HDL Cholesterol: A "Pathogen Lipid Sink" for Sepsis?

The lack of a specific medical therapy for sepsis, a dysregulated response to infection that is responsible for up to half of all inpatient deaths (1), plagues critical care. Numerous clinical trials have failed to improve mortality (2, 3). Furthermore, the failure of many anticytokine therapies challenges the classic paradigm of observing an association between plasma levels of a purported marker and sepsis mortality, testing the marker’s causality and modifiability in animal models, and then moving toward clinical trials to test marker blockade. Correlation does not equate with causation, and strategies to modify a correlated but noncausal biomarker are unlikely to improve sepsis survival. Although controlled interventional trials provide strong evidence for causality, it is frequently unethical or impractical to randomize subjects to high- or low-biomarker infusions, leaving us to bridge this gap with observational designs. Our field needs smarter tools to dissect correlation from causality and identify the causal biomarkers of sepsis to speed the development of sepsis therapy.

Fortunately, tools to infer causality from observational data do exist. Classically, such methods have been applied to avoid making policy decisions based on biased or inconsistent association estimates (4) due to measurement error, uncontrolled confounding, or reverse causation. One potential solution is to use an instrumental variable analysis. This approach is valid if the instrument, or reliably measured variable, has a strong association with a potential mediator variable, and there is no correlation between the instrument and the outcome being studied (5). For example, if distance from a grocery store reliably predicts intake of fresh produce, then the association between grocery store distance and lean weight can be used to infer whether produce intake has a causal relationship with weight. Genetic researchers extended this approach by using genotype(s) to predict biomarkers, testing the association between biomarker-predicting genotypes and disease, and inferring whether the biomarker has a causal relationship with disease. Called Mendelian randomization (MR) analysis, this genetic instrumental variable strategy is attractive because genotypes are assigned at random by gametogenesis and genotype assignment always precedes outcome. Such MR analyses have provided evidence that low-density lipoprotein cholesterol (LDL-C) plasma concentrations are causally related to cardiovascular disease and mortality (6), whereas markers such as C reactive protein are not (7), thus focusing therapeutic efforts on modifying plasma LDL-C. Furthermore, MR has enabled the identification of novel genetic regulators of LDL-C, which has translated to a new class of lipid-lowering agents (8).

Could similar strategies be applied to identify key causal intermediates for sepsis death? In this issue of the Journal, Trinder and colleagues (pp. 854–862) implicate serum high-density lipoprotein cholesterol (HDL-C) as a potentially causal contributor to sepsis survival, and suggest that medications that boost HDL deserve investigation for sepsis (9). The foundation for this work was the group’s prior observation that low HDL-C was a strong predictor of organ dysfunction or death among patients presenting to an emergency room with suspected sepsis (10). Because HDL-C can bind and sequester pathogen lipids, including endotoxin, patients with lower HDL-C may have worse sepsis outcomes. The authors used an astute approach to identify genetic predictors of serum HDL-C and performed an MR analysis of the effect of HDL-C on sepsis survival. First, they performed targeted resequencing of 10 HDL-C–associated genes in 200 subjects with suspected sepsis, focusing on SNPs that influence coding sequence or splicing. For each candidate gene, they tested whether subjects with low HDL-C had an excess of coding SNPs compared with subjects with normal or high HDL-C, and the gene CETP—encoding CETP (cholesteryl ester transfer protein)—was the only one to demonstrate an HDL-C association during sepsis. Furthermore, one missense CETP SNP, rs1800777, drove the association between CETP, HDL-C, and increased sepsis-related organ failures. The SNP seems to be a CETP gain-of-function variant, with rs1800777 carriers exhibiting higher plasma CETP activity. Because the sequencing was performed in the same 200 subjects in whom the authors first reported an association between low HDL-C and sepsis death, raising concerns about selection bias and generalizability, the authors validated that rs1800777 was associated with decreased sepsis survival in two additional sepsis populations. Finally, the authors used rs1800777 as a genetic instrument to predict a portion of HDL-C variance. By MR analysis, each log decrease in genetically predicted HDL-C during early sepsis was associated with an increase in the adjusted hazard ratio for mortality, leading to the causal inference that lower serum HDL-C during sepsis has a causal effect on reduced sepsis survival.

This study has several strengths, including its sophisticated design to test suspected functional genetic variants via a sequencing approach. By focusing on genome-wide association study–validated loci that influence HDL-C, the authors were more likely to discover a strong relationship between genotype and HDL-C, and they showed the SNP’s gain-of-function action by testing plasma levels of CETP activity. The consistency of the SNP’s association with both HDL-C and sepsis organ failure and survival in multiple populations lends confidence that rs1800077 is a valid genetic instrument for making a causal inference about HDL-C. Most importantly, by establishing serum HDL-C as a potential causal intermediate in sepsis survival, this study introduces HDL-C modification as a highly novel therapeutic strategy for sepsis, which is an exciting concept in that agents to inhibit CETP already exist.

Trinder and colleagues acknowledge that although their genetic instrument meets validity criteria, it is rare: only 10 of the 200 subjects in the early infection cohort carried this SNP, and small sample sizes are at risk for unstable effect estimates. However, the validation of the SNP–mortality association in additional sepsis populations is reassuring. HDL-C is less established than other potential sepsis prognostic biomarkers, and thus it will be important to ensure the consistency of this association in much larger populations. Finally, the authors acknowledge some inconsistencies in the data supporting a strategy of CETP blockade in sepsis, including worrisome observations of increased infections and excess mortality in randomized trials of one CETP inhibitor, torcetrapib, for coronary arterial disease (11). In addition, the recent disappointing results of the EUPHRATES (Evaluating the Use of Polymyxin B Hemoperfusion in a Randomized controlled trial of Adults Treated for Endotoxemia and Septic shock) trial, which randomized subjects with septic shock and elevated endotoxin activity assays to a hemofiltration therapy targeted at reducing endotoxin activity (12), likewise dampens enthusiasm for the notion that modifying endotoxin availability is a helpful approach in sepsis.

Although it remains to be seen whether CETP inhibition might be a viable therapeutic option in sepsis, the study by Trinder and colleagues is nonetheless a robust example of employing genomic and statistical tools in observational clinical cohorts to identify novel therapeutic targets in sepsis. Similar approaches should be embraced by investigators in our field, with dedicated attempts to replicate prior findings while generating new discoveries. The validation of causal intermediates should accelerate translation from observation to safe, testable interventions, ideally leading to improved sepsis therapy.