Genetic Heterogeneity and Risk of Acute Respiratory Distress Syndrome

Abstract Genetic variation explains some of the observed heterogeneity in patients' risk for developing the acute respiratory distress syndrome (ARDS). Although the lack of extant family pedigrees for ARDS precludes an estimate of heritability of the syndrome, ARDS may function as a pattern of response to injury or infection, traits that exhibit strong heritability. A total of 34 genes have now been reported to influence ARDS susceptibility, the majority of which arose as candidate genes based on the current pathophysiological understanding of ARDS, with particular focus on inflammation and endothelial or epithelial injury. In addition, novel candidate genes have emerged from agnostic genetic approaches, including genome-wide association studies, orthologous gene expression profiling across animal models of lung injury, and human peripheral blood gene expression data. The genetic risk for ARDS seems to vary both by ancestry and by the subtype of ARDS, suggesting that both factors may be valid considerations in clinical trial design.

The completion of the Human Genome Project in 2003 ushered in a new era of biomedical research. With a reference deoxyribonucleic acid (DNA) sequence for all 3 billion base pairs and roughly 24,000 human genes, researchers worldwide had the tools to interrogate the association between genetic variation and disease. 1 Capitalizing on this investment, technological advances in high-throughput genotyping allowed the International HapMap Consortium to catalogue genetic variation between different global populations. [2][3][4] This exponentially growing body of genetic information has been harnessed to create higher resolution genotyping platforms on a genome-wide level, such that genetic linkage studies-once the exclusive territory of family-based genetics-could be adapted to investigate the genetics of unrelated cases sharing a phenotype: the modern genome-wide association study (GWAS). 5 The number of traits and disease states investigated for a genetic basis has exploded, fueled by rapidly advancing information and technologies; the number of published GWASs grew from fewer than 10 in 2005 to almost 1,000 by 2011. 6 Lending support to the utility of genetic investigations to advance our mechanistic understanding of disease, the polymorphisms that have been associated with diseases or traits by GWAS display an overrepresentation of promoter or nonsynonymous coding variants, polymorphisms with an intuitively functional role. 7 The variants detected by GWAS also underrepresent the intergenic or gene dessert regions of the genome. 7 Despite this exciting era for genetic research, it was not immediately apparent whether genomic science could help to unravel the complexities of a trait-like acute respiratory distress syndrome (ARDS). No family pedigrees of ARDS exist to help focus attention on a region of the genome, because ARDS does not arise spontaneously but rather follows severe illness, and most predisposing critical illnesses are themselves sporadic. It remains uncertain if ARDS triggered by different insults represents a single pathophysiology with Keywords ► genetic association study ► genetic variation ► polymorphism ► ARDS ► ALI shared genetic underpinnings or if distinct mechanistic processes merely elaborate a similar clinical phenotype. The acute nature of the illness also raises challenges for genetic study design, as patients' proxy decision makers are asked to consider participation in research while their loved one is in the throes of critical illness. Nonetheless, despite these formidable challenges to the genetic investigation of ARDS, the field can claim numerous successes. This article reviews the rationale for investigating the genetic risk for ARDS, the currently recognized genetic contributors to ARDS risk, and future directions in which the field is likely to evolve.

Why Study the Genetics of ARDS?
Heterogeneity is a clinical hallmark of ARDS. Heterogeneous conditions can incite the syndrome; only a minority of patients with predisposing factors will manifest the syndrome; and once triggered, the clinical course can vary from resolution within hours to prolonged, protracted ventilatory dependence. Although clinical risk factors for the syndrome are well documented, 8,9 clinicians find it challenging to predict which individual patients will develop severe hypoxemia and low lung compliance, and of those who do develop ARDS, which patients will recover. [10][11][12] Because available clinical risk factors alone do not explain the risk for ARDS, attention has turned to genetics as one possible explanation for interindividual variation in development of lung injury. 13,14 ARDS epitomizes a complex genetics trait, which cannot be explained by one polymorphism or even one gene. In contrast to Mendelian or monogenetic disorders, ARDS requires a fairly extreme environmental insult-sepsis, pneumonia, aspiration, major trauma, or exposure to a ventilator-to be superimposed on an underlying predisposition to manifest the clinical phenotype. Because the precipitants triggering ARDS are so severe, they tend to be sporadic, leading to an absence of pedigrees with notable ARDS risk.
Despite the lack of classical evidence for ARDS heritability such as might come from segregation analysis or twin studies, there is strong evidence for heritability of an individual's response to infection or injury. Infection and trauma are two major evolutionary threats that have shaped our present-day populations, 15 and individual responses to these forces have almost certainly been subjected to selective pressure. Genetic variants with notably different frequencies in different global populations, including the genes encoding for cytokines, cytokine receptors, and cytokine antagonists, 16 offer some clues to selection pressure. Genome-wide maps of positive selection hint at cultural, geographic, and genetic coevolution. [17][18][19] For instance, the hemoglobin S polymorphism and the Duffy null genotype, which are both highly frequent in West African populations and have been implicated with a protection from malarial infection, are almost nonexistent in European populations. 3 Functional variants in the alternative pathway of complement activation may have been selected for during yellow fever and dengue outbreaks among Dutch colonialists, 20 and some of the same "proinflammatory" variants are now implicated in age-related macular degener-ation. 21 Furthermore, a study of Danish adoptees in the preantibiotic era found that a child's risk of premature death from infection was sixfold higher when the child's biological parent had also died prematurely from infection, establishing death from infection as a trait with stronger heritability than cancer or vascular disease. 22 In this context, ARDS could be viewed as an evoked phenotype triggered by certain patterns of injury, and the rationale to seek a heritable basis for ARDS becomes more apparent.

Designing Genetic Association Studies for ARDS
One way to conceptualize the influence of genotype on the risk for ARDS is as a multistage model, much like the multistage model of carcinogenesis, as outlined in ►Fig. 1. 23, 24 We can think of genotypes affecting exposure to ARDS risk factors (G E , including genetic factors predisposing to sepsis, pneumonia, or trauma); affecting the risk of developing ARDS once exposed (G D ); or influencing the risk of mortality once ARDS has developed (G O ). Ideally, as specific pharmacotherapies become available, it will also be possible to test for evoked genotypes associated with response to therapy (G T ) or preventive approaches (G P ). ►Fig. 1 also illustrates the importance of study design in amplifying the genetic signal; to hone in on genotypes influencing ARDS development, it is important to capture cohort or case-control populations in which the entire population has a predisposing risk for ARDS (G E ) or the study risks capturing both genotypes associated with both exposure to ARDS risk factors (G E ) and development of ARDS itself (G D ), likely diluting the strength of association for each.
Once the study population is defined, there are several genomic approaches to investigate the role of genetic variants in determining ARDS risk (►Table 1). A comprehensive review of genetic variation and genomic study design is

Sepsis Pneumonia Trauma
Fig. 1 Multistage model of genetic risk. A useful conceptual model illustrates the many ways in which genotype can influence acute respiratory distress syndrome (ARDS) risk and outcome. Genotype can affect exposure to ARDS risk factors (G E , including genetic factors predisposing to sepsis, pneumonia, or trauma); the risk of developing ARDS once exposed (G D ); or the risk of mortality once ARDS has developed (G O ). To maximize the ability to detect a genetic signal influencing exposure, disease, or outcome, an effective study design will capture only the population that tests one category at a time. Ideally, as specific pharmacotherapies become available, it will also be possible to test for evoked genotypes associated with response to therapy (G T ) or preventive approaches (G P ). (Adapted from Rebbeck 23 and Christie. 24 ) beyond the scope of this article, but the reader is directed to excellent recent reviews. 25,26 If the study population is small, or if the desire is to test a limited number of hypothesis-driven variants in one or a few genes, the design is that of a candidate gene association study. The bulk of published information on genetic contribution to ARDS risk comes from candidate gene studies, in which the researcher selects putatively functional variants in a gene hypothesized to play a role in ARDS pathogenesis, and tests for enrichment of the polymorphism in either the case or the control population. Advantages of this approach are that it is relatively simple to design probes and test the samples, that the study results are intuitive and straightforward to interpret by virtue of testing a putatively functional variant, and that there is minimal statistical penalty for multiple comparisons because the number of tests performed is small. Hypothesis-driven questions focused on a specific polymorphism, gene, or gene pathway are well suited to this design because it will maximize power to answer a focused question. The main shortcoming of the candidate gene approach is that it relies on the investigator to select a priori the "correct" candidate gene, and further, the correct genotyping strategy to capture the functional variant of the gene. Moreover, recurrent tests of individual candidate genes are an inefficient use of DNA in the era of multiplexed genotyping platforms.
At the other extreme, the GWAS approach tests > 500,000 polymorphisms with an intent to capture > 80% of all the variation in the genome, and tests all variants for enrichment in the diseased population. Given the staggering number of comparisons, GWASs employ a strict statistical thresholdtypically a p-value < 5 Â 10 À8 , accounting for the approximate number of individual tests after accounting for linkage disequilibrium (LD) across the genome-to declare significance. By hypothesizing that significant differences in genotype will exist, rather than predicting which genes or variants will differ, GWASs afford no weight to potentially functional variants, or to high-priority genes with a potential role in the disease. In fact, many platforms do not directly test the purported functional variant itself but rely on LD, the predictable correlation of a group of single nucleotide polymorphisms (SNPs) inherited together in a block of the genome, to locate the chromosomal region of association with a trait. Given this design, one criticism of the GWAS approach is that the stringent statistical threshold risks missing important associations, whose strength of associa-tion may be more modest than p < 5 Â 10 À8 . In addition, to adequately power to such an extreme p-value, one needs a minimum of approximately 600 cases and ideally twice as many at-risk controls. 27 Critically ill sample sizes of this number have been challenging to accrue. Furthermore, because the strategy relies on knowing the LD structure of the population being studied to hone in on the precise genomic region, it can be difficult to know exactly which gene and which variation in the gene is driving the association with phenotype. The data output generated by each GWAS is massive and typically requires bioinformatic interrogation to render meaningful results. Finally, even when an association signal is extremely strong and replicated, it can be difficult to know what the association signifies. [28][29][30] Intermediate between the single candidate gene association study and GWAS are multiplexed genotyping platforms or high-throughput SNP arrays, which can simultaneously test numerous polymorphisms on the same patient samples, testing from as few as 15 gene variants to more than 50,000. Medium-and high-throughput approaches offer efficiency and value, with a lower computational or bioinformatic workload compared with a GWAS, and a less extreme statistical threshold to declare significance. The main drawback to high-throughput genotyping is the lack of consensus on an appropriate statistical threshold, particularly as pertains to a custom SNP array containing high-priority genes. 31,32 Some investigators recommend adjusting only for the number of genes tested, whereas others advocate using the haplotype block as the unit of interest, and still others correct the number of SNPs tested for LD to approximate the number of independent tests. [33][34][35] Several publications using the 50,000 SNP array designed for the National Heart Lung and Blood Institute Candidate Gene Association Resource group to interrogate cardiopulmonary and metabolic phenotypes, known as the ITMAT-Broad-CARe chip, have used discovery thresholds ranging from 10 À4 to 10 À6 , 32,35-38 and then replicated positive results.
Modern genotyping platforms measure predominantly or exclusively SNPs, which are single base-pair alterations in the DNA sequence. Some multiplex platforms are also facile at the detection of structural variants, such as copy number variants (CNVs), tandem repeats, insertions, deletions, or inversions, but these remain the minority. Bioinformatic programs exist to infer CNVs from large-scale SNP array data, though it remains difficult to reliably call CNVs if the structural variant is small or lacks tight LD with the genotyped SNPs. As sequencing costs decrease and quality assurance standards evolve, future endeavors will focus on high-throughput, "next generation sequencing (NGS)," which allows a massive parallel sequencing of DNA across the genome and should be informative about all SNPs and structural variants. Many of the available NGS platforms focus on sequencing only the protein-coding regions of the genome, known as the exome, to look for an accumulation of multiple rare variants in one gene encoding the same protein, and this technique has been effective at defining the molecular defect for some rare disorders with a handful of unrelated cases. 39 Sequencing overcomes the reliance of GWAS on tagging SNPs to approximate the genomic region associated with a trait, but comes with the burden of enormous data generation and the need for strenuous quality assurance steps. It remains uncertain how much ARDS genetic risk will be explained by rare variants clustered within a small number of genes, but this is an area of intense interest for future research. 40,41 Beyond changes in genomic DNA, there may be important variation in the level of gene expression, DNA methylation, or DNA histone modification that might impact ARDS susceptibility. These fields are largely in their infancy with respect to the ARDS phenotype; however, these techniques may feature prominently in future studies as technology advances and patient populations accrue. Gene expression profiling might improve our ability to molecularly subclassify ARDS and may yield novel insight to ARDS pathogenesis. 42,43 Furthermore, linking gene expression changes to variation in DNA sequence can allow the identification of expression quantitative trait loci, genetic variants that are likely to have functional consequences and which may prove fruitful as candidate risk variants. 43 Methylation and histone modification are considered epigenetic processes, in that they affect gene expression but through mechanisms that do not alter the DNA sequence. DNA hypermethylation typically suppresses gene transcription, whereas hypomethylation may leave transcription unchecked. 44 Further, the significance of small noncoding ribonucleic acids, including micro-RNA, is becoming apparent in other disease models and deserves further study in ARDS. 45,46 Progress could also be made in performing interaction studies testing how different genetic variants jointly influence ARDS susceptibility and how genetic and environmental risk factors interact. [47][48][49]

Genetic Variation Implicated in ARDS Risk
The past decade has witnessed an explosion in the number of genetic association studies of ARDS. Candidate gene studies led the charge, given their feasibility and the only recent maturity of genome-wide technology and informatics. Because candidate genes are understandably chosen based on the current pathophysiology of ARDS, many of those found to associate with ARDS fall into the broad categories of immune regulation, endothelial barrier function, and respiratory epithelial function, as described in the following section. However, investigators have also employed innovative translational and bioinformatic approaches to expand the current pool of ARDS-susceptibility genes (►Table 2), and have extended our understanding of the mechanisms driving ARDS.

Genes Influencing Respiratory Epithelial Function
The first gene described to influence ARDS susceptibility was surfactant protein B (SFTPB). Building on the finding that congenital alveolar proteinosis, a rare familial form of neonatal respiratory distress, is due to coding region mutations in SFTPB, 50,51 Lin et al genotyped tagging SNPs in the genes encoding surfactant proteins A, B, and D and found a common nonsynonymous coding variant in SFTPB exon 4 (rs1130866) to associate with ARDS compared both to healthy and at-risk controls. 52 Subsequent studies have confirmed the association of either the coding variant or an intronic SFTPB insertion/deletion polymorphism with primarily infectionassociated ARDS in both adults and children, as well as with mortality among ARDS subjects. [53][54][55][56] Interestingly, SFTPB variation may influence only pulmonary-specific risks for ARDS (►Fig. 2); in two of the studies testing nonpulmonary and pulmonary risks for ARDS, only the subjects with pneumonia or aspiration as their ARDS risk factors demonstrated an association with SFTPB. 52, 55 Gong et al also reported an interaction between gender and the SFTPB polymorphism on ARDS risk, with women demonstrating a stronger effect of this variant, though that result has not been replicated. 55 The fact that the genetic risk of SFTPB is relatively specific to pulmonary forms of ARDS is particularly relevant given that the effect of exogenous surfactant repletion on reducing mortality in ARDS was also more pronounced for the pulmonary form. 57 Although other epithelial-specific genes have not been as well studied, the epidermal growth factor gene (EGF) has also been implicated in ARDS risk in one mixed intensive care unit (ICU) population, where the observation seemed to be limited to males at risk for ARDS. 58 Genes Influencing Immune Regulation ARDS has long been considered a syndrome of dysregulated inflammation, with a historical focus on neutrophilic inflammation given the observed accumulation of neutrophils in bronchoalveolar lavage (BAL) fluid and in the lungs of patients with ARDS. 59 Furthermore, global gene expression profiling of many animal models of lung injury reveal significant dysregulation of inflammatory genes. 60,61 Not surprisingly, the genes encoding inflammatory cytokines and those that modify response to injury or infection have been among the most replicated ARDS susceptibility genes. Interleukin-6 (IL6), the gene encoding the proinflammatory cytokine IL-6, an abundant molecule in the plasma and BAL fluid of patients with ARDS, demonstrates consistent association with ARDS in European Americans. [62][63][64] In a gene-based replication study, IL6 haplotypes were also associated with ARDS following trauma among African Americans. 49 Four studies in individuals of European ancestry have found an association between a promoter polymorphism in the interleukin 10 (IL10) gene associated with higher levels of the anti-inflammatory cytokine IL-10 and the risk for ARDS, though the same allele was associated with decreased ARDS risk in one population 65  increased ARDS risk in the others. 49,66,67 The same polymorphism has demonstrated conflicting results with respect to sepsis susceptibility. 68,69 Although alleles that demonstrate conflicting directions of association for the same phenotype may represent spurious findings, alternative explanations are that the "flip-flop" in odds ratios masks differential genetic background or population substructure, different environmental interactions, or unrecognized clinical differences in the populations being studied. 70 Interestingly, the same study that found the G allele to associate with increased ARDS susceptibility found that among subjects who developed ARDS, the high IL-10-expressing rs1800896G allele was associated with decreased severity of illness scores and decreased mortality, suggesting that this polymorphism may have pleiotrophic effects. Given the interplay between host immunity and risk for infection, it may also be that overexpressing an anti-inflammatory molecule such as IL-10 could predispose to developing more severe infection, yet mitigate complications relating to excessive inflammation. A similar observation has been reported in severe sepsis with respect to the CXCL2 gene, whereby a promoter short tandem repeat seems to associate with increased susceptibility to sepsis but a decreased mortality among septic patients. 71 Our group recently reported an association between reduced ARDS risk and an SNP in the IL-1 receptor antagonist gene (IL1RN). 72 The association was present in two trauma populations and a large mixed ICU population, and, although the SNP is synonymous and does not alter the protein coding sequence, it was associated with higher plasma levels of IL-1 receptor antagonist (IL-1RA). This raises the possibility that higher circulating IL-1RA might help protect against ARDS, which is consonant with the finding that nonsurviving patients with ARDS had lower lung edema fluid IL-1RA levels than survivors. 73 In one small study of trauma subjects, an IL-8 promoter variant (rs4073A) was associated with increased ARDS incidence, and homozygous A carriers had persistently elevated levels of plasma IL-8 compared with homozygous TT individuals. 74 This finding has not yet been replicated in additional trauma or nontrauma populations.
Another innate immune gene that demonstrates a replicated association with ARDS is the mannose-binding lectin (MBL2) gene. Mannose-binding lectin protein (MBL) binds to ligands expressed by many microorganisms and triggers complement activation and opsonization. 75 Reduced circulating levels of MBL are associated with increased susceptibility to infection and pediatric sepsis, and the level of circulating MBL seems to be largely determined by MBL2 genotype. [76][77][78][79][80] Because there is an association between low-MBL-expressing genotypes and a decreased risk for some mycobacterial infections, it may be that low MBL levels are an infection risk early in life, before adaptive immunity is mature, but may offer an advantage in adulthood, 81,82 and this may explain the persistence of low-MBL-expressing variants in some populations. The level of MBL also influences the secretion of other cytokines, decreasing tumor necrosis factor-α (TNF-α) and stimulating IL-10. 81 Gong and colleagues tested critically ill patients at risk for ARDS for MBL2 genotypes and found a strong signal for ARDS susceptibility among patients homozygous for one of the low-MBLexpressing variants, codon 52B. The same variant associated with a higher severity of illness on admission, higher risk of septic shock, and, among ARDS patients, increased organ failure and mortality. 83 Additional studies have reported an association between low-MBL-expressing MBL2 genotype and pneumonia, 84,85 although the association may vary by microorganism. 86 Importantly, of the four studies examining the association between MBL2 and genotype, one was performed in a Chinese population; all others studied exclusively European ancestry subjects. The lack of ancestral diversity among genetic association studies of ARDS has been a historic shortcoming of the field, though recent studies are attempting to redress this gap. 37

Trauma-associated ARDS
Mixed ICU ARDS Fig. 2 Genetic variants differ by acute respiratory distress syndrome (ARDS) at-risk population. Whereas some genes like IL6 and IL10 seem to influence ARDS susceptibility in the context of multiple potential precipitants, others have been associated with only sepsis-associated or traumaassociated ARDS. More large-scale replication studies are necessary to conclude that a gene is truly specific to one ARDS endophenotype, however. In some cases, the gene simply has not been tested in all types of ARDS. Interest in the role of both inflammation and apoptosis in lung injury pathogenesis led one group to test the FAS gene for association with ARDS, and they reported an association between a common FAS haplotype and ARDS in two populations. 88 The same haplotype was associated with higher stimulated whole blood FAS mRNA expression in response to endotoxin stimulation. 88 With experimental evidence supporting strong roles for FAS driving both inflammatory and apoptotic roles in lung injury, [89][90][91] it remains uncertain whether FAS genetic variation preferentially influences inflammation, apoptosis, or both processes.
Multiple members of the toll-like receptor (TLR) signaling pathway have also been implicated in ARDS susceptibility, particularly for septic populations. Wurfel and colleagues performed an elegant study to screen for genetic variants associated with variation in response to TLR agonists, and found that the TLR1 upstream SNP rs5743551 associated with higher TLR1-mediated cytokine response was also associated with gram-positive infection, sepsis-associated organ dysfunction including ARDS, and death. 43 In functional assays, the authors demonstrated that a nonsynonymous coding SNP (rs5743618) in LD with rs5743551 (r 2 ¼ 0.76) resulted in higher TLR1 expression on the cell surface of peripheral monocytes as well as markedly higher nuclear factor kappa B (NF-κB) expression in response to stimulation of TLR2/1 heterodimers. 43 Coding variants in TLR1 have now been reported to associate with septic outcomes in three distinct populations. 43,92 The association between TLR1 variation and ARDS has not been replicated in nonseptic populations at risk for ARDS, though rs5743618 did show a signal in the same direction in a small European ancestry cohort of trauma subjects at risk for ARDS (additive model odds ratio 1.55, 95% confidence interval 0.96 to 2.48, p ¼ 0.07), 49 suggesting that TLR1 variation may be relevant in noninfection-related ARDS as well.
Additional pathway members of TLR signaling may also influence ARDS susceptibility. Mal protein, encoded by the gene TIRAP, is an important adaptor protein in the activation of TLR2 and TLR4 signaling. 93 Coding variants in TIRAP have been implicated in altering TLR2/4-stimulated cytokine profiles and susceptibility to infections and sepsis, whereby high cytokine-producing individuals carrying rs8177374T may have some protection from infections, but homozygous TT individuals may be predisposed to septic shock due to an excessive cytokine release. [94][95][96][97] Among Chinese subjects, in whom rs1877374 is very rare, there was no association between this variant and sepsis or ARDS, but tagging variants in the 3' end of TIRAP were strongly associated with both severe sepsis and sepsis-associated ARDS. 98 The association between TIRAP and ARDS serves as a potent example of the need to test whether genetic variants maintain similar relationships in diverse populations, and provide rationale to attempt replication at the gene level rather than at the level of the individual SNP or variant. 33,49 IL-1 receptor associated kinase 3 (IRAK3), another member of the TLR signaling pathway, has been associated with ARDS susceptibility in two diverse populations: a European septic case population and an African American trauma cohort. 49,99 The gene product of IRAK3, IRAK-M, is upregulated in some septic populations and animal models and seems to influence TLR ligand tolerance, perhaps predisposing to sepsis-induced immunosuppression. [100][101][102] Downstream of TLR activation, the transcription factor NF-κB is typically bound in the cytoplasm and kept inactive by its inhibitor NFKBIA (also called inhibitor kappa B-α, IκB) under unstimulated conditions. 103 Two studies have reported association between NFKBIA variation and ARDS in European populations, 49,104 though the functional implication of NFKBIA variation is not yet fully elucidated. A structural variant in a different gene in this pathway, NFKB1, which encodes the precursor for the repressive NF-κB subunit p50, has also been reported to associate both with risk for ARDS among nonelderly subjects and with ARDS severity as measured by lung injury score. 105,106 The variant tested was a four-base-pair deletion in the NFKB1 promoter, which is associated in vitro with diminished p50 expression, 107 though functional studies in critically ill patients are thus far unavailable.
Additional cytokines have also been associated with ARDS susceptibility. The gene encoding pre-B cell colony enhancing factor or visfatin (NAMPT, also denoted PBEF), is considered a proinflammatory adipocytokine that augments IL-1β-mediated inflammation and contributes to delayed neutrophil apoptosis. 108 Grigoryev and colleagues identified PBEF as consistently upregulated across multiple animal and cell culture models of lung injury, 109 prompting Ye et al to sequence the gene and look for an association with sepsis and sepsis-induced ARDS. 110 The PBEF promoter SNP (T-1001G) and its haplotype demonstrate a reproducible association with increased risk of sepsis-associated ARDS 111,112 and may increase mortality. 111 Single-population studies have also reported association between ARDS susceptibility and the cytokines tumor necrosis factor-α (TNF) 113 and macrophage migration inhibitory factor (MIF). 114

Genes Influencing Endothelial Function
Permeability of the alveolocapillary membrane is a pathophysiological hallmark of ARDS. Accordingly, vascular barrierregulating genes have been attractive candidates for ARDS susceptibility, and several have demonstrated reproducible associations with ARDS. The angiotensin-converting gene (ACE) was originally reported to strongly associate with ARDS susceptibility with an odds ratio > 2.6, 115 though two subsequent studies have failed to replicate that association. 116,117 It seems that the ACE insertion/deletion polymorphism influences mortality in ARDS, 115 at least in the era before widespread adoption of low stretch ventilation, because an association between the high-ACE-expressing DD genotype and death from ARDS was reported in three populations. The mortality was > 50% overall for the two studies reporting a link between ACE-DD and ARDS, whereas more recent interventional trials, employing a more highly screened population, report placebo-arm mortality of only $ 20%. 116,[118][119][120] The myosin light chain kinase gene MYLK, which encodes both smooth muscle and non-smooth-muscle myosin light chain kinase proteins (smMLCK and nmMLCK) as well as Telokin protein, has also demonstrated association with ARDS of multiple etiologies. There is ample evidence supporting a critical role for nmMLCK in vascular barrier regulation. [121][122][123] Nonsynonymous coding variants in the non-smooth-muscle region of MYLK have been associated with both sepsis and sepsis-associated ARDS as well as with ARDS following trauma. 124,125 The functional effects of these coding variants remain unknown. Importantly, the ARDS-associated allele was different between ARDS incited by sepsis compared with trauma, and it remains unclear whether the "flipflop" in directionality of association represents different population structure, confounding with a clinical variable such as blood transfusion, or basic mechanistic differences between different ARDS subtypes. 125 An improved understanding of nmMLCK's role in different subtypes will be imperative if the isoform is to be a target of drug discovery. 42,126 Two members of the angiogenesis pathway have also demonstrated association with ARDS. Vascular endothelial growth factor A (VEGF), encoded by VEGFA, potently induces microvascular permeability. 127 However, VEGF also appears to play a protective role for alveolar epithelium, with low levels of lung VEGF during ARDS and rising levels as patients recover. 128,129 A promoter SNP in VEGFA (rs3025039, C þ 936 T), which has been associated with both reduced plasma levels of VEGF and with reduced BAL levels of VEGF, 130,131 has been associated with ARDS susceptibility and mortality. 132,133 In a gene-based replication study of trauma-associated ARDS, the promoter region was again associated with ARDS among European ancestry subjects, whereas a haplotype at the 3' end of the gene was associated with ARDS among African ancestry subjects, again highlighting the necessity to test genetic associations across varied populations. 49 Another angiogenesis gene, angiopoietin-2 (ANGPT2), associates with ARDS susceptibility in both European and African ancestry populations. An increased risk for ARDS with intronic ANGPT2 variation has been reported in three populations; one population was predominantly European ancestry and septic, 134 whereas the others were African ancestry and European ancestry trauma populations. 37 In African ancestry trauma subjects, ANGPT2 variants emerged as highly associated with ARDS (p < 10 À4 ) using a multiplex moderate throughput genotyping platform of almost 50,000 SNPs. 37 The functional effect of intronic variation is not clear, though carriers of the ARDS-associated variant also demonstrated higher plasma levels of fulllength angiopoietin-2 protein (ANG2). 37 With in vivo and in vitro evidence demonstrating a barrier-disruptive effect of ANG2 on respiratory endothelium and epithelium, which can be rescued by antagonizing ANG2, 135-140 the angiopoietin axis may evolve to a pharmaceutical target for sepsis and ARDS. 141 Members of the coagulation cascade are critical mediators between endothelium and circulating inflammatory cells, and dysregulated coagulation is frequently observed in ARDS. 67,[142][143][144] Several coagulation genes have been associated with mortality among ARDS patients. A rare haplotype in the urokinase gene PLAU was associated with 60-day mortality and ventilator-free days in a European ancestry septic population. 145 The factor V Leiden mutation, a nonsynonymous coding substitution in the coagulation factor V gene (F5), which increases the risk of thrombotic complications, was surprisingly associated with decreased mortality from ARDS in one German population. 146 A more robust association has been reported between mortality and an insertion/ deletion polymorphism in the plasminogen-activating inhibitor (PAI-1, encoded by the gene SERPINE1) promoter. The single base-pair insertion/deletion polymorphism results in a 5-guanine (5G) or 4-guanine (4G) allele, with 4G being the minor allele state in European ancestry populations with a frequency of $ 46%. 147 The 4G allele associates with higher plasma levels of circulating PAI-1 due to diminished transcriptional repression relative to the 5G allele, resulting in a higher basal transcription rate. 148 In addition to associating with pneumonia in both pediatric and elderly patients, the 4G allele has been associated with mortality, multisystem organ failure, and ARDS among patients with infection. [149][150][151][152][153] Genes Influencing Injury: Oxidative Stress In addition to the broad categories already defined here, several genes in the oxidant stress pathway have been implicated in ARDS susceptibility. Reactive oxygen and nitrogen species incite endothelial and epithelial permeability in vitro, and antioxidants can mitigate lung injury in animal models. 154 Furthermore, the alveolar fluid of patients with ARDS has higher levels of nitrite and nitrate than those of hydrostatic pulmonary edema, suggesting that reactive nitrogen species play a role in the pathogenesis of ARDS. 155 The transcription factor NF-EF related factor 2 (NRF2), encoded by the NFE2L2 gene, was identified as a candidate gene through linkage analysis and positional cloning of hyperoxia sensitivity in mice. 156 Marzec and colleagues subsequently sequenced the NFE2L2 promoter and found rs6721961 (-617 C/A) to associate both with diminished promoter activity as well as with a strongly increased risk for trauma-associated ARDS. 157 The same SNP influences vasodilator responses as measured by forearm blood flow. 158 In a similar design, a functional promoter SNP in human NAD(P)H:quinine oxidoreductase 1 (NQO1) was associated with reduced NQO1 expression but with reduced risk of trauma-associated ARDS. 159 The authors hypothesized that while NQO1 expression is upregulated in response to oxidative stress and typically functions as an antioxidant, there are some contexts in which NQO1 may itself be pro-oxidant, which could contribute to ARDS susceptibility. 159 It is not yet known if either NFE2L2 or NQO1 variation influences nontrauma ARDS. A haplotype in the extracellular superoxide dismutase gene SOD3 demonstrated marginal association with infection-associated ARDS susceptibility compared with healthy controls. 160 More convincing was the association between an alternative, rare SOD3 haplotype, occurring in < 4% of European subjects, which exhibited a reproducible association with reduced mortality and reduced ventilator days compared with noncarriers with sepsis or ARDS. 160 The functional consequences of the haplotype remain unknown.
Because iron catalyzes several reactions that produce reactive oxygen or nitrogen species, 161 the family of iron metabolism genes has also been investigated for association with ARDS susceptibility. In a case-control population, Lagan and colleagues reported association between an SNP in the promoter region of ferritin light chain (FTL) and increased risk of ARDS compared with healthy control subjects. 162 In the same population, the authors reported reduced risk for ARDS among carriers of a haplotype in the heme oxygenase 2 (HMOX2) gene, which encodes for the constitutively active heme degradation enzyme. 162 Both the FTL and HMOX2 associations appeared to be more pronounced in subjects with pulmonary-specific ARDS risk. Sheu and colleagues focused on the inducible form of heme oxygenase and genotyped HMOX1 in a large population of European ancestry subjects with ARDS compared with at-risk critically ill controls, and found that longer copies of a short tandem repeat were associated both with decreased risk of ARDS and with increased plasma HMOX1 levels. 163 A subsequent study in a mixed ICU population replicated the association between long copies of the tandem repeat and plasma HMOX1, and reported reduced multiorgan dysfunction in carriers of the long tandem repeat. 164 The role of iron metabolism in ARDS pathogenesis may be a fertile line of investigation in the future.

Using Genetics to Inform ARDS Mechanistic Targets
The vast majority of the genetic associations with ARDS presented in the preceding text arose as hypothesis-driven association studies in candidate genes, based on the current understanding of ARDS pathogenesis. A few genetic risk factors for ARDS have been discovered using large-scale genotyping approaches. 37,49 The first published GWAS in ARDS was published in 2012. Christie et al reported the association of an SNP in the gene liprin-α (PPFIA1) with trauma-associated ARDS based on a association in 2 populations and evidence that the SNP influenced gene expression levels of its transcript in peripheral blood mononuclear cells. 67 Liprin-α was not previously an ARDS candidate gene, and the Christie study has highlighted the possibility for genetic studies to stimulate further mechanistic work, even for complex multigenic traits. 41 In addition to the PPFIA1 variant, 158 SNPs replicated their association with ARDS in a case-control trauma population at risk for ARDS, and 19 of those occurred in genes already hypothesized to play a role in ARDS, including the thrombospondin and chitinase families. 67 The Christie GWAS thus generates a rich list of potential new ARDS candidate genes to be prioritized in future ARDS research.
Another example of genetics informing mechanism comes from a study by Wang et al to evaluate the circulating gene expression differences between patients with acute and convalescent ARDS, which identified pre-elafin as strongly dysregulated between the phases. 165 The pre-elafin gene (PI3) was expressed at threefold lower levels in the blood of patients with acute ARDS compared with the same patients' convalescent blood, and plasma levels of PI3 protein were low for subjects with ARDS. 165 The same group associated a nonsynonymous SNP in the P13 gene with both increased ARDS susceptibility and higher plasma elfin levels during ARDS. 166 PI3 functions as a low-molecular weight neutrophil elastase inhibitor produced locally at the site of neutrophil activation in the lung. 167 Animal evidence supports a protective role against lung injury for pre-elafin and/or elafin, though in humans, the effect of the human neutrophil elastase inhibitor sivelestat seems limited to a short-term improvement in oxygenation, without affecting survival. 168 It has yet to be tested whether genetic variation in PI3 interacts with sivelestat treatment.
Novel ARDS candidate genes may also arise from animal or in vitro models of lung injury. The gene PBEF, now a replicated candidate gene influencing ARDS susceptibility, was first identified through a bioinformatic approach to select genes with consistent gene expression changes across multiple models of lung injury, performed in multiple species (mouse, rat, dog, and human cells). 109 The same study was influential in highlighting genes with mechanistic importance to stretch-induced or endotoxin-mediated lung injury. 61 Type 2 deiodinase (encoded by the gene DIO2), an enzyme involved in thyroid hormone metabolism, was identified in a similar manner from animal models, and a nonsynonymous coding SNP in DIO2 displayed decreased risk of sepsis-associated ARDS in one population. 169 Another group took the interesting approach of performing GWAS upon more than 20 strains of inbred mice, using the strain-specific sensitivity to lung insults-oxidant stress, acrolein, chlorine-as the trait of interest. [170][171][172][173] Injury-prone strains were compared with injury-resistant, and this approach has identified both accepted (SFTPB) and novel (ACRV1, activin A receptor 1) candidate genes. 170,172 In a similar approach, positional cloning studies focused on mouse strain susceptibility to hyperoxia identified the oxidative stress genes NFE2L2 and NQO1 as ARDS susceptibility factors. 157,159 Even when a candidate gene implicated in animal or in vitro studies fails to associate with ARDS susceptibility or outcome in human populations, these investigations often focus attention on critical gene or protein pathways that were previously overlooked.

Ancestry-Specific ARDS Factors?
A discussion about the genetic heterogeneity underlying ARDS susceptibility would be incomplete without some attention to potential ancestry-specific differences in ARDS risk. There is no strong evidence to suggest that ancestry itself is differentially associated with ARDS, though the published literature is scant in this field. 174 There have been reports that Americans of non-European ancestry suffer a higher mortality with ARDS, though some of this excess mortality was Seminars in Respiratory and Critical Care Medicine Vol. 34 No. 4/2013 explained by a higher severity of illness on admission. 175 Because the vast majority of genetic association studies have been performed in either predominantly or exclusively European ancestry populations, there are limited data about which genetic variants may be specific to any ancestry. As shown in ►Table 2, only nine genes have been reported in African ancestry populations and three have been reported in Asian populations. However, the number of ancestry-specific genetic risk factors is likely to grow as more populations are tested, especially considering that genes which influence individual response to injury or infection seem to have undergone selection. 16,95,176 One very interesting gene in this regard is Duffy antigen/ receptor for chemokines (DARC). The so-called Duffy null polymorphism, a promoter SNP that results in loss of expression of the Duffy antigen on the erythrocyte membrane, is a marker so divergent between African and European ancestries that it is frequently used by genotyping platforms as an ancestry informative marker. 25 African Americans are reportedly 60 to 80% homozygous for the Duffy null SNP. 25,87 Because the Duffy antigen is the erythrocyte receptor for Plasmodium vivax, the high frequency of the Duffy null polymorphism in African populations is considered an example of selective pressure shaping genetic architecture. 177 However, in addition to recognizing malarial species, the Duffy antigen recognizes and scavenges various CXC and CC chemokines, and experimentally, the Duffy null state promotes transfusion-related lung injury. 178,179 Among African ancestry subjects enrolled in ARDS network studies, mortality was significantly increased for Duffy null subjects compared with those with at least one copy of the rs2814778T allele. 87 Duffy null subjects also had higher levels of circulating IL-8, consistent with the role of Duffy antigen as a chemokine scavenger. Particularly notable was that this association was robust to adjustment for other ancestry informative markers, and that the effect of the polymorphism is large enough-at least in one population-to be detected in a study of only 132 subjects. 87 Whether DARC variation influences susceptibility to ARDS remains unclear; in one study of trauma-associated ARDS in a cohort of African Americans, DARC did not emerge as strongly associated with ARDS incidence. 37 Future Directions: Can Genes Help to Personalize Therapy?
The critical care community lacks a specific pharmacological therapy for ARDS. Only one agent has shown a mortality reduction in a randomized, controlled trial, and that trial was confined to pediatric patients. 57 Furthermore, with recent trials reporting mortality rates of only 20% in placebo arms, the sample size needed to show a benefit will continue to grow. 119,120 However, it may be that genetic risk factors could give some clue as to which patients are likely to respond to which agent. Combined with changes in gene or protein expression, DNA variants could stimulate a new era of molecular phenotyping in ARDS, which in turn could facilitate biomarker development and the translation of molecular profiles into therapeutic targets. 42 As mentioned, a preparation including surfactant protein B reduced mortality among pediatric ARDS patients, and SFTPB polymorphisms have been implicated in ARDS risk and outcomes in both European and African populations. [52][53][54] Both the genetic risk factor and the benefit of exogenous repletion seem to be specific to direct pulmonary injury. Could it be that SFTPB polymorphism carriers with pneumonia-associated ARDS stand the most to gain from exogenous surfactant repletion? A similar story may emerge for subjects with markers for an endothelial-predominant lung injury, with variation in ANGPT2, VEGFA, ACE, or MYLK. Perhaps these are the patients most likely to benefit from statin therapy, or from newly developed antipermeability agents. 139,180 Genetic risk factors may then be harnessed as molecular diagnostic tools to help personalize ARDS therapy. As we eagerly await such developments, genetic associations are already yielding important mechanistic insights that refine our understanding of ARDS.