Evaluation of plasma sphingolipids as mediators of the relationship between kidney disease and cardiovascular events

Summary

Background

Sphingolipids are a family of circulating lipids with regulatory and signaling roles that are strongly associated with both eGFR and cardiovascular disease. Patients with chronic kidney disease (CKD) are at high risk for cardiovascular events, and have different plasma concentrations of certain plasma sphingolipids compared to patients with normal kidney function. We hypothesize that circulating sphingolipids partially mediate the associations between eGFR and cardiovascular events.

Methods

We measured the circulating concentrations of 8 sphingolipids, including 4 ceramides and 4 sphingomyelins with the fatty acids 16:0, 20:0, 22:0, and 24:0, in plasma from 3,463 participants in a population-based cohort (Cardiovascular Health Study) without prevalent cardiovascular disease. We tested the adjusted mediation effects by these sphingolipids of the associations between eGFR and incident cardiovascular disease via quasi-Bayesian Monte Carlo method with 2,000 simulations, using a Bonferroni correction for significance.

Findings

The mean (±SD) eGFR was 70 (±16) mL/min/1.73 m²; 62% of participants were women. Lower eGFR was associated with higher plasma ceramide-16:0 and sphingomyelin-16:0, and lower ceramides and sphingomyelins-20:0 and -22:0. Lower eGFR was associated with risk of incident heart failure and ischemic stroke, but not myocardial infarction. Five of eight sphingolipids partially mediated the association between eGFR and heart failure. The sphingolipids associated with the greatest proportion mediated were ceramide-16:0 (proportion mediated 13%, 95% CI 8–22%) and sphingomyelin-16:0 (proportion mediated 10%, 95% CI 5–17%). No sphingolipids mediated the association between eGFR and ischemic stroke.

Interpretation

Plasma sphingolipids partially mediated the association between lower eGFR and incident heart failure. Altered sphingolipids metabolism may be a novel mechanism for heart failure in patients with CKD.

Funding

This study was supported by T32 DK007467 and a KidneyCure Ben J. Lipps Research Fellowship (Dr. Lidgard). Sphingolipid measurements were supported by R01 HL128575 (Dr. Lemaitre) and R01 HL111375 (Dr. Hoofnagle) from the National Heart, Lung, and Blood Institute (NHLBI).

Research in context.

Evidence before this study

Between 4/1/2022 and 2/28/2023, PubMed and Google Scholar were searched using the following terms: “sphingolipids”, “ceramides”, “sphingomyelins”, “kidney disease”, “CKD”, “eGFR”, “cardiovascular disease”, “heart failure”, “stroke”, “myocardial infarction”, and “mediate” in any position. These searches disclosed multiple studies investigating associations between kidney disease and sphingolipids, sphingolipids and cardiovascular disease, and kidney disease and cardiovascular disease, but none evaluating sphingolipids as mediators of associations between kidney disease and cardiovascular disease.

Added value of this study

We described associations between eGFR and sphingolipid concentrations in the Cardiovascular Health Study cohort, and performed formal mediation analyses evaluating sphingolipids as mediators of the associations between kidney disease and cardiovascular disease, which may suggest sphingolipids lie on the causal pathway between kidney disease and cardiovascular disease.

Implications of all the available evidence

Our findings suggest that sphingolipids may partially mediate associations between lower eGFR and incident heart failure. In light of previous findings, our findings may suggest sphingolipids to be mechanistically involved in the development of heart failure in patients with CKD.

Introduction

Patients with chronic kidney disease (CKD) are at disproportionately high risk for cardiovascular disease (CVD) compared to the general population, and are more likely to suffer adverse outcomes from cardiovascular events.1, 2, 3 Though lipoprotein cholesterol levels have been strongly associated with CVD in the general population, they have been inconsistently associated with CVD in patients with CKD, and the roles of their constituent lipids are largely unknown.4, 5, 6

Recently, targeted lipidomics has emerged as a powerful tool to quantify changes in lipid metabolism beyond the traditional metrics available in standard lipid panels.⁷ These approaches have identified sphingolipids (including ceramides and sphingomyelins) as a class of lipid molecules with signaling and regulatory functions.⁸ Untargeted lipidomic approaches have been applied to patients with CKD to explore the lipidomic signature of decreased eGFR and alterations associated with progressive CKD.9, 10, 11 These studies have suggested that lower eGFR is associated with greater plasma levels of total ceramides.^12,13 However, these prior studies were limited by their use of relatively imprecise untargeted approaches that evaluated hundreds of lipids at a time, so the potential associations between CKD and plasma sphingolipid levels remain incompletely understood.

Lipidomic techniques have also been leveraged to investigate novel mechanisms of cardiovascular disease. Higher plasma levels of specific species of sphingolipids have been associated with increased risk of heart failure, myocardial infarction, mortality, and atrial fibrillation in several studies.14, 15, 16, 17, 18, 19, 20 Despite the reported associations between decreased eGFR and sphingolipids, and the known associations between decreased eGFR and cardiovascular endpoints including heart failure,²¹ myocardial infarction,^22,23 and stroke,22, 23, 24, 25 to our knowledge no study has evaluated whether sphingolipids mediate the associations between decreased eGFR and cardiovascular events. It is unknown whether dysregulated sphingolipid metabolism may lie on the causal pathway between CKD and CVD. To investigate this gap in knowledge, we measured plasma sphingolipids via a more precise targeted approach and formally assessed their roles as potential mediators of the associations between decreased eGFR and several distinct cardiovascular events: heart failure, myocardial infarction, and ischemic stroke.

Methods

Study population

We utilized data from the Cardiovascular Health Study cohort (CHS), a population-based cohort study of cardiovascular disease among older adults living in 4 U.S. communities (Forsyth County, NC; Sacramento County, CA; Washington County, MD; and Allegheny County, PA).²⁶ The cohort included 5,887 participants: 5,201 were recruited in 1989–1990, and 687 were recruited in 1992–1993.

To measure sphingolipids, we used stored blood samples from 4,026 participants with available specimens at the 1994–1995 clinic exam and from 586 additional participants at the 1992–1993 clinic exam because they lacked specimens at the 1994–1995 visit; sphingolipids were measured once in each participant. We excluded 935 participants with prior diagnoses of heart failure, myocardial infarction, and stroke (as prior cardiovascular disease has been associated with sphingolipids in previous studies),¹³ and 214 participants without creatinine and cystatin C levels, resulting in a final analytic cohort of 3,463 participants with and without CKD, and without known baseline cardiovascular disease (Fig. 1). Each participant was followed from their sphingolipid collection visit until disenrollment, death, cardiovascular event, or the end of administrative follow-up on 12/31/2014.

Fig. 1.

CONSORT diagram.

Ethics

All CHS activities are currently approved and participants appropriately consented under the University of Washington Institutional Review Board approval number STUDY00000109.²⁶

Sphingolipid measurement

Sphingolipids were measured on fasting and non-fasting EDTA plasma which had been continuously stored at −80 °C without thawing since collection. EDTA inhibits sphingomyelinases, and such samples have been used extensively for sphingolipid determination in previous studies.^15,19,27, 28, 29, 30, 31 Plasma lipids were extracted and sphingolipids quantified by liquid chromatography-tandem mass spectroscopy on a Sciex 6500 (Framingham, MA); detailed methodology and quality control have been previously reported.³⁰ Sphingolipid concentrations were determined using a single-point calibrator that was made from a pooled EDTA anticoagulated plasma sample; the calibrator was used throughout the study and 5 replicates were analyzed in each batch. Two quality control materials made from independent pools of plasma was included in each batch and run in duplicate; over 52 batches, coefficients of variation for each of the sphingolipid measurements were <20%. Plasma concentrations of sphingolipids were expressed as μM. We assumed the presence of the most common d (18:1) backbone for all sphingolipids other than ceramide-24:0; for instance, ceramide (d18:1/16:0), ceramide with an acylated fatty acid containing 16 carbons and 0 double bonds, is called ceramide-16:0 throughout. As ceramide-24:0 contained d18:0 and d18:1 backbones, we summed the two species and report ceramide-24:0. The ceramides measured were ceramide-16:0, -20:0, -22:0, and -24:0; the sphingomyelins measured were sphingomyelin-16:0, -20:0, -22:0, and -24:0, for a total of 8 exposure variables. Molecular structures are illustrated in Supplemental Fig. S1. These specific sphingolipids were measured given the associations of their acylated fatty acid chains with outcomes in previous work.^32,33

Measurement of eGFR

Assays for eGFR (creatinine and cystatin C) were performed on serum obtained from participants at the 1992–1993 study visit, stored at −70 °C. Serum creatinine was measured by a colorimetric method (Ektachem 700, Eastman Kodak); Cystatin C was measured via particle-enhanced immunonephelometric assay (N Latex Cystatin C, Dade Behring) with a nephelometer (BNII, Dade Behring). The coefficients of variation for each have been previously reported.²³ Estimated GFR was calculated from serum creatinine and cystatin C using the 2021 combined Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation.³⁴

Ascertainment of cardiovascular events

Study participants were followed by annual clinic examinations with interim phone contacts until 1999, telephone contacts twice a year thereafter, and an additional clinic visit in 2005–2006. Cardiovascular events included heart failure, myocardial infarction, and stroke; all were adjudicated by centralized CHS cardiac and stroke committees based on information from inpatient and outpatient medical records, diagnostic tests, consultations, and interviews. Events have been adjudicated through 12/31/2014. This adjudication process was rigorous and has been found to be superior to claims-based event determination strategies.^35,36 Confirmation of definitive heart failure required each of 3 criteria: 1. Diagnosis of heart failure by a treating physician; 2. Either heart failure symptoms (shortness of breath, fatigue, orthopnea, or paroxysmal nocturnal dyspnea) plus signs (edema, rales, tachycardia, gallop on cardiac exam, or displaced apical impulse) or supportive clinical findings on echocardiography, contrast ventriculography, or chest radiography; and 3. Medical therapy for heart failure, defined as diuretics plus either digitalis or a vasodilator (angiotensin-converting enzyme inhibitors, hydralazine, or long-acting nitrates).³⁷

Myocardial infarction was defined as a clinical event of myocardial ischemic with abnormal cardiac enzymes (creatinine kinase and lactate dehydrogenase) and/or electrocardiogram findings (evolving Q wave or ST-T abnormalities) using a previously published algorithm.³⁵ We considered fatal and non-fatal myocardial infarctions.

Stroke was defined as a clinical event of rapid onset of neurologic deficit not caused by trauma, tumor, or infection. Available documentation and imaging were reviewed to identify transient ischemic attacks and to classify strokes as ischemic, hemorrhagic, or uncertain.^38,39 We considered fatal and non-fatal ischemic and uncertain stroke types, but excluded hemorrhagic strokes.

For all outcomes, time to event was measured as the number of days from the date of sphingolipid collection to the event, death, or last date of follow-up. Participants could have an incident of each event during follow-up, and these participants were included in all qualifying analyses. For participants without an event, censoring times were calculated according to the last date of follow-up or the date of death.

Covariates

Information on covariates was obtained during each study visit; factors evaluated included medical history, lifestyle, and clinical risk factors.²⁶ Prevalent transient ischemic attack (TIA) was evaluated by questioning, with confirmation by review of medical records and imaging.^38,39 Information on age, sex, race and ethnicity, alcohol use (any versus none), smoking status, and fasting status for laboratory studies was by self-report. Weight, height, body mass index (BMI) and blood pressure were measured using standardized protocols.²⁶ Diabetes mellitus was defined by a fasting glucose level ≥126 mg/dL (7 mmol/L) or use of insulin or oral hypoglycemic agents. Total cholesterol, triglycerides, and high-density lipoprotein cholesterol (HDL-C) were measured on an Olympus Demand system (Lake Success, NY), standardized according to the Centers for Disease Control. Low density lipoprotein cholesterol (LDL-C) was calculated by the standard Friedewald equation.⁴⁰ Medications were ascertained by inventory of prescribed medications taken in the 2 weeks prior to each study visit. Most covariates were obtained from the same clinic exam from which the sphingolipids were measured, except for some laboratory measurements (LDL-C, HDL-C, cystatin C, and creatinine), which were measured at the 1992–1993 exam.

Statistical analysis

Sphingolipid species concentrations were log-transformed to reduce skewness. We tabulated baseline study characteristics overall and by quintiles of ceramide-16:0. Correlations were assessed using Pearson’s correlation coefficients among sphingolipid species, and between each sphingolipid and total HDL-C, total LDL-C, triglycerides, statin use, and fibrate use; heatmaps of each correlation coefficient were generated. We evaluated the functional forms of the association between sphingolipid species concentrations and eGFR by fitting generalized additive models based on a penalized regression spline approach with multiple smoothing parameter estimation by (Restricted) Marginal Likelihood, using the “mgcv” package in R.⁴¹ Based on these assessments, we evaluated cross-sectional associations between the exposure of eGFR and the outcome of log-transformed sphingolipids using linear regression analyses (Supplemental Fig. S2). To understand the effects of adjustment for various covariates, we fit a series of models. The first was unadjusted. Model 1 was adjusted for age, biologic sex, clinical location, fasting, smoking, BMI, alcohol use, LDL-C, HDL-C, triglycerides, SBP, diabetes, prior TIA, and use of fibrates, statins, and antihypertensive medications. Model 2, our “final model”, was adjusted for the components of Model 1 with additional adjustment for one other sphingolipid species as previously described17, 18, 19: analyses of ceramide-16:0 were additionally adjusted for ceramide-22:0 given their strong correlation and previous work showing opposed associations with cardiovascular outcomes to control for potential mutual confounding,17, 18, 19 and analyses of ceramide-20:0, -22:0, and -24:0 were adjusted for ceramide-16:0. An analogous “co-adjustment” strategy was used for sphingomyelins. Variance inflation factors were <2.5 for all sphingolipids in co-adjusted models, suggesting collinearity was not a concern.⁴² Models satisfied standard assumptions of linearity, independence, homoscedasticity, and normality.

To evaluate associations between eGFR (per 15 mL/min/1.73 m² lower eGFR) and each incident cardiovascular event, we fit parametric Weibull regressions, adjusting for the components of Model 1 above. Though associations between eGFR and cardiovascular events may not be completely linear when utilizing older eGFR formulae,^22,24 we modeled eGFR linearly for ease of interpretation and compatibility with our mediation analyses. We further adjusted each analysis for each sphingolipid to assess for possible attenuation of association as an estimate of mediation in the approach of Baron and Kenny.⁴³ If eGFR was not significantly associated with a particular cardiovascular event, we did not proceed with mediation analysis for that event, as there was no relationship to explore potential mediators for. Our primary, formal mediation analyses were conducted using a quasi-Bayesian Monte Carlo method based on the normal approximation with 2,000 simulations via the mediation package in R.^44,45 We calculated the proportion of each association mediated by each sphingolipid by dividing the indirect effect (the effect of the exposure on the outcome which is explained by the mediator) by the total effect of the exposure on the outcome as described by VanderWeele.⁴⁶ To correct for multiple comparisons in all analyses, we utilized a Bonferroni adjusted p-value for significance of 0.00625 (0.05/8 species) in all analyses.

Sensitivity analysis

To understand the mediation of relationships between eGFR and incident cardiovascular events by groups of sphingolipids, we utilized a joint mediation analysis in the mediation package. The groups (calculated by log-adjusting the sums of each constituent sphingolipid) analyzed were: sphingolipids with long fatty acids (ceramide-16:0 and sphingomyelin-16:0), sphingolipids with very long fatty acids (ceramide-20:0, -22:0, and -24:0, and sphingomyelin-20:0, -22:0, and -24:0), all ceramides, all sphingomyelins, and all sphingolipids.

Missingness was low overall (5.7% for diabetes status, 1.7% for LDL, 1.3% for smoking status, and <1% otherwise). All missing covariates were multiply imputed by chained equations via the mice package in R.⁴⁷ Twenty imputed datasets were generated and model fitting results were pooled using standard methods.⁴⁸ All analyses were performed using R 4.0.2 (R Foundation for Computing, Vienna, Austria).

Role of funders

Salary support for investigators and measurement of sphingolipids were paid for by grants from the National Institutes of Health and the American Society of Nephrology. Funders had no role in study design, data collection, data analysis, interpretation, or writing of this report.

Results

Characteristics of the study population

Among 3463 participants, the mean (±SD) age was 76 (±5) years. A total of 2,155 (62%) were women. The mean (±SD) eGFR was 70 (±16) mL/min/1.73 m², and the mean systolic BP was 134 (±20) mmHg. A total of 535 (15%) had diabetes. Compared to participants with lower plasma concentrations of ceramide-16:0, participants with higher plasma concentrations were more often women, had lower mean eGFR, higher LDL-C and total triglycerides, and were less often alcohol users (Table 1). We additionally tabulated baseline characteristics by categories of eGFR (Supplemental Table S1).

Table 1.

Baseline characteristics overall and by quintiles of ceramide-16:0.

	Overall	Quintile 1 (≤0.216 μM)	Quintile 2 (>0.216 μM, ≤0.244 μM)	Quintile 3 (>0.244 μM, ≤0.270 μM)	Quintile 4 (>0.270 μM, ≤0.308 μM)	Quintile 5 (>308 μM)
N	3463	693	692	693	692	693
Age, years, mean (SD)	76 (5)	76 (5)	76 (5)	76 (5)	76 (5)	77 (5)
Women, N (%)	2155 (62)	398 (57)	437 (63)	421 (61)	444 (64)	455 (66)
Race and ethnicity
Non-hispanic white	2881 (83)	556 (80)	561 (81)	593 (86)	580 (84)	591 (85)
Non-hispanic black	525 (15)	124 (18)	117 (17)	92 (13)	98 (14)	94 (14)
Hispanic	2 (0)	10 (1)	10 (1)	4 (1)	10 (1)	7 (1)
American Indian/Native Alaskan	2 (0)	0 (0)	1 (0)	0 (0)	1 (0)	0 (0)
Asian or Pacific Islander	41 (1)	0 (0)	0 (0)	2 (0)	0 (0)	0 (0)
Other	12 (0)	3 (0)	3 (0)	2 (0)	3 (0)	1 (0)
eGFR, mL/min/1.73 m2, mean (SD)	70 (16)	73 (15)	71 (15)	71 (15)	69 (16)	67 (18)
Diabetes, N (%)	535 (15)	98 (14)	102 (15)	82 (12)	115 (17)	138 (20)
Prior transient ischemic attack, N (%)	77 (2)	16 (2)	11 (2)	17 (2)	14 (2)	19 (3)
Systolic BP, mmHg, mean (SD)	134 (20)	135 (20)	133 (19)	133 (20)	135 (20)	136 (21)
BMI, mean (SD)	27 (5)	27 (4)	27 (5)	27 (5)	27 (5)	27 (5)
Current or prior tobacco use, N (%)	1795 (52)	362 (52)	340 (49)	355 (51)	349 (50)	389 (56)
Alcohol use, N (%)	1576 (46)	364 (53)	324 (47)	337 (49)	300 (43)	251 (36)
LDL-C, mg/dL, mean (SD)	121 (34)	108 (32)	116 (30)	121 (30)	126 (32)	133 (38)
HDL-C, mg/dL, mean (SD)	54 (14)	56 (15)	55 (14)	55 (14)	53 (14)	52 (14)
Total triglycerides, mg/dL, mean (SD)	142 (81)	117 (67)	128 (67)	138 (74)	154 (90)	171 (94)
Antihypertensive(s), N (%)	1685 (49)	349 (50)	335 (48)	323 (47)	337 (49)	341 (49)
Statins, N (%)	202 (6)	44 (6)	39 (6)	46 (7)	36 (5)	37 (5)
Fibrates, N (%)	43 (1)	29 (4)	4 (1)	5 (1)	2 (0)	3 (0)

BP, blood pressure; BMI, body mass index; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein.

Correlation between sphingolipid species, lipoproteins, and medication use

All sphingolipids were positively correlated with each other (Supplemental Fig. S3). The correlation was especially strong between sphingomyelin-20:0, -22:0, and -24:0 (coefficients ranging from 0.68 to 0.90). Ceramides were negatively correlated with HDL-C, and sphingomyelins were positively correlated with HDL-C. All sphingolipids were positively correlated with LDL-C and weakly correlated with medication use (coefficients ranging from −0.12 to 0.05).

Association of eGFR with sphingolipids

In unadjusted models, eGFR was significantly associated with ceramide-16:0 and sphingomyelin-20:0, -22:0, and -24:0. In Model 1, ceramide-16:0 and sphingomyelin-16:0 were significantly inversely associated with eGFR. In Model 2, our final model, each 15-point lower eGFR was associated with a 2.2% greater geometric mean of ceramide-16:0 (95% CI 1.6%, 2.8%), and 1.0% greater geometric mean of sphingomyelin-16:0 (95% CI 0.6%, 1.4%). In contrast, lower eGFR was associated with lower plasma concentrations of ceramide-20:0, -22:0, and sphingomyelin-20:0, -22:0, and -24:0; all these associations were significant at the Bonferroni-adjusted p-value < 0.00625 (Table 2).

Table 2.

Percent difference in geometric means of plasma sphingolipids per 15-point lower eGFR.

	Unadjusted		Model 1		Model 2
	% difference in geometric mean (95% CI)	p-value	% difference in geometric mean (95% CI)	p-value	% difference in geometric mean (95% CI)	p-value
Ceramide-16:0	2.8 (2.0, 3.5)	<0.001	2.3 (1.6, 3.1)	<0.001	2.2 (1.6, 2.8)	<0.001
Ceramide-20:0	0.8 (−0.2, 1.9)	0.13	−0.8 (−1.9, 0.2)	0.13	−2.5 (−3.4, −1.5)	<0.001
Ceramide-22:0	1.3 (0.3, 2.3)	0.008	0.4 (−0.5, 1.3)	0.44	−1.2 (−1.9, −0.4)	0.002
Ceramide-24:0	0.1 (−0.8, 1.1)	0.77	0.4 (−0.6, 1.3)	0.44	−0.7 (−1.5, 0.1)	0.09
Sphingomyelin-16:0	0.4 (−0.1, 0.9)	0.14	0.7 (0.2, 1.2)	0.003	1.0 (0.6, 1.4)	<0.001
Sphingomyelin-20:0	−1.5 (−2.1, −0.8)	<0.001	−0.5 (−1.1, 0.1)	0.09	−0.9 (−1.5, −0.3)	0.002
Sphingomyelin-22:0	−2.8 (−2.0,−3.5)	<0.001	−2.3 (−1.6,−3.1)	<0.001	−2.2 (−1.6,−2.8)	<0.001
Sphingomyelin-24:0	0.8 (−0.2, 1.9)	0.13	−0.8 (−1.9, 0.2)	0.13	−2.5 (−3.4, −1.5)	<0.001

Bold font indicates significance at the Bonferroni-adjusted p-value < 0.00625.

Model 1: adjusted for age, sex, clinical location, smoking, BMI, alcohol use, LDL-C, HDL-C, triglycerides, SBP, diabetes, prior TIA, and use of fibrates, statins, and antihypertensive medications.

Model 2: Adjusted for components of Model 1; additionally, ceramide-16:0 was adjusted for ceramide-22:0, sphingomyelin-16:0 was adjusted for sphingomyelin-22:0, ceramides-20:0–24:0 were adjusted for ceramide-16:0, and sphingomyelins-20:0–24:0 were adjusted for sphingomyelin-16:0.

Association of eGFR with cardiovascular events

In total, there were 1,130 heart failure events, 637 myocardial infarctions, and 537 strokes (including 481 ischemic strokes and 56 of unknown type) during follow-up. Median (IQR) time to event was 9.8 (5.1, 15.7) years for heart failure, 10.2 (5.7, 16.1) years for myocardial infarction, and 10.7 (5.8, 16.4) years for stroke. After adjustment for the components of Model 1, lower eGFR was significantly associated with greater risk of incident heart failure and ischemic stroke. However, eGFR was not significantly associated with risk of myocardial infarction. Adjusting for sphingolipids slightly attenuated the associations between eGFR and heart failure, but not the associations between eGFR and stroke (Supplemental Table S2).

Mediation analyses

Formal mediation analyses demonstrated that ceramide-16:0 and -22:0, and sphingomyelin-16:0, -20:0, and -22:0 each partially mediated the relationship between eGFR and incident heart failure when adjusting for the components of Model 2, with p-values for tests of mediation less than the pre-specified Bonferroni-adjusted p-value of 0.00625 (Table 3). For instance, ceramide-16:0 mediated 13% (95% CI 8%, 22%) of the association (Fig. 2), and sphingomyelin-16:0 mediated 10% (95% CI 5%, 17%) of the association. None of the sphingolipids in question significantly mediated the association between eGFR and stroke (all p-values > 0.00625) (Table 3).

Table 3.

Mediation by individual sphingolipids of the associations between eGFR and cardiovascular outcomes using a quasi-Bayesian Monte Carlo method with 2,000 pulls.

Sphingolipid	Heart failure		Ischemic stroke
	Proportional effect (95% CI)	p-value	Proportional effect (95% CI)	p-value
Ceramide-16:0	0.13 (0.08, 0.22)	<0.001	0.06 (−0.14, 0.51)	0.32
Ceramide-20:0	0.04 (0.01, 0.09)	0.02	0.02 (−0.19, 0.27)	0.66
Ceramide-22:0	0.05 (0.02, 0.10)	<0.001	0.02 (−0.10, 0.21)	0.52
Ceramide-24:0	0.02 (0.00, 0.05)	0.06	0.01 (−0.06, 0.14)	0.47
Sphingomyelin-16:0	0.10 (0.05, 0.17)	<0.001	0.00 (−0.21, 0.21)	0.96
Sphingomyelin-20:0	0.04 (0.01, 0.09)	<0.001	0.01 (−0.11, 0.15)	0.79
Sphingomyelin-22:0	0.07 (0.03, 0.13)	<0.001	0.03 (−0.13, 0.27)	0.53
Sphingomyelin-24:0	0.03 (0.01, 0.07)	0.01	0.01 (−0.07, 0.15)	0.54

Bold font indicates significance at the Bonferroni-adjusted p-value < 0.00625.

Proportional effect is the percentage of the association between eGFR and the cardiovascular outcome that is mediated by the sphingolipid of interest.

Adjusted for age, sex, clinical location, smoking, BMI, alcohol use, LDL-C, HDL-C, triglycerides, SBP, diabetes, prior TIA, and use of fibrates, statins, and antihypertensive medications. Additionally, ceramide-16:0 was adjusted for ceramide-22:0, sphingomyelin-16:0 was adjusted for sphingomyelin-22:0, ceramides-20:0–24:0 were adjusted for ceramide-16:0, and sphingomyelins-20:0–24:0 were adjusted for sphingomyelin-16:0.

Fig. 2.

Partial mediation of the association between eGFR and heart failure by ceramide-16:0. X axis are unadjusted Weibull regression coefficients; more negative numbers imply shorter event-free survival. ACME: Average causal mediation effect (portion of total effect mediated by ceramide-16), ADE: Average direct effect of eGFR on heart failure risk (accounting for mediation). Total Effect: association between eGFR and heart failure risk before accounting for ceramide-16.

Sensitivity analysis

Sphingolipids with 16-carbon fatty acids and sphingolipids with very long fatty acids each partially mediated the associations between eGFR and incident heart failure (proportion mediated (95% CI) 10% (5%, 17%) and 7% (3%, 15%)) (Table 4). No group of sphingolipids mediated the associations between eGFR and stroke.

Table 4.

Joint mediation by classes of sphingolipids of the associations between eGFR and cardiovascular outcomes.

Sphingolipid subclass	Heart failure		Ischemic stroke
	Proportional effect (95% CI)	p-value	Proportional effect (95% CI)	p-value
Sphingolipids with 16-carbon fatty acidsa	0.10 (0.05, 0.17)	<0.001	0.00 (−0.22, 0.22)	0.95
Sphingolipids with very long fatty acidsb	0.07 (0.03, 0.15)	<0.001	0.02 (−0.15, 0.28)	0.57
All ceramides	0.00 (−0.01, 0.01)	0.73	0.00 (−0.06, 0.04)	0.90
All sphingomyelins	0.01 (−0.01, 0.03)	0.16	0.00 (−0.08, 0.04)	0.71
All sphingolipids	0.01 (−0.01, 0.03)	0.15	0.00 (−0.07, 0.04)	0.73

Bold font indicates significance at the Bonferroni-adjusted p-value < 0.01.

Proportional effect is the percentage of the association between eGFR and the cardiovascular outcome that is mediated by the sphingolipid of interest.

Adjusted for age, sex, clinical location, smoking, BMI, alcohol use, LDL-C, HDL-C, triglycerides, SBP, diabetes, prior TIA, and use of fibrates, statins, and antihypertensive medications. Additionally, sphingolipids with 16-carbon fatty acids were adjusted for ceramide-22:0 and sphingomyelin 22:0, and sphingolipids with very-long fatty acids were adjusted for ceramide-16:0 and sphingomyelin 16:0.

^aSphingolipids with 16-carbon fatty acids: sum of ceramide-16:0 and sphingomyelin-16:0.

^bSphingolipids with very-long fatty acids: sum of ceramide-20:0, ceramide-22:0, ceramide-24:0, sphingomyelin-20:0, sphingomyelin-22:0, and sphingomyelin-24:0.

Discussion

In a large cohort of older adults, plasma sphingolipids partially mediated the associations between eGFR and incident heart failure. This observation suggests that alterations in sphingolipid metabolism may lie on the causal pathway between CKD and incident heart failure, potentially identifying a novel contributor to the increased heart failure risk in this patient population. In contrast, sphingolipids did not significantly mediate the association between eGFR and ischemic stroke. Further studies are needed to confirm the directionality and magnitude of these associations in more diverse, cohorts.

We found that sphingolipids mediated the associations between eGFR and heart failure, but not the association between eGFR and ischemic stroke. In our analyses, eGFR was not significantly associated with myocardial infarction, and we were therefore unable to assess sphingolipids as mediators of this nonsignificant association. These associations may point to underlying pathophysiologic differences. For instance, ischemic stroke and myocardial infarction are thought to be macrovascular diseases, while heart failure (especially with preserved ejection fraction) may represent a microvascular disease.49, 50, 51, 52, 53 While sphingolipids have been related to macrovascular disease risk in non-CKD populations,^14,20 it is possible that in CKD they specifically enhance microvascular disease pathways, leading to increased heart failure risk.

Sphingolipids are a large class of biologically active lipid molecules with structural, signaling, and regulatory properties.⁸ Central to sphingolipid metabolism are the ceramides, which can be converted to sphingomyelins via sphingomyelin synthase⁵⁴; sphingomyelins may be converted back to ceramides via sphingomyelinase, with conservation of the acylated fatty acid chain.⁵⁵ Sphingolipids include an acylated fatty acid chain of varying length and unsaturation; the length of these fatty acids has been associated with differential outcomes in several studies. For instance, ceramide-16:0 and sphingomyelin-16:0 have been associated with greater risk for incident heart failure, atrial fibrillation, and mortality, whereas sphingolipids with very long saturated fatty acids have been associated with lesser risk of these outcomes.17, 18, 19 In particular, ceramide-16:0 has been consistently associated with greater risk for major adverse cardiac events and mortality in meta-analysis.²⁰ While the mechanisms underlying these differential risks are not well understood, it is hypothesized that differential enrichment of cell membranes in sphingolipids with long versus very long fatty acids leads to greater activation of inflammatory and pro-apoptotic pathways.56, 57, 58, 59 Similarly, sphingolipids have been shown to promote apoptotic pathways via mitochondrial instability, activation of cRAF removing AKT-PKB inhibition of Caspase-9 and the Caspase 3 apoptosis pathway, and activation of the PP2A/Bcl-2/BAD pro-apoptotic pathway.^{14,15,28,60,61} Sphingolipids may also induce senescence pathways via dephosphorylation of retinoblastoma protein and activation of serine/threonine protein phosphatases.62, 63, 64 Future mechanistic work to more fully describe these biologic pathways, and demonstrate causality, may be needed.

The associations between CKD and sphingolipids are incompletely understood. Several studies have noted associations between lower eGFR and greater proteinuria and greater plasma concentrations of various sphingolipids.^31,65 For example, Mantovani et al. recently reported higher plasma levels of ceramide-16:0, -18:0, -20:0, and -24:1 in patients with eGFR < 60 mL/min/1.73 m² versus those with greater eGFR; however, these authors did not consider the correlation between ceramides with long fatty acids and ceramides with very long fatty acids.¹³ Afshinnia et al. noted greater concentrations of sphingomyelins with long fatty acids (≥16 carbons) in patients with stage 5 versus less severe CKD, though they utilized an untargeted approach.¹⁰ More recently, our group found that greater albuminuria was significantly associated with greater relative HDL composition of ceramide-22:0, -24:1, and sphingomyelin-16:0 in a Seattle-based cohort of adults with CKD (we did not measure ceramide-16:0).⁶⁶ In our present analyses, we noted significant (but numerically small) inverse associations between eGFR and plasma concentrations of ceramide-16:0 and sphingomyelin-16:0, and positive associations between eGFR and concentrations of ceramide-20:0 and -22:0 and sphingomyelin-20:0, -22:0, and -24:0. This inconsistency may be due to our previous evaluation of HDL alone, whereas our present analyses consider whole plasma.

The pathophysiology of cardiovascular disease in patients with CKD is complex and multifactorial, and has been related to pathways of inflammation, volume overload and hypertension, vascular calcification, fibrosis, and endothelial dysfunction.67, 68, 69, 70, 71, 72, 73, 74 Further, the mechanisms by which sphingolipids may mediate the relationship between CKD and cardiovascular disease are not completely understood. However, recent studies in humans and animal models of CKD have demonstrated that ceramide synthetases (CerS) likely play an important role.⁵⁷ CerS2 is involved in the biosynthesis of ceramides with very long fatty acids. CerS2 is highly expressed in the kidney, and activity is thought to decrease with worsening CKD. Conversely, CerS5 and CerS6 are involved in the biosynthesis of ceramide-16:0. Greater activity of CerS5 and CerS6 have been noted in patients with fibrotic kidneys. These ceramides may then be converted to sphingomyelins via sphingomyelinase, with conservation of the acylated fatty acid. These data would suggest that worsening CKD may cause decreased activity of CerS2, and increased activity of CerS5 and CerS6, leading to lower levels of sphingolipids containing very long acylated fatty acids, and higher levels of ceramide-16:0 and sphingomyelin-16:0. Though our data on the associations of eGFR and sphingolipids are cross-sectional in nature, they generally support this hypothesis; we noted associations between lower eGFR and higher ceramide and sphingomyelin-16:0, and lower ceramide and sphingomyelin-20:0 and -22:0. As higher ceramide and sphingomyelin-16:0 and lower ceramide and sphingomyelin-20:0 and -22:0 are associated with increased risk of heart failure,¹⁹ we speculate that these sphingolipids partially mediate the association between CKD and heart failure, though we cannot establish the directionality of the proposed mediation with our present analyses.

These findings may have important implications for the care of patients with CKD regarding prevention of cardiovascular disease should they hold true in other cohorts. If so, therapies decreasing ceramide and sphingomyelin with long fatty acids and increasing ceramides and sphingomyelins with very long fatty acids may be potential therapeutic options to prevent the development of incident cardiovascular disease in CKD. For instance, several authors have proposed ceramide synthase inhibitors as potentially meritorious options.^57,75 However, our use of the CHS cohort may limit generalizability to other populations, specifically those with more advanced CKD (given the limited number of patients with eGFR < 45 mL/min/1.73 m²), and those at higher risk for cardiovascular disease than our population of generally healthy older adults.

This study has several notable strengths, including the use of a large, well-characterized cohort; the targeted, hypothesis-driven evaluation of specific sphingolipids; and multiple approaches to our primary analyses to ensure the robustness of findings. However, our approach does have several limitations. First, not all data were collected at the same time point, and individuals’ measurements were un-replicated at later time points. Second, given the cross-sectional nature of our analyses of associations between eGFR and sphingolipids, we are unable to state directionality of associations; sphingolipids may influence eGFR, rather than eGFR influencing sphingolipids. However, this alternative seems unlikely, as most of our sphingolipids were measured after eGFR. The observational nature of the study limits causal inference, and residual confounding remains possible, although rich data in CHS allow for adjustment of many potential confounders. Age, alcohol use, and smoking status were by self-report, which may have resulted in under-ascertainment and incomplete adjustment. We modeled eGFR continuously for sake of model fitting; however, this may oversimplify associations between eGFR and cardiovascular events, especially given previous work suggesting greater risk for events at the lowest categories of eGFR. Ascertainment of myocardial infarction events was limited by reliance on lactate dehydrogenase and creatinine kinase rather than modern, myocardium-specific markers. Our study was not designed to investigate factors that might alter sphingolipid metabolism as potential therapeutic targets. Finally, the cohort was comprised of older, black and white research study participants who were relatively healthy (i.e., with relatively high eGFR), which may limit generalizability to younger populations, patients of different ethnicities, or those with more severe CKD.

In conclusion, we observed that sphingolipids may partially explain the associations between eGFR and incident heart failure. These results may implicate certain sphingolipids as partial mediators of incident heart failure in CKD. Further studies are needed to replicate these findings in younger populations, confirm the directionality of associations from decreased eGFR to altered sphingolipids, and investigate the potential roles of sphingolipids as contributors to cardiovascular disease risk in CKD.

Contributors

BL was responsible for conceptualization, project administration, formal analysis, investigation, methodology, validation, visualization, and writing—original draft. NB was responsible for conceptualization, funding acquisition, and resources. LZ was responsible for data curation, methodology, formal analysis, software, validation, and visualization. AH was responsible for conceptualization, funding acquisition, and resources. AF was responsible for conceptualization and methodology. RL was responsible for conceptualization, funding acquisition, methodology, project administration, resources, and supervision. BL and LZ verified the underlying data. All authors (BL, NB, LZ, AH, AF, WTL, MS, DS, JU, RL) contributed equally to writing—review and editing. All authors read and approved the final version of the manuscript.

Data sharing statement

The data and study materials will not be made available to other researchers for the purpose of reproducing the results or replicating the procedure. The authors are not authorized to share CHS data. Information on CHS data requests can be found at https://chs-nhlbi.org.

Declaration of interests

Dr. Zelnick reports a consultancy agreement with Veterans Medical Research Foundation, and serves as a Statistical Editor for the Clinical Journal of the American Society of Nephrology. Dr. Hoofnagle reports providing expert testimony for Kilpatrick Townsend and Stockton LLP; receipt of equipment and instrument support from Waters, Inc.; and serves as an Associate Editor for Clinical Chemistry.