 was higher in non-survivors compared to survivors (p = 0.014) as shown in e-Figure 1 (https://links.lww.com/CCM/H728). In unadjusted analysis, higher sICAM-1 levels were associated with increased odds of in-hospital mortality (1.49 per sd; 95% CI, 1.03–2.19; p = 0.04; Table 2). After adjusting for age, sex, race, and severity of oxygenation failure (OI), the odds of mortality were 1.47 times higher (95% CI, 0.97–2.23; p = 0.06) per sd of log-increase of sICAM-1, with the p value trending toward significance. We did observe a significant difference in sICAM levels when stratified by race, cancer status, and vasoactive use (e-Figs. 1–3, https://links.lww.com/CCM/H728). However, there was no significant interaction between sICAM and any of these variables with respect to mortality (e-Table 2, https://links.lww.com/CCM/H728).

The AUROC 0.62 (95% CI, 0.51–0.72) for mortality was similar to the AUROC of PRISM alone (AUROC, 0.68; 95% CI, 0.57–0.79). There was an observed increase when PRISM was added to the logistic regression model (AUROC, 0.70; 95% CI, 0.51–0.73; p = 0.05; e-Fig. 5, https://links.lww.com/CCM/H728). Additional analysis incorporating other endothelial markers, as shown in the PALI cohort, was not carried out due to limited biomarker measurements in this cohort.

sICAM-1 and Secondary Outcomes

PALI (ARDS) Cohort.

We defined PELOD-2 quartiles and observed that sICAM-1 increased as PELOD-2 quartile increased (p < 0.001; e-Fig. 6, https://links.lww.com/CCM/H728). Increased sICAM-1 levels were significantly associated with PELOD-2 score in both univariate and multivariable linear regression analysis adjusting for age, sex, race, and OI (Table 2). On univariate analysis, higher sICAM-1 levels were associated with fewer VFDs, but not in the multivariable model (Table 2). sICAM levels were not associated with ICU LOS in either of the univariate or multivariable linear regression model.

CAF-PINT (ARF) Cohort.

sICAM also increased in a stepwise fashion as PELOD-2 quartile increased (e-Fig. 6, https://links.lww.com/CCM/H728). Increased sICAM-1 levels were also significantly associated with PELOD-2 score in both the univariate and multivariable linear regression analysis adjusting for age, sex, race, and OI (Table 2). Increased sICAM-1 levels were associated with fewer VFDs on univariate analysis, but not in the multivariable model. Increased sICAM-1 levels were significantly associated with longer ICU LOS in both our univariate and multivariable linear regression model.

Biomarker Correlation Network-PALI Cohort

Higher plasma levels of sICAM-1 positively correlated with most of the 32 biomarkers in the panel, including those associated with inflammation, endotheliopathy, thrombosis, extracellular matrix degradation, and epithelial injury (Fig. 1). The force-directed graph illustrating these relationships is shown in Figure 2. Network topological analysis identified tissue inhibitor of metalloproteinase-1 (TIMP-1) (composite centrality, 0.99), tumor necrosis factor receptor-1 (TNFR-1) (0.83), sICAM-1 (0.74), and interleukin (IL)-8 (0.74) as network hubs, with composite centrality values greater than or equal to 90th percentile (Table 3). Applying a similar rank-based criterion to individual centrality metrics—degree, strength, closeness, betweenness, and eigenvector—sICAM-1 displayed degree (0.96) and betweenness (0.43), and centrality values greater than or equal to 90th percentile (Table 3). High degree and betweenness centrality indicated that sICAM-1 had numerous direct connections to other nodes and lied on a high number of shortest paths between nodes. TIMP-1 values exceeded the 90th percentile across all individual centrality metrics. The fast-greedy method detected three interconnected node communities within the broader correlation network (Fig. 2). When the data was restricted to include only patients with less than 10% missing biomarker data (compared to the initial threshold of 25%), the results of the network analysis remained consistent with the initial analysis of 33 biomarkers (e-Table 3 and e-Figs. 7 and 8, https://links.lww.com/CCM/H728).

Figure 1.

Correlation heat map of 33 biomarkers in Pediatric Acute Lung Injury cohort. Ang-2 = angiopoietin-2, A1AT = α-1 antitrypsin, A2MB = α-2 macroglobulin, BDNF = brain-derived neurotrophic factor, B2MB = β-2 microglobulin, C3 = complement component 3, CRP = C-reactive protein, etoxan = eosinophil chemotactic protein, EPCR = endothelial protein C-receptor, fibrino = fibrinogen, MCP = monocyte chemotactic protein, IL-6 = interleukin-6, IL-8 = interleukin-8, IL-10 = interleukin-10, IL-18 = interleukin-18, MIP-1 β = macrophage inflammatory protein-1β, MMP-9 = matrix metalloproteinases-9, MMP-3 = matrix metalloproteinases-3, RAGE = soluble receptor of advanced glycation end-products, SCF = stem cell factor, sICAM = soluble intercellular adhesion molecule-1, SPD = surfactant protein-D, TIMP-1 = tissue inhibitor of metalloproteinase-1, TM = thrombomodulin, TNFR-1 = tumor necrosis factor receptor-1, TSP = T-cell specific protein, VCAM-1 = vascular cell adhesion molecule-1, VDBP = vitamin-D protein, VEGF = vascular endothelial growth factor, vWF = von Willebrand factor.

Figure 2.

Force-directed graph: The biomarker correlation network was visualized using Cytoscape (33). Edges (i.e., correlations) are weighted based on the absolute value of the Spearman coefficient between connected nodes (i.e., biomarkers), with edge color indicating a positive (black) or negative (red dotted) correlation. Node color represents communities identified using the fast-greedy algorithm, weighted by the absolute value of the Spearman coefficient. Node size reflects the composite centrality metric. The layout was determined using the prefuse force-directed algorithm, based on the normalized value of edge weights. Soluble intercellular adhesion molecule-1 (sICAM-1) is highlighted with a bold edge. Ang-2 = angiopoietin-2, A1AT = α-1 antitrypsin, A2MB = α-2 macroglobulin, BDNF = brain-derived neurotrophic factor, B2MB = β-2 microglobulin, CRP = c-reactive protein, C3 = complement component 3, etoxan = eosinophil chemotactic protein, EPCR = endothelial protein C-receptor, IL-6 = interleukin-6, IL-8 = interleukin-8, IL-10 = interleukin-10, IL-18 = interleukin-18, RAGE = soluble receptor of advanced glycation end-products, MCP = monocyte chemotactic protein, MIP-1β = macrophage inflammatory protein-1β, MMP-3 = matrix metalloproteinases-3, MMP-9 = matrix metalloproteinases-9, SCF = stem cell factor, SPD = surfactant protein-D, TIMP-1 = tissue inhibitor of metalloproteinase-1, TM = thrombomodulin, TNFR-1 = tumor necrosis factor receptor-1, TSP = T-cell specific protein, VCAM-1 = vascular cell adhesion molecule-1, VEGF = vascular endothelial growth factor, vWF = von Willebrand factor.

TABLE 3.

Centrality Metrics for 33 Protein Biomarkers in Pediatric Acute Lung Injury Cohort

Marker	Degreea	Betweennessa	Closenessa	Strengtha	Eigenvectora	Compositea
Tissue inhibitor of metalloproteinase-1	1	1	1	1	0.96	0.99
Tumor necrosis factor receptor-1	0.83	0.38	0.96	0.96	1	0.83
Soluble intercellular adhesion molecule-1	0.96	0.43	0.86	0.74	0.72	0.74
Interleukin-8	0.92	0.13	0.91	0.88	0.89	0.74
β-2 microglobulin	0.88	0.11	0.89	0.83	0.86	0.72
Interleukin-18	0.88	0.16	0.85	0.77	0.8	0.69
Ferritin	0.92	0.08	0.84	0.79	0.79	0.68
Angiopoietin-2	0.96	0.18	0.83	0.67	0.61	0.65
Interleukin-6	0.92	0.22	0.82	0.66	0.63	0.65
Vascular cell adhesion molecule-1	0.83	0	0.82	0.74	0.79	0.64
Monocyte chemotactic protein	0.79	0.02	0.83	0.72	0.77	0.63
C-reactive protein	0.88	0.33	0.76	0.57	0.5	0.61
Stem cell factor	0.79	0.07	0.81	0.67	0.68	0.6
Interleukin-10	0.83	0	0.78	0.67	0.72	0.6
T-cell specific protein	0.88	0.11	0.74	0.6	0.58	0.58
Macrophage inflammatory protein-1β	0.63	0	0.77	0.56	0.65	0.52
Eosinophil chemotactic protein	0.63	0.17	0.67	0.45	0.48	0.48
Matrix metalloproteinases-3	0.71	0	0.7	0.46	0.47	0.47
Thrombomodulin	0.58	0.02	0.68	0.41	0.44	0.43
Brain-derived neurotrophic factor	0.58	0.01	0.65	0.4	0.41	0.41
Endothelial protein C-receptor	0.58	0	0.64	0.38	0.42	0.4
Haptoglobin	0.54	0.01	0.62	0.34	0.18	0.35
Complement component 3	0.42	0.18	0.6	0.33	0.12	0.33
von Willebrand factor	0.46	0.08	0.62	0.26	0.22	0.33
Factor VII	0.33	0.19	0.63	0.22	0.13	0.3
α-1 antitrypsin	0.38	0.06	0.58	0.26	0.11	0.28
Fibrinogen	0.29	0.06	0.55	0.23	0.06	0.24
Soluble receptor of advanced glycation end-products	0.29	0	0.54	0.16	0.16	0.24
Vitamin-D protein	0.33	0	0.51	0.23	0.07	0.23
α-2 macroglobulin	0.25	0	0.52	0.16	0.06	0.2
Vascular endothelial growth factor	0.21	0.05	0.49	0.13	0.05	0.19
Surfactant protein-D	0.21	0	0.5	0.11	0.09	0.18
Matrix metalloproteinases-9	0.17	0	0.42	0.08	0.05	0.14

^aScaled values between 0 and 1 (values closer to 1 represent higher centrality metrics).

DISCUSSION

In two large independent multicenter pediatric cohorts of ARDS (PALI) and ARF (CAF-PINT), early elevation in sICAM-1 levels was significantly associated with higher in-hospital mortality, multiple organ failure, and fewer 28-day VFDs. In our multivariable analysis, these associations were independent of age, sex, race, and severity of oxygenation failure, for the outcome of multiple organ failure in both cohorts and for mortality in the PALI cohort, with trend to significance in the CAF-PINT cohort. To contextualize ICAM’s putative role in the pathophysiology of pediatric ARDS, we constructed a biomarker correlation network incorporating prognostic biomarkers of diverse mechanistic pathways. This approach identified sICAM-1 as a network hub, strongly associated with biomarkers linked to key biological processes in ARDS, including dysregulated inflammation (e.g., IL-8, soluble TNFR-1 [sTNFR-1]), endotheliopathy (e.g., Ang-2), and extracellular matrix degradation and remodeling (e.g., TIMP-1). These findings suggest ICAM-1 is not only an important biomarker for outcome prediction but also a central role in ARDS pathogenesis that may have potential therapeutic implications.

The results are well-aligned with several earlier, smaller single center pediatric ARDS studies that have shown increased plasma levels of sICAM-1 to be associated with prolonged length of mechanical ventilation, ARDS severity, and mortality (20–23, 34, 35). In a recent small cohort of pediatric ARDS patients, sICAM levels were associated with lung injury severity, outperforming 15 other plasma biomarkers of endothelial, epithelial, and hyperinflammation (34). Early elevated levels of alveolar and plasma sICAM-1 levels are known to be associated with poor clinical outcomes in adults with ALI (20).

However, the biological interest of sICAM extends beyond its prognostic utility for clinical outcomes and illness severity. Similar to ICAM’s multifaceted physiologic role in the normal functioning of essential tissues, dysregulated ICAM pathways play a multifaceted and complex role in primary pathogenesis of ARDS and ARF. ICAM enhances pro-inflammatory signaling pathways via nuclear factor kappa B and extracellular signal-regulated kinase, activates macrophages and the innate immune host response (36), facilitates transcellular migration, and contributes to epithelial and endothelial hyperpermeability and extracellular matrix degradation. Targeted ICAM therapies have shown promising therapeutic benefits in animal models of ALI and sepsis. In a lipopolysaccharide-induced murine lung injury model, anti-ICAM-1 antibodies loaded with dexamethasone traversed the pulmonary endothelium and modulated the infiltration of inflammatory cells into the lung and decreased cytokine production (37). In a murine sepsis model, anti-ICAM-1 antibody therapy improved survival and bacterial clearance, further attenuating the dysregulated immune response leading to end-organ damage that included the lung (38). Key mechanistic insights from these in vivo models provide additional evidence for the critical role of ICAM-1 in both ARDS and sepsis-induced organ dysfunction.

The “hyper-inflammatory” phenotype, identified in adult and later validated in pediatric ARDS (27, 39–41), is characterized by differential elevation of protein biomarkers of systemic inflammation and endothelial injury that include IL-6, IL-8, sTNFR-1, Ang-2, and ICAM-1 (among others). Among these biomarkers, ICAM-1 is unique in its critical role in leukocyte adhesion and transcellular migration, central to ARDS pathophysiology. Beyond biomarkers characterized by primary endothelial injury and/or inflammation, elevated sICAM in ARDS patients further reflects key disruptions in normal cellular interactions and progression of leukocyte-driven inflammation, integral to ARDS development and severity. For this reason, we sought to specifically contextualize sICAM within a network of diverse ARDS biomarkers reflective of varied pathobiological pathways. Complimentary to contemporary methods to biomarker-based molecular phenotyping, network biology explores molecular connections with network topology (i.e., arrangement of nodes and corresponding edges) and enables identification of key biomarkers (i.e., network hubs) that anchor these complex connections and interactions.

Our findings substantiate preclinical research mentioned above, highlighting ICAM-1’s intricate role as a central orchestrator of multiple pathogenic processes (transcellular migration, epithelial and endothelial injury, and immune cell dysregulation) (16, 17, 19, 32, 37). Taken together with preclinical studies, these findings underscore the role of ICAM-1 in the molecular underpinnings of this complex disease. Investigation of sICAM-1 as a potential therapeutic target in pediatric ARDS maybe warranted. However, it is important to note the direction of relationships between nodes—and, by extension, the causal mechanistic pathways—cannot be inferred from a correlation-based network alone. Thus, the functional and biological significance of key network hubs identified within this analysis (i.e., TIMP-1, sTNFR-1, sICAM-1, and IL-8), and their connections require rigorous prospective and preclinical validation.

This study has several strengths. To date, it is the largest multicenter study of critically ill children with respiratory failure to examine sICAM-1’s association with salient clinical outcomes. Patient recruitment involved multiple sites resulting in a large and diverse pediatric cohort, validating earlier smaller pediatric studies. Though these results derived in an ARDS cohort were replicated in an independent ARF cohort, severity of illness, mortality, and oxygenation metrics were comparable. In addition, sICAM-1 levels were collected prospectively and at comparable time points early in the course of ARDS/ARF development. The relatively large sample size permitted an analysis that can further adjust for key covariates known to influence variation in biomarkers. Finally, the availability of a large and diverse panel of ARDS biomarkers allowed for a network analysis approach. This network analysis, inclusive of 33 total protein biomarkers of varied pathobiological significance, is arguably the largest protein-protein biomarker network analysis currently within both adult and pediatric ARDS literature.

Limitations of our study include our inability to temporally assess changes to sICAM-1 levels and evaluation of sICAM-1 at the alveolar level—a common limitation of pediatric ARDS cohort. Plasma-measured biomarkers are often indirect surrogates of their membrane-bound form and cellular function, with plasma levels potentially influenced by various other biological factors and need consideration in interpretation of results. Since patient enrollment started prior to publication of the pediatric-specific Pediatric Acute Lung Injury Consensus Conference criteria in 2015, patients were initially screened using the adult criteria outlined by the American-European Consensus Conference and Berlin definition (42, 43). The proportion of African American participants represented in our PALI cohort (7.9%) was somewhat lower than in U.S. population as a whole (14%) (44). Finally, in correlation-based (or “undirected”) networks, connections are bidirectional, meaning causality cannot be established. However, these analyses provide a novel approach for assessing complex pathobiological interactions and generating plausible hypotheses that could then inform validation in mechanistic studies in preclinical ARDS models.

In conclusion, early elevation in sICAM-1 levels were associated with higher mortality, severity of organ dysfunction, and fewer VFDs in critically ill children with ARDS and ARF in two cohorts. sICAM-1 was identified as a key biomarker in a correlation-based network that included biomarkers associated with inflammation, endothelial and epithelial injury, thrombosis, and extracellular matrix damage. The association of sICAM-1 with poor outcomes suggests that it may be useful for helping to identify critically ill children at risk of poor outcomes from pediatric ARDS and ARF. Further, its central role within this network paired with its known multifaceted mechanistic role in ARDS disease warrants future evaluation of its potential role as a target for treatment in the context of children with ARDS.

Supplementary Material

Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (https://journals./ww.com/ccmjournal).

KEY POINTS.

Question:

What is the association of early plasma soluble intercellular adhesion molecule-1 (sICAM) and clinical outcomes in children with acute respiratory distress syndrome (ARDS), and is there a central role of sICAM in ARDS pathophysiology within the context of a network analysis incorporating other validated prognostic ARDS biomarkers?

Findings:

In two separate large multicenter pediatric cohorts of ARDS and acute respiratory failure, higher early plasma sICAM-1 was significantly associated with in-hospital mortality in each of the two cohorts. Further, sICAM was found to be a central hub among a network analysis of 32 other plasma ARDS biomarkers.

Meaning:

These findings underscore the critical role of sICAM in ARDS pathophysiology.

APPENDIX

We thank the contributions of the PALI research site collaborators for their assistance in recruiting patients for this study: Heidi Fiori, MD (Children’s Hospital of Oakland); Robinder Khemani, MD (Children’s Hospital of Los Angeles); Ana Graciano, MD (Children’s Hospital Central California); and Juan Boriosi, MD (American Family Children’s Hospital) in addition to the Study Site Investigators for CAF-PINT.

CAF-PINT investigators include: 1) Boston Children’s Hospital: Investigators, Michael Agus, MD, Lisa Asaro, MS.; Research Coordinators, Kerry Coughlin-Wells, Kyle Hughes, Jaclyn French, Meghan Fitzgerald; 2) Children’s Hospital of Philadelphia: Investigator, Vijay Srinivasan, MD; Research Coordinator, Martha Sisko; 3) Cincinnati Children’s Hospital Medical Center: Investigator, Ranjit S. Chima, MD, FAAP; Research Coordinators, Kelli Howard, Rhonda Jones; 4) PennState Children’s Hospital: Investigator, Neal J. Thomas, MD, MS; Research Coordinator, Debbie Spear; 5) Maria Fareri Children’s Hospital: Investigators, Simon Li, MD, Alan Pinto, MD; Research Coordinator, Peter Eldridge; 6) Children’s Hospital of Los Angeles: Investigator, Christopher Newth, MD, FRCPC; Research Coordinator, Jeni Kwok; 7) John R. Oishei Children’s Hospital, Buffalo, NY: Investigator, Amanda B. Hassinger, MD, MSc; Research Coordinator, Haiping Qiao; 8) Medical City Children’s Dallas: Investigator, Kris Bysani, MD; Research Coordinator, Tracey Monjure; 9) Yale-New Haven Children’s Hospital: Investigator, Edward Vincent Faustino, MD, MHS; Research Coordinators, Joana Tala, Sarah A. Kandil, Tyler Quinn; 10) Primary Children’s Medical Center: Investigator, Eliotte Hirshberg, MD; Research Coordinator, Jennifer Lilley; 11) University of Louisville and Norton Children’s Hospital: Investigators, Kupper Wintergerst, MD; Janice E. Sullivan, MD; Research Coordinator, Kristen Lee; 12) Children’s Hospital of Orange County: Investigator, Adam Schwarz, MD; Research Coordinators, Cathy Flores, Ofelia Vargas-Shiraishi. Ann & Robert H; 13) Lurie Children’s Hospital of Chicago: Investigators, Lauren Sorce, APN NP, CPNP, AC/PC, FCCM, Lauren Marsillio, MD; Research Coordinator, Avani Shukla; 14) Children’s Hospital and Research Center Oakland: Investigator, Natalie Cvijanovich, MD, Heidi Flori, MD; Research Coordinator, Becky Brumfield; 15) Children’s Healthcare of Atlanta (Emory): Investigator, Nga Pham, MD; Research Coordinator, Cheryl Stone; 16) C.S. Mott Children’s Hospital; Investigator, Mary Dahmer, PhD; Research Coordinator, Chaandini Jayachandran; 17) Mattel Children’s Hospital (U.C.L.A.): Investigators, Myke Federman, MD, Kayley Wong, BS, Sitaram S. Vangala, MS, Matteo Pellegrini, PhD, Brunilda Balliu, PhD, Kinisha P. Gala, PhD; Research Coordinators, Theresa Kirkpatrick, Tanaya Deshmukh, Manvita Mareboina, Nguyen Do, Neda Ashtari, Anna Ratiu; 18) Dartmouth Hitchcock Medical Center: Investigators, Sholeen Nett, MD, Marcy Singleton, ARNP; Research Coordinators, Dean Jarvis, Mary McNally, Karlyn Martini; 19) University of Chicago Comer Children’s Hospital: Investigators, Neethi Pinto, MD, MS, Grace Chong, MD; Research Coordinator, Chiara Rodgers; 20) AI DuPont/Nemours: Investigator, Shirley Viteri, MD; Research Coordinators, Ramany John, Teresa Mulholland, Gwen Pellicciotti; and 21) UCSF Benioff Children’s Hospital: Investigators, Anil Sapru, MD, Patrick McQuillen, MD, Matt Zinter, MD; Research Coordinators, Shrey Goel, Mustafa Alkhouli, Anne McKenzie, Denise Villarreal-Chico.