Peeking under the Hood of Acute Respiratory Distress Syndrome Phenotypes: Deeper Insights into Biological Heterogeneity

Given the biological heterogeneity inherent to acute respiratory distress syndrome (ARDS), it is unsurprising, or perhaps even inevitable, that a pharmacotherapy “magic bullet” targeting a specific mechanistic pathway has failed to emerge. A silver lining in the plethora of failed clinical trials is the foresight of the original trial investigators to judiciously collect and store biospecimens, which, coupled with the richness of the trial data, has permitted informative secondary analyses. Investigators have used these data to perform unsupervised subgrouping analyses and have consistently identified two biologically and clinically distinct phenotypes in ARDS, referred to as the “hypoinflammatory” and “hyperinflammatory” phenotypes (1–4). The hyperinflammatory phenotype is associated with elevated plasma levels of proinflammatory biomarkers and an increased incidence of shock and organ dysfunction. More recently, researchers identified two phenotypes using only plasma biomarkers in an observational cohort of patients with ARDS (5). These phenotypes, termed “uninflamed” and “reactive,” share similarities with the hypoinflammatory and hyperinflammatory phenotypes, respectively. In both classification schemes, the two phenotypes have divergent clinical outcomes, and the phenotypes derived from trial data were also associated with distinctly different responses to randomized treatments (6).

Although the nomenclature attached to these ARDS phenotypes focuses on inflammation, in reality, our understanding of the biological pathways associated with the phenotypes remains very superficial. A deeper understanding of the true biological differences between the phenotypes will be required before we can move toward personalized therapies directed at one or the other specific subtype. In a study reported in this issue of the Journal, Bos and colleagues (pp. 42–50) used whole-blood transcriptomic data to study key differences in biological pathways associated with the reactive and uninflamed phenotypes in an observational cohort of 210 patients with sepsis-related ARDS (7). First, the authors used a previously developed five-biomarker-based classifier model (5) to assign either the reactive or the uninflamed phenotype. Next, in plasma samples drawn simultaneously, they used microarray technology to compare mRNA expression of 11,443 genes between the two phenotypes. A significant difference in expression was observed in 29% of these genes. In the reactive phenotype, genes associated with leukocyte activation were more highly expressed, with MMP8, OLFM4, RETN, and GPR84 showing the greatest fold difference in expression compared with the uninflamed phenotype. By pathway analysis, the gene enrichment pattern of the reactive phenotype was most associated with oxidative phosphorylation, mitochondrial dysfunction, and cholesterol metabolism pathways.

In contrast, the uninflamed phenotype was associated with higher expression of genes that were previously found to be negatively associated with ARDS (e.g., MME and HCAR3) (8). Pathway analysis revealed expression patterns associated with cell proliferation/survival (mitogen-activation protein kinase and RAF-1 pathways). Many of the discriminatory genes associated with pathways in the uninflamed phenotype are ubiquitously expressed during cellular stress from a multitude of insults (9). Interestingly, in contrast to the reactive phenotype, where the altered pathways were associated mostly with the innate immune system, in the uninflamed phenotype several of the altered pathways were associated with the adaptive immune system.

What are the key “big-picture” implications of this study? In the reactive phenotype, several of the top differentially expressed genes were neutrophil related and known to be immune modulatory (10, 11). Coupled with elevated levels of proinflammatory biomarkers, these data suggest that the innate immune response is a key contributor to the pathogenesis of the reactive phenotype (6). From a biological standpoint, these findings further undermine our current approach to defining ARDS, which mandates “lumping” these disparate patients together (12). Another notable finding of this study is the analysis of pathways that are relatively overexpressed in the uninflamed phenotype. To date, the uninflamed phenotype has been primarily defined by its relative lack of inflammatory markers, rather than by any distinct biology of its own. Along these lines, it is interesting to note that the gene-expression pattern of the uninflamed phenotype overlapped more closely with sepsis than with the reactive phenotype of ARDS in a principal component analysis, bringing into question the accuracy with which the current definition of ARDS identifies the pathophysiologic entity it seeks.

This study has several important strengths, including the fact that it is the first to explore differences in gene expression and associated pathway analyses in ARDS phenotypes. The study was performed in a carefully curated and prospectively accrued population of patients with ARDS. In the context of a clinical syndrome already plagued by heterogeneity, studying gene expression in unselected and/or poorly defined populations that are extracted from registries can often lead to increased noise-to-signal ratios, resulting in data that are difficult to interpret (13). Interestingly, some of the key neutrophil-related upregulated genes identified in the reactive phenotype have also been identified in prior ARDS studies (8, 13). Kangelaris and colleagues identified upregulation of several genes in sepsis-related ARDS that were also identified by the investigators of the current study, including OLFM4 and LCN2, among others (8). This corroboration of previous findings is notable and adds to the validity of the study.

The study also has some limitations. First, the use of microarray technology rather than RNA sequencing may lead to some inherent bias, as only the genes on the array can be detected. Second, as with all transcriptomic analyses, definitive causal mechanistic pathways and downstream functional protein expression (other than that of MMP-8, which the authors did measure in plasma) remain unknown. Third, the nature of the studied population (sepsis specific) limits the generalizability of the findings, and it may be that the authors’ findings are sepsis related rather than ARDS specific. Future analyses should include true replication in a general independent ARDS cohort.

Many important questions regarding ARDS phenotypes remain unanswered. It is unknown whether the presented reactive phenotype is the same as or similar to the hyperinflammatory phenotype described elsewhere in the literature (6). In contrast to studies of sepsis (14), investigators have not yet used gene-expression data in unsupervised analyses to seek de novo discrete subtypes of ARDS, and it is unknown whether the same two phenotypes, or more, would emerge with such an approach. Although the authors have made a start in understanding the underlying biology of the uninflamed/hypoinflammatory phenotype, much remains unknown about this phenotype that constitutes 38–70% of the ARDS population. Biological differences between phenotypes in the lung compartment also remain unknown. Observational studies in humans cannot identify causal pathways, so experimental work will be necessary to truly test mechanistic hypotheses. Finally, to identify the phenotypes rapidly, at the bedside, real-time tests to quantify key biomarkers are urgently needed.

As the field of seeking phenotypes in ARDS matures, a wider array of high-dimensional data, either in isolation or in concert, will be used to explore systems biology in clinical syndromes. Bos and colleagues have elegantly demonstrated that phenotypes identified using plasma biomarkers have distinct gene-expression profiles and biological attributes. This study represents a substantial step forward on the challenging journey toward delivering precision medicine in ARDS.