While the impact of HDL levels on cardiovascular health is well recognized, its role in the immune response is less well known. This paper discusses the causes of the decrease in HDL-C in septic patients and describes the genetic mechanisms that underlie the variability of that decrease. Specifically, it describes how a genetic variation in the cholesteryl ester transfer protein (CETP) affects the decline in HDL-C as well as survival in a sepsis cohort.
High-density lipoprotein (HDL) cholesterol (HDL-C) levels are inversely correlated with cardiovascular disease. However, negative results from clinical trials (1, 2) and genetic studies (3) have cast doubt on the causal role of HDL-C in atherosclerosis, raising the question of the true physiological function of HDL.
Beyond its established role in cholesterol metabolism, HDL is also an important component of the innate immune system by virtue of its ability to sequester pathogen lipids from invading bacteria (4-7). HDL-C levels drop acutely during sepsis, and lower HDL-C levels are associated with increased risk of multiorgan dysfunction syndrome, prolonged hospital admission, and death (8-15). In the general population, low levels of HDL-C are associated with increased risk of infection (16). Preclinical animal models support a causal role of HDL in sepsis pathogenesis. Mice receiving HDL infusions and transgenic mice overexpressing human apolipoprotein AI (the predominant protein component of HDL) have reduced inflammation and improved survival rates after intraabdominal sepsis (17). In contrast, mice deficient for apolipoprotein AI have low HDL-C levels and exacerbated sepsis (18).
Collectively, these observations suggest that HDL may play an important role in the response to infection. However, the underlying mechanisms responsible for the decrease in HDL-C levels during sepsis, and for the interindividual variability in the magnitude of decline, are incompletely understood (5, 9, 19). Furthermore, whether the association between HDL-C and infectious outcomes is causal or correlative remains unknown. We hypothesized that genetic variation in genes known to modulate HDL metabolism would be associated with changes in HDL-C levels during sepsis, and that genetically determined changes in HDL-C levels would be associated with clinical outcomes. Some of the results of these studies have been previously reported in the form of an abstract (20).
Studies were approved by the University of British Columbia and Providence Health Care research ethics boards. Patients or substitute decision-makers provided written informed consent.
Early Infection Cohort
The Early Infection cohort has been previously reported, and consists of 200 adult patients presenting to St. Paul's Hospital Emergency Department (Vancouver, BC, Canada) from January 2011 to June 2014 with sepsis (see the supplemental Methods and Table E1 in the online supplement) (9, 12).
Early Infection Cohort DNA Library Preparation and Next-Generation Sequencing
DNA was extracted from 200 ml of buffy coat using DNeasy Blood and Tissue kits (catalog no. 69582; QIAGEN). Library preparation, targeted resequencing on the MiSeq instrument, and quality filtering was performed as previously described (21) (see also the Methods in the online supplement). We analyzed potentially functional variants with mean allele frequency less than 5% in the ABCA1, APOA1, APOA2, CETP, GALNT2, LCAT, LIPG, NPC1, PLTP, and SCARB1 genes.
Vasopressin versus Norepinephrine in the Severe Septic Shock Trial Cohort
VASST (Vasopressin versus Norepinephrine Infusion in Patients with Septic Shock Trial) was a multicenter, double-blinded, randomized, control trial of 778 adult patients with septic shock and need for vasopressors (22). Genotyping was performed as previously described in 632 patients (23).
St. Paul's Hospital Intensive Care Unit 2 Cohort
A subset of 203 patients from the previously reported SPHICU2 (St. Paul's Hospital Intensive Care Unit 2) cohort with DNA available were genotyped for candidate genetic variants by Genome Quebec on the Agena iPLEX platform (24).
We used summary statistics of additive perallele ß coefficients and standard errors from the MAGNETIC NMR GWAS (Nuclear Magnetic Resonance GenomeWide Association Study) dataset (25) to assess the effects of rs1800777 carrier status on 111 metabolic measures of ambulatory patients without sepsis (see supplemental Methods).
A cohort of ambulatory patients without sepsis with biobanked plasma was genotyped using TaqMan genotyping assays (see supplemental Methods).
Blood Collection and Lipid Measurements in the Early Infection Cohort
Blood was collected at the time of the first clinical encounter in the emergency department, and a complete lipid panel was performed on an ADVIA 1800 Chemistry System (Siemens) (9). Where available (48% of patients), another sample was taken on Day 7 (±3) of admission. From chart review, we determined HDL-C levels between 5 years before and 2 years after (but not within one month of) the index sepsis episode, referred to as the "baseline" level.
CETP Activity Assays
CETP activity assays were performed on plasma samples (catalog no. MAK106; Sigma-Aldrich).
Analyses were performed with Prism 7 (GraphPad Software, Inc.), unless otherwise stated. Fischer's exact tests were used for all contingency analyses. Normally distributed data were analyzed with unpaired t tests, whereas nonnormal data were analyzed with Mann-Whitney tests. SNPs were tested for Hardy-Weinberg equilibrium using a x2 test. Multiple testing correction for geneand variant-based HDL-C associations used the Bonferroni method.
Survival analyses were analyzed by logrank tests and Cox regression models using the "survival" R package. Meta-analyses were performed using the "metafor" R package. Variant(s) identified to be associated with HDL-C levels were used as an instrumental variable in a Mendelian randomization analysis on 28-day survival using the "MendelianRandomization" R package (v0.2.2). Statistical significance was claimed when two-sided P values were less than or equal to 0.05.
A Genetic Variant in CETP Predicts Very Low HDL-C during Sepsis
To identify genes associated with changes in HDL-C levels during sepsis we performed targeted resequencing of candidate genes involved in HDL metabolism (21, 26). We compared subgroups of patients defined by an HDL-C of 0.65 mmol/L or less versus a value greater than 0.65 mmol/L at the time of presentation with sepsis, based on prior evidence that this cut-off value of HDL-C strongly predicts sepsis survival in this cohort (9). Only the CETP gene displayed a significant (P = 0.006 unadjusted; P = 0.022 Bonferroni corrected) difference in the frequency of rare variants between the lowand high-HDL-C subgroups (rare variant detected in 14.7% of patients with HDLC C 0.65 mmol/L vs. 3.0% of patients with HDL-C > 0.65 mmol/L; P < 0.01) (Figure 1).
Out of the 14 patients with rare CETP variants, 10 patients were heterozygous (G>A) for a coding, missense SNP (rs1800777 G>A, p.Arg468Glu, mean allele frequency = 0.03) that was in HardyWeinberg equilibrium (Table E2). Among carriers of rs1800777, 9/10 (90.0%) had HDL-C levels of 0.65 mmol/L or less at the time of presentation to the emergency department, compared with 81/190 (42.6%) of noncarriers (P < 0.01, Figure 2A). We also investigated the association between genetic variation and HDL-C during the early stages of sepsis by performing a linear regression of continuous, ln-transformed HDL-C levels against rare variants in HDLrelated genes that was adjusted for APACHE (Acute Physiology and Chronic Disease Health Evaluation) II score and sex. Only the rs1800777 variant was significantly associated with HDL-C levels (raw P = 0.006, adjusted P = 0.03; Table E3), providing additional evidence for the role of this variant in HDL-C levels during sepsis and indicating that this finding was robust as to whether HDL-C levels were considered as a continuous or dichotomous trait.
Carriers of rs1800777 had a significantly greater decline in HDL-C during hospitalization for sepsis compared with noncarriers at all time points measured (Figure 2B). At the time of presentation to the emergency department, 9 out of 10 (90.0%) carriers of rs1800777 had HDL-C levels below the ambulatory populationbased fifth percentile for age and sex, compared with 83/190 (43.6%) of noncarriers (P < 0.01; Figure 2C). Carriers of rs1800777 demonstrated slightly, statistically nonsignificant lower levels of HDL-C at baseline (i.e., before or after sepsis), relative to noncarriers (means 6 SD: 1.26 6 0.47 vs. 1.40 6 0.61 mmol/L; Figure 2B), with a magnitude of difference comparable to that observed in a prior study of this variant in healthy individuals (27). This difference became much more profound during sepsis, with carriers of rs1800777 displaying a significantly larger intraindividual decrease in HDL-C compared with noncarriers (P < 0.001; Figure 2D).
The demographics of individuals with and without rs1800777 at the time of their presentation to the emergency department are shown in Table 1. There was no significant difference between rs1800777 carriers and noncarriers in demographic variables, preexisting comorbidities, or baseline laboratory values, except for lipid levels at the time of their presentation to the emergency department (Table 1). Both total cholesterol and HDL-C were significantly lower in carriers of the rs1800777 relative to noncarriers.
rs1800777 Is a Gain-of-Function Variant
Based on the lower levels of HDL-C in carriers of rs1800777, we hypothesized that this represented a gain-of-function variant. To investigate this, we explored the MAGNETIC NMR GWAS dataset to assess the effects of rs1800777-A allele carriage on 111 metabolic lipid measures in a large cohort (n 20,000) of ambulatory patients without sepsis (25) (Table E4). The allele frequency of rs1800777 was 0.03. Total cholesterol in HDL showed the largest negative effect size between rs1800777 carriers and noncarriers (ß [SE] = -0.255 [0.035]; P = 5.88 X 10 -13), along with reductions in other HDL-associated lipids, as well as apoAI (Figure 3). Smaller and opposite effects were seen in very lowdensity lipoprotein lipids (Figure 3). These findings are consistent with previous reports of rs1800777-A allele being a gainof-function variant in ambulatory patients without sepsis (27-29).
In the Early Infection cohort, carriers of rs1800777 displayed significantly greater levels of CETP activity (P < 0.01) compared with age-, sex-, and ethnicitymatched noncarriers at the time of presentation to the emergency department (Figure 4). We also determined CETP activity in rs1800777 carriers and matched controls from a cohort of ambulatory individuals without sepsis (Figure 4). Interestingly, the difference in CETP activity between rs1800777 carriers and noncarriers during the early stages of sepsis in the Early Infection cohort was much greater than that observed in ambulatory patients without sepsis (Figure 4), suggesting that rs1800777 may lead to an inability to downregulate CETP during sepsis, as is normally thought to occur (30-32). This hypothesis is supported by our observation that CETP activity in rs1800777 carriers continued to show a trend to be approximately 1.7-fold higher than age-, sex-, and ethnicitymatched noncarriers in the SPHICU2 cohort, a later time point in the disease process of sepsis when CETP activity should be notably suppressed (P = 0.06; Figure 4).
rs1800777 Is Associated with Increased Organ Failure and Decreased Survival
Because carriers of rs1800777-A have lower HDL-C levels during an infection, and because low HDL-C is associated with a poorer clinical outcome during sepsis (8-15), we next examined clinical outcomes in carriers of this variant versus noncarriers in the Early Infection cohort. Relative to noncarriers, carriers of rs1800777 had higher APACHE II scores, significantly increased prevalence of acute kidney injury (70.0% vs. 25.8%; P < 0.05), and significantly increased need for vasopressor support (50.0% vs. 15.8%; P < 0.05; Table E5) (12, 33).
Carriers of rs1800777 in the Early Infection cohort had a trend to decreased 28-day survival rates in comparison to noncarriers (P = 0.07, log-rank test; Figure 5A). The rs1800777 carrier genotype conferred an unadjusted hazard ratio of 3.59 (95% confidence interval [CI] = 0.80-16.02; P = 0.09) for mortality from sepsis, and a hazard ratio of 1.80 (95% CI = 0.39-8.20; P = 0.4) when adjusted for APACHE II score and sex (Figure 6; Table E6).
Because the Early Infection cohort was relatively underpowered for survival analysis (its utility was the prospective lipid analyses to associate with genetic data), we next examined the association of rs1800777 with survival in the larger VASST cohort (n = 632). Power analysis based on Early Infection cohort survival data determined that 521 patients (26 rs1800777 carriers GA, 495 noncarriers GG) would be required to detect a statistically significant difference in survival between rs1800777 carriers and noncarriers (a = 0.05, ß = 0.20). Relative to the Early Infection cohort, patients were enrolled in VASST later in their disease process, a median time of 12 hours after the development of septic shock and more than 24 hours after initial presentation. As such, the mortality rates in the VASST cohort are higher. The demographics, baseline clinical measurements, and baseline laboratory investigations of individuals with and without rs1800777-A in the VASST cohort are shown in Table E7.
The distribution of rs1800777 genotypes in the VASST cohort was in Hardy-Weinberg equilibrium, with 597 noncarriers (GG), 34 rs1800777 heterozygotes (GA), and 1 rs1800777 homozygote (AA) (Table E2). In VASST, carriers of rs1800777 had significantly decreased 28-day (log-rank test P = 0.007) and 90-day survival (log-rank test P = 0.04) relative to noncarriers (Figure 5B). The rs1800777 carrier genotype conferred a hazard ratio of 1.98 when unadjusted (95% CI = 1.26-3.14; P = 0.003) and 2.02 (95% CI = 1.28-3.18; P = 0.002) when adjusted for APACHE II score and sex (Figure 6; Table E6). To determine if this result was influenced by population stratification, we also performed adjustment for the first two principal components as a measure of genetic ancestry. When adjusted for APACHE II score, sex, and the first two principal components, the rs1800777 genotype conferred a hazard ratio of 2.05 (95% CI= 1.30-3.23; P = 0.002; Table E8).
As additional validation of this finding, we next examined the association between the rs1800777 variant and survival in the SPHICU2 cohort. The distribution of rs1800777 genotypes in SPHICU2 was in Hardy-Weinberg equilibrium with 191 noncarriers (GG) and 12 heterozygotes (GA) (Table E2). There were no significant differences between rs1800777 carriers and noncarriers in terms of demographics, baseline clinical measurements, and baseline laboratory investigations (Table E9). However, carriers of rs1800777 had significantly decreased 28-day survival (logrank test P = 0.02) and a trend to decreased 90-day survival (log-rank test P = 0.06) relative to noncarriers (Figure 5C). The rs1800777 carrier genotype conferred a hazard ratio of 2.70 when unadjusted (95% CI= 1.14-6.38; P = 0.02) and 3.12 (95% CI = 1.31-7.42; P = 0.01) when adjusted for APACHE II score and sex (Figure 6; Table E6).
Finally, we performed a fixed-effect meta-analysis of the effect of rs1800777 genotype on 28-day survival in the Early Infection, VASST, and SPHICU2 cohorts. The Q statistic for heterogeneity was 0.83 (P = 0.7), which suggested that a fixed-effect meta-analysis is valid. The meta-analysis found that rs180077 carrier status conferred a significant hazard ratio of 2.19 (95% CI = 1.48-3.23; P < 0.0001) when adjusted for APACHE II score and sex (Figure 6).
Causal Inference of rs1800777determined HDL-C on Sepsis Survival
The availability of a genetic instrument (e.g., rs1800777 genotype) associated with the change in HDL-C during sepsis provides the opportunity to determine whether the epidemiologically observed association between HDL-C and survival during sepsis represents a causal or a correlative relationship. To investigate this, we performed Mendelian randomization analysis, based on the principal that genotypes are randomly assorted at meiosis, analogous to the randomization in a clinical trial. This approach can therefore provide causal inference on an observational relationship between a biomarker (exposure) and a disease outcome of interest (outcome) using genetic variant(s) as an instrumental variable (34, 35). In the first step of the two-stage Mendelian randomization, we examined the effect of the rs1800777 risk allele carrier status (instrumental variable) on ln-transformed HDL-C levels (exposure) in the Early Infection cohort. The rs1800777 variant was used as a single instrumental variable, because it was the only variant significantly associated with sepsis HDL-C levels in the Early Infection cohort, which is the primary assumption of Mendelian randomization analysis (P = 0.0006; F statistic = 12.1; 5.7% variance explained; Table E10) (34). For the second step of the Mendelian randomization, we used the estimate of rs1800777-mediated HDL-C levels during sepsis to examine the effect of genetically lowered HDL-C on the outcome of 28-day survival in the VASST and SPHICU2 cohorts. We found that genetically lowered HDL-C was associated with significantly decreased survival from sepsis in both the VASST and SPHICU2 cohorts (1.38 ln and 2.30 ln increase in the adjusted hazard ratio per 1 ln reduction in HDL-C [ln mmol/L], respectively; P = 0.02; Figures E1A and E1B). Furthermore, an inverse varianceweighted meta-analysis of the VASST and SPHICU2 cohorts observed a 1.56 ln increase in the adjusted hazard ratio per 1 ln reduction in HDL-C (ln mmol/L; P < 0.001; Figure E1C).
Here, we report that genetic variation in CETP influences HDL-C levels and clinical outcomes during sepsis. Importantly, the rs1800777 gain-of-function variant in CETP was associated with elevated CETP activity, exacerbated decline in HDL-C, and lower survival rates during sepsis. To our knowledge, this is the first report describing a genetic mechanism that contributes to changes in HDL-C levels during sepsis, establishing a gene-environment interaction for the observed drop in HDL-C observed in patients with sepsis. Mendelian randomization results were consistent with HDL playing a causal role in survival from sepsis. Our results provide strong evidence for an important role of genetically mediated changes in HDL-C via CETP in the outcome of sepsis. Our results lend support to the concept that a central function of HDL is to modulate immunity and infection (4-9, 13, 16, 17).
Previous studies have reported that rs1800777 increases CETP activity due to reduced degradation of the protein (29) and that carriers of rs1800777 have slightly lower levels of HDL-C than noncarriers (91.6-93.1% that of noncarriers) (27). Consistent with this, we found that carriers of rs1800777 had slightly lower HDL-C at baseline (i.e., before or after recovery from the episode of sepsis). However, rs1800777 was associated with a profound drop in HDL-C during sepsis. We did not observe association of other rare coding variants in HDL-related genes with HDL-C levels during sepsis (Figure 1). Whether common variants in these genes, or in other genes, also contribute to changes in HDL-C during sepsis remains to be investigated.
CETP transfers cholesteryl esters from HDL to very low-density lipoproteins in exchange for triglycerides. High levels of CETP activity result in HDL that is depleted of cholesterol and more rapidly removed from the circulation, leading to lower HDL-C levels (36, 37). CETP expression and plasma CETP activity are inhibited by LPS (30-32), which may mitigate the decline in HDL that occurs during sepsis to promote HDL-mediated sequestration of pathogen lipids, or other immunomodulatory effects of HDL particles. HDL-sequestered LPS is partially cleared from the circulation through scavenger receptor class B type 1 -mediated hepatic elimination and detoxification (38-41). Our finding of increased CETP activity during sepsis in carriers of rs1800777 suggests that an inability to downregulate CETP activity may lead to the large drop in HDL-C levels in these patients, which could impair HDL-mediated sequestration and clearance of pathogen lipids.
Our findings are in contrast with reports that mice transgenic for human CETP have improved survival rates when subjected to the cecal ligation and puncture-induced polymicrobial sepsis (42) and intraperitoneal-induced endotoxemia (43). This may relate to differences between mice and humans in that mice have higher HDL and lower very low-density lipoprotein compared with humans (44, 45). In fact, expression of CETP in wildtype mice impacts HDL-C levels to only a small extent, because of lack of triglyceridedonor particles (46). Alternatively, differences between the mouse model of sepsis and the clinical condition in humans may account for this discrepancy. Our findings are in agreement with a human sepsis study showing that higher initial CETP levels predict death (47).
One CETP inhibitor, torcetrapib, significantly increased the risk of death from infection in the ILLUMINATE trial of patients with high cardiovascular disease risk (48), which would appear to be in contradiction to our finding that genetic CETP gain-of-function led to decreased sepsis survival. However, other large, randomized, controlled clinical trials of the CETP inhibitors, including dalcetrapib (2), evacetrapib (49), and anacetrapib (1), have not reported an increased risk of severe infection, suggesting that the findings from torcetrapib may relate to off-target pharmacology specific to that compound. Whether pharmacological inhibition of CETP would improve clinical outcomes in the context of sepsis merits further study.
Our study has some limitations. Most importantly, despite the use of three independent cohorts, the sample size of patients with the rare rs1800777 variant (mean allele frequency, ~3%) was relatively small, and future validation of these findings will be critical. Furthermore, lipid levels were not measured as part of the VASST protocol. In addition, in the Early Infection cohort, presepsis and postsepsis lipid levels were not available in all patients, and the indication for making these measurements was not known in all cases, which might have biased these results.
Limitations of our Mendelian randomization approach are the assumptions that the rs1800777 variant is: 1) not associated with confounders; and 2) influences the outcome of 28-day sepsis survival only through the risk factor of low HDL-C. For a single-instrument variable Mendelian randomization, these assumptions can only be argued theoretically (4). In support of these assumptions, we observed no significant differences in demographic variables and preexisting comorbidities in the Early Infection, VASST, or SPHICU2 cohorts between rs1800777 carriers and noncarriers. Our rs1800777 Cox proportional hazard ratio analyses adjusted for APACHE II and sex also supports the validity of this assumption. Second, we searched the Genotype-Tissue Expression database (https://gtexportal.org/), National Human Genome Research Institute-European Bioinformatics Institute GWAS catalog (https://www.ebi.ac.uk/gwas/), PubMed, and LitVar for reports of rs1800777 variant associations. To the best of our knowledge from the literature and data presented (25), the rs1800777 variant is associated with phenotypes related to decreased HDL and increased very low-density lipoprotein. Because increased very low-density lipoprotein is associated with protection from sepsis (6, 7, 50), the low HDL-C associated with rs1800777 appears to be the most likely explanation for the observation of poor sepsis outcomes in these patients. However, there remains the possibility of unbalanced horizontal pleiotropy biasing our results; for instance, if CETP influences survival from sepsis independently of its effect on HDL-C.
In summary, we report that genetic variation in CETP influences HDL-C levels and clinical outcomes in patients with sepsis. Our results provide insight into the mechanisms regulating HDL-C levels during acute infection and systemic inflammation, and establish that genetically determined changes in HDL levels are a critical determinant to sepsis survival.
Author disclosures are available with the text of this article at www.atsjournals.org.
Acknowledgment: The authors thank the patients and families who participated in both the Early Infection cohort and VASST (Vasopressin versus Norepinephrine Infusion in Patients with Septic Shock Trial) studies, which have helped increase the scientific and clinical understanding of sepsis. The authors also thank the Centre for Heart Lung Innovation's clinical research team for facilitating patient recruitment and data collection in the Early Infection cohort study. Data on metabolic measures were provided by MAGNETIC NMR GWAS (Nuclear Magnetic Resonance Genome-Wide Association Study) and downloaded from http://www.computationalmedicine. fi/data#nmr_gwas. The authors also thank Alexander Blauw for writing the Python script to design the circular figure.
Copyright American Thoracic Society Apr 1, 2019
This article was written by Liam Brunham, Kelly Genga, Patrick Rensen, Mihai Cirstea, Mark Trinder, Keith Walley, Cody Lo, HyeJin Kong, Lisanne Blauw, Yanan Wang, Xuan Li, James Russell and John Boyd from American Journal of Respiratory and Critical Care Medicine and was legally licensed through the NewsCred publisher network. Please direct all licensing questions to firstname.lastname@example.org.