High-Density Lipoprotein Lipidomics in Chronic Kidney Disease

Abstract

Background

Patients with chronic kidney disease (CKD) have dysfunctional high-density lipoprotein particles (HDL) as compared with the general population. Understanding the lipid composition of HDL may provide mechanistic insight. We tested associations of estimated glomerular filtration rate (eGFR) and albuminuria with relative HDL abundance of ceramides, sphingomyelins, and phosphatidylcholines in participants with CKD.

Methods

We studied 490 participants with CKD from the Seattle Kidney Study. HDL was isolated from plasma; targeted lipidomics was used to quantify the relative abundance of ceramides, sphingomyelins, and phosphatidylcholines per 10 μg of total HDL protein. We evaluated the associations of eGFR and albuminuria with levels of individual lipids and lipid classes (including 7 ceramides, 6 sphingomyelins, and 24 phosphatidylcholines) using multivariable linear regression, controlling for multiple comparisons via the false discovery rate.

Results

The mean eGFR was 45 (±24) mL/min/1.73 m²; the median (IQR) albuminuria was 108 (16, 686) mg/g [12.2 (1.8, 77.6) mg/mmol] urine creatinine. After adjusting for demographics, past medical history, laboratory values, and medication use, eGFR was not associated with higher relative abundance of any class of lipids or individual lipids. Greater albuminuria was significantly associated with a higher relative abundance of total ceramides and moderate-long R-chain sphingomyelins, ceramides 22:0 and 24:1, hexosylceramide 16:0, sphingomyelin 16:0, and phosphatidylcholines 29:0, 30:1, and 38:2; the strongest association was for hexosylceramide 16:0 (increase per doubling of urine albumin-creatinine ratio 0.022, 95% CI 0.012, 0.032).

Conclusions

Greater albuminuria was significantly associated with specific alterations in the lipid composition of HDL in participants with CKD.

BRIEF SUMMARY

Patients with chronic kidney disease (CKD) have dysfunctional high-density lipoprotein (HDL) compared with the general population. Understanding the lipid composition of HDL may provide mechanistic insight. We isolated HDL from 490 participants with CKD, and quantified the relative abundance of ceramides, sphingomyelins, and phosphatidylcholines using liquid chromatography-tandem mass spectrometry (LC-MS/MS). EGFR was not associated with abundance of lipid classes or individual lipids. Albuminuria was positively associated with relative abundance of ceramides and moderate-long R-chain sphingomyelins, ceramides 22:0 and 24:1, hexosylceramide 16:0, sphingomyelin 16:0, and phosphatidylcholines 29:0, 30:1, and 38:2. Albuminuria was associated with alterations in the lipid content of HDL in CKD.

INTRODUCTION

High-density lipoproteins (HDL) are a complex family of lipoprotein particles composed of lipid and protein components.(1, 2) Patients with chronic kidney disease (CKD) typically have lower circulating concentrations of HDL-cholesterol (HDL-C) compared with patients who have normal kidney function.(3) Furthermore, HDL particles in CKD tend to be dysfunctional, lacking the typical anti-inflammatory and cholesterol efflux capabilities of HDL found in healthy individuals.(4, 5) However, the molecular derangements underlying these functional deficiencies have not been completely characterized. To date, studies have characterized the protein components of HDL particles in CKD, rather than investigating specific constituent lipids or classes of lipids.(6) Several lipid classes are present on the surface of HDL (e.g., sphingolipids and phosphatidylcholines) and serve important signaling and regulatory roles; it is possible that lipid alterations in CKD are at least partially culpable for the functional alterations observed in the HDL of patients with CKD.(7, 8)

Whole-plasma lipidomics has emerged as a powerful tool to characterize lipid alterations beyond traditionally investigated lipoprotein size, apolipoprotein composition, and total serum lipid levels. These approaches have suggested that β-oxidation is impaired in patients with CKD, and that lower eGFR is associated with greater polyunsaturated long-chain complex lipids.(9) However, specific lipidomic alterations in HDL particles have not been characterized in CKD and may provide further mechanistic insight into the excess cardiovascular risk observed in this setting. To investigate this gap in knowledge, we utilized targeted lipidomics to characterize the surface lipid composition of HDL particles in a cohort study of CKD. We tested associations between kidney function, defined by eGFR and albuminuria, and the relative concentrations of defining lipid molecules within HDL: ceramides, sphingomyelins, and phosphatidylcholines.

METHODS

Study Population

We utilized data from the Seattle Kidney Study (SKS), which began in 2004 and recruited 693 individuals with CKD from outpatient nephrology clinics in Seattle, Washington. Samples were collected from 01/23/2006 to 12/4/2012, and frozen at −80°C. Inclusion criteria were age ≥18 years and either eGFR ≤90 mL/min/1.73 m² or a urine protein-creatinine ratio ≥30 mg/g (≥3.4 mg/mmol creatinine). Exclusion criteria included receipt of maintenance dialysis at the baseline visit, expectation of dialysis initiation within the coming 3 months, prior kidney transplantation, and inability to provide informed consent. All participants provided written informed consent; the study protocol was approved by the Institutional Review Board at the University of Washington. For this study, we measured HDL composition in 490 SKS participants with available samples and covariate data for analysis.

HDL Isolation and Lipid Measurement

We evaluated both individual lipids and classes of lipids (ceramides, sphingomyelins, and phosphatidylcholines). For each ceramide and sphingomyelin, we assumed the presence of the d(18:1) backbone (the proportion of these lipids with non-d18:1 backbones is unclear, though d18:1 is by far the most common in humans);(10) our nomenclature reflects this by omitting the backbone chain (e.g. sphingomyelin d(18:1)16:0, sphingomyelin with an acylated fatty acid containing 16 carbons and 0 double bonds, is called sphingomyelin 16:0 throughout). Each class of lipids was calculated as the sum of the relative amounts of constituent lipid species for each participant. The ceramides included ceramide 22:0, ceramide 24:0, ceramide 24:1, hexosylceramide 16:0, hexosylceramide 22:0, hexosylceramide 24:0, and lactosylceramides 16:0. The sphingomyelins included moderate-long R-chain sphingomyelins 14:0, 16:0, and 18:0, and very-long R-chain sphingomyelins 20:0, 22:0, and 24:0. The phosphatidylcholines measured were phosphatidylcholine 28:0, 28:1, 29:0, 30:1, 30:2, 32:1, 32:2, 33:3, 34:1, 34:2, 35:4, 35:5, 36:1, 36:2, 36:3, 36:4, 36:5, 38:1, 38:2, 38:3, 38:4, 38:5, 40:5, and 40:6. We further evaluated phosphatidylcholines by degree of unsaturation (polyunsaturated were considered as having 2 or more double bonds; monounsaturated and saturated phosphatidylcholines contained 1 and 0 double bonds, respectively). The total lipid content was calculated by adding the relative abundance of each lipid for each participant.

Complete details of the purification of HDL particles in this cohort have been previously published.(6) Briefly, the HDL fraction of plasma (density 1.063 to 1.210 g/mL) was purified by a 2-step density gradient ultracentrifugation process utilizing potassium bromide between May and November 2013. First, all lipoproteins were floated from serum using a 1.210 g/mL potassium bromide solution and were transferred to a new tube. Second, all lipoproteins less dense than HDL were floated using 1.063 g/mL potassium bromide; the lipoproteins at the bottom of each sample after this second step were subsequently dialyzed and frozen at −80°C before use. HDL lipids were isolated using an organic protein precipitation method with analog internal standards, as previously described.(11)

Each sample was analyzed using liquid chromatography tandem-mass spectrometry (LC-MS/MS) on a Sciex 6500 (Framingham, MA) in May 2015. Peak areas for each endogenous lipid were normalized to the peak areas from isotope-labeled internal standards that were added prior to lipid extraction/protein precipitation with organic solvent. The internal standards used to normalize each endogenous lipid are listed in Supplemental Table 1. The calculated ratio of endogenous to internal standard peak area for each lipid in each sample (peak area ratio) was then standardized to peak area ratios from calibrator samples included in each extraction batch to reduce inter-batch variability. This standardized peak area ratio is therefore a relative abundance of each lipid in the sample. Because an aliquot containing 10 μg of HDL protein was extracted for each sample, the standardized peak area ratio is the relative abundance of each lipid per 10 μg of total HDL protein.

Measurement of kidney function

Serum concentrations of urine creatinine were measured on a Beckman-Coulter DxC autoanalyzer and are traceable to isotope dilution mass-spectrometry standards. Serum cystatin C concentrations were calibrated by reconstitution of cystatin C reference material (ERM-DAY7/IFCC) per its Certificate of Analysis. Spot urine albumin concentrations were measured by immunoturbidimetry and were indexed to urine creatinine. Estimated GFR was calculated from serum creatinine and cystatin C using the 2021 Chronic Kidney Disease Epidemiology Collaboration equation (CKD-EPI).(12)

Covariates

At the initial study visit, participants provided information on sociodemographic characteristics, medical history, medication usage, and lifestyle behaviors. Race and ethnicity were self-reported, and were categorized as non-Hispanic White, non-Hispanic Black, Asian or Pacific Islander, American Indian or Native Alaskan, and Other (including “African”, “Armenian”, “Egyptian”, several European ethnicities, various combinations of the above, and several non-responders). Total cholesterol, HDL-C, and triglycerides were measured on a standard Beckman-Coulter DxC autoanalyzer (Indianapolis, IN). Low density lipoprotein cholesterol (LDL-C) was calculated via the Friedewald calculation.

Statistical Analyses

Baseline characteristics of the analytic cohort were tabulated overall and by tertiles of ceramide 24:1, given prior literature suggesting that ceramides and long R-chain polyunsaturated lipids may be associated with eGFR.(9, 13) Correlation between each lipid species, and between each lipid and total HDL-C, total LDL-C, total cholesterol, statin use, and use of other lipid-lowering medications were assessed using Pearson correlation coefficients; heatmaps of each correlation coefficient were generated. The sums of relative abundance of lipids in various functional classes (ceramides, sphingomyelins, moderate-long R-chain [14–18 carbons] and very long R-chain [20–24 carbons] sphingomyelins, phosphatidylcholines, saturated and monounsaturated phosphatidylcholines, polyunsaturated phosphatidylcholines, and total lipids) were plotted versus continuous eGFR. We evaluated the functional form of the association between relative lipid class abundance 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.(14)

We modeled the relative lipid abundances linearly (rather than log-transformed) based on the functional forms of their associations with eGFR and albuminuria assessed using generalized additive models as above.(14) These analyses demonstrated relatively linear relationships (Supplemental Figure 1), supporting modeling lipids continuously rather than log-adjusted.

To evaluate associations between eGFR, albuminuria, and relative HDL lipid concentrations, linear regression analyses were performed. We modeled eGFR continuously, and albuminuria per doubling in the urine albumin-creatinine ratio. For each model, the outcome was modeled as the sums of abundance of lipids in each functional class as defined above, and as relative abundance of each lipid alone. We adjusted for age, biologic sex, BMI, diabetes, hypertension, prior stroke or myocardial infarction, HDL-C, LDL-C, and use of statins and other lipid-lowering medications. We controlled for multiple comparisons by utilizing a false discovery rate (FDR) of 5%.(15)

In a sensitivity analysis, to explore if statin use was related to lipid levels, we fitted linear models with statin use as the exposure, and concentrations of various classes of lipids or individual lipids as the outcome; similar to our primary analyses, these models were adjusted for age, biologic sex, BMI, diabetes, hypertension, prior stroke or myocardial infarction, eGFR, albuminuria, HDL-C, LDL-C, and use of other lipid-lowering medications. For this analysis, we evaluated a subset of the population that had non-missing data for statin use at baseline (N = 486).

To investigate associations between the length and number of double bonds present in phosphatidylcholines within each stratum of eGFR (≥60, 45 - <60, 30 - <45, and <30 mL/min/1.73 m²) and albuminuria (<30, 30 - <300, and ≥300 mg/g of urine creatinine; equivalent to <3.4, 3.4 - <33.9, ≥33.9 mg/mmol creatinine) we linearly regressed the mean relative HDL concentration of each species at each strata on the number of carbons, the number of double bonds, and the interaction of the two. We created heatmaps plotting the average relative HDL concentrations in each combination of number of carbons and double bonds, within each stratum of eGFR and albuminuria.

Missingness was low overall (5.5% for LDL-C levels, and ≤1% for all other covariates); all missing covariates were multiply imputed using chained equations via the mice package in R.¹⁶ The multiple analyses over imputations were combined using standard Rubin’s rules to account for the variability in the imputation procedure.(17)

All analyses were performed using R 4.0.2 (R Foundation for Computing, Vienna, Austria).

RESULTS

Characteristics of the Study Population

Among 490 participants, the mean (±SD) age was 58 (±14) years. A total of 161 (33%) were women. The mean (±SD) eGFR was 45 (±24) mL/min/1.73 m², and the median (IQR) urine albumin-creatinine ratio was 108 (16–686) mg/g [12.2 (1.8, 77.6) mg/mmol]. In total 50% of the population had diabetes, and 54% were on statins. The mean (±SD) LDL-C and HDL-C levels were 102 (±44) and 42 (±17) mg/dL [2.6 (±1.1) and 1.1 (±0.4) mmol/L], respectively. Compared to participants with lower relative HDL concentrations of ceramide 24:1, those in tertile 3 versus tertile 1 were more often women with lower eGFR, greater albuminuria, greater HDL-C, and lower BMI (Table 1).

Table 1:

Baseline characteristics of the analytic cohort overall and by tertiles of Ceramide 24:1

Variable	Overall	Tertile 1 (Cer 24:1 ≤0.530)	Tertile 2 (Cer 24:1 >0.530, ≤0.786)	Tertile 3 (Cer 24:1 > 0.786)	p-value
N	490	164	163	163
Age in years, mean (SD)	58 (14)	59 (13)	56 (14)	59 (15)	0.18
Women, N (%)	161 (33)	44 (27)	52 (32)	65 (40)	0.04
Race and ethnicity, N (%)
Non-Hispanic White	287 (59)	105 (64)	94 (58)	88 (54)	0.18
Non-Hispanic Black	112 (23)	30 (18)	46 (28)	36 (22)	0.10
Asian or Pacific Islander	39 (8)	12 (7)	10 (6)	17 (10)	0.34
American Indian / Native Alaskan	10 (2)	2 (1)	3 (2)	5 (3)	0.49
Hispanic	24 (5)	6 (4)	7 (4)	11 (7)	0.39
Other	18 (4)	9 (5)	3 (2)	6 (4)	0.22
Estimated glomerular filtration rate, mL/min per 1.73 m2, mean (SD)	45 (24)	46 (24)	45 (24)	44 (24)	0.62
Urine albumin : creatinine ratio  mg/g creatinine, median (IQR)  mg/mmol creatinine, median (IQR	108 (16–686) 12.2 (1.8, 77.6)	69 (13–438) 7.8 (1.5, 49.5)	111 (17–889) 12.6 (1.9, 101)	213 (22–1003) 24.0 (2.5, 113)	0.02
Diabetes mellitus, N (%)	243 (50)	82 (50)	87 (53)	74 (45)	0.35
Prior atherosclerotic disease, N (%)	155 (32)	50 (30)	51 (31)	54 (33)	0.87
Peripheral vascular disease, N (%)	15 (3)	5 (3)	3 (2)	7 (4)	0.44
Myocardial infarction, N (%)	84 (17)	37 (23)	26 (16)	21 (13)	0.06
Stroke, N (%)	66 (13)	24 (15)	23 (14)	19 (12)	0.70
COPD, N (%)	29 (6)	6 (4)	11 (7)	12 (7)	0.31
Systolic blood pressure, mmHg, mean (SD)	133 (23)	128 (21)	135 (24)	134 (24)	0.02
Diastolic blood pressure, mmHg, mean (SD)	76 (14)	74 (13)	77 (15)	76 (14)	0.08
Body mass index, kg/m2, mean (SD)	31 (8)	33 (8)	31 (7)	30 (8)	<0.001
Current smoker, N (%)	97 (20)	25 (15)	38 (23)	34 (21)	0.18
Alcohol use, N (%)	91 (19)	26 (16)	29 (18)	36 (22)	0.57
LDL-C, mg/dL, mean (SD)  mmol/L, mean (SD)	102 (44) 2.6 (1.1)	102 (38) 2.6 (1)	106 (42) 2.7 (1.1)	99 (51) 2.6 (1.3)	0.41
HDL-C, mg/dL, mean (SD)  mmol/L, mean (SD)	42 (17) 1.1 (0.4)	39 (14) 1.0 (0.4)	42 (16) 1.1 (0.4)	45 (19) 1.2 (0.5)	0.004
Statins, N (%)	266 (54)	95 (58)	86 (53)	85 (52)	0.60
Other cholesterol meds, N (%)	34 (7)	14 (9)	7 (4)	13 (8)	0.26

COPD, chronic obstructive pulmonary disease; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; IQR, Interquartile range; SD, standard deviation

Correlation Between Lipid Species, Lipoproteins, and Statin Use

All lipid species were positively correlated with each other (Figure 1). The correlation was especially strong between sphingomyelins, with correlation coefficients ranging from 0.73 to 0.97 (Supplemental Figure 2). Additionally, strong correlations were noted among phosphatidylcholines with similar numbers of carbons (for instance, phosphatidylcholines 36:1 and 36:2), and among phosphatidylcholines with the same degree of unsaturation (e.g. phosphatidylcholines 34:2 and 36:2). Of note, each lipid species was only minimally correlated with total cholesterol, HDL-C, LDL-C, and statin use (correlation coefficients ranging from −0.17 to 0.34).

Figure 1:

Correlation of individual lipid species, lipoproteins, and statin use at baseline

Associations of eGFR with HDL Lipids

The mean levels of various classes of lipids were similar across the range of eGFR (Supplemental Figure 3). In unadjusted models, after accounting for the False Discovery Rate of 5%, there were no significant associations between eGFR and either classes of lipids (Supplemental Table 2) or individual lipid species (Supplemental Figure 4). After adjusting for potential covariates (age, biologic sex, diabetes, hypertension, prior stroke or myocardial infarction, HDL, LDL, and use of statins and other lipid-lowering medications), eGFR was not significantly associated with relative abundance of any class of lipids (Table 2). Lower eGFR was associated with greater relative abundance of ceramide 24:1, sphingomyelins 14:0 and 16:0, and phosphatidylcholines 28:0, 28:1, 29:0, 30:1, 30:2, and 35:4) though these associations were non-significant after controlling the FDR at 5% (Figure 2). Finally, the interaction between R-chain length and number of double bonds present in phosphatidylcholines was not significantly associated with mean relative abundance of phosphatidylcholines at any eGFR strata (Supplemental Figure 5).

Table 2:

Association between eGFR, albuminuria and levels of classes of lipids

	eGFR (per 15-point decrement) N = 497		Natural log-adjusted albuminuria (per doubling), N = 495
	Average difference in relative lipid abundance (95% CI)	p-value	Average difference in relative lipid abundance (95% CI)	p-value
Ceramides	0.10 (−0.05, 0.25)	0.18	0.11 (0.04, 0.18)	0.002
Sphingomyelins	0.06 (−0.06, 0.18)	0.35	0.05 (−0.01, 0.11)	0.08
Moderate-long R-chain	0.06 (0.00, 0.12)	0.05	0.04 (0.01, 0.07)	0.008
Very long R-chain	0.00 (−0.07, 0.06)	0.93	0.01 (−0.02, 0.04)	0.46
Phosphatidylcholines	0.19 (−0.28, 0.65)	0.43	0.09 (−0.13, 0.32)	0.41
0–1 double bonds	0.14 (−0.02, 0.30)	0.08	0.05 (−0.03, 0.12)	0.24
≥2 double bonds	0.05 (−0.27, 0.37)	0.77	0.05 (−0.11, 0.21)	0.54
Total lipids	0.35 (−0.36, 1.06)	0.34	0.26 (−0.08, 0.59)	0.13

Adjusted for age, gender, BMI, diabetes, hypertension, prior stroke or MI, HDL-C, LDL-C, and use of statins and other cholesterol-lowering medications

Moderate-long R-chain sphingomyelins had acylated fatty acids containing 14–18 carbons; very long R-chain sphingomyelins had acylated fatty acids containing 20–24 carbons

Monounsaturated and saturated phosphatidylcholines have 0–1 double bonds in the fatty acid chain; polyunsaturated phosphatidylcholines have 2 or more double bonds in the fatty acid chain

BOLD FONT indicates significant findings at the False Discovery Rate of 5%

Figure 2:

Association of eGFR with HDL lipids (Footnote: eGFR – estimated glomerular filtration rate; FDR – false discovery rate; BMI – body mass index; DM – diabetes mellitus; HTN – hypertension; MI – myocardial infarction; HDL-C – high-density lipoprotein cholesterol; LDL-C; low-density lipoprotein cholesterol).

Associations of Albuminuria with HDL Lipids

When modeled continuously with adjustment for multiple potential covariates as above, each doubling of the urine albumin-creatinine ratio was associated with a higher average relative abundance of ceramides of 0.11 (95% CI 0.04, 0.18) and moderate-long R-chain sphingomyelins (0.04, 95% CI 0.01, 0.07) (Table 2). In unadjusted analyses, albuminuria was significantly associated with higher relative abundance of ceramides 22:0, 24:0, and 24:1, hexosylceramide 16:0, sphingomyelin 16:0, and phosphatidylcholine 30:1, though these associations were not significant when controlling the FDR at 5% (Supplemental Figure 6). After adjustment as above, each doubling of the urine albumin-creatinine ratio was associated with higher relative HDL concentrations of ceramide 22:0 (increase in relative abundance 0.016, 95% CI 0.005, 0.027), ceramide 24:1 (increase in relative abundance 0.016, 95% CI 0.006, 0.026), hexosylceramide 16:0 (increase in relative abundance 0.023, 95% CI 0.011, 0.035), sphingomyelin 16:0 (increase in relative abundance 0.022, 95% CI 0.012, 0.032), phosphatidylcholine 29:0 (increase in relative abundance 0.015, 95% CI 0.005, 0.024), phosphatidylcholine 30:1 (increase in relative abundance 0.022, 95% CI 0.012, 0.033), and phosphatidylcholine 38:2 (increase in relative abundance 0.017, 95% CI 0.007, 0.027) (Figure 3). These associations remained significant after controlling the false discovery rate of 5%. Ceramide 24:0, lactosylceramide 16:0, and phosphatidylcholine 30:2 were also associated with albuminuria, though these associations were not significant at the false discovery rate <5%. The interaction of R-chain length and number of double bonds in phosphatidylcholines were not significantly associated with the mean relative abundance of phosphatidylcholines at any albuminuria class (<30 mg/g, 30–300 mg/g, and ≥300 mg/g, equivalent to <3.4, 3.4 - <33.9, ≥33.9 mg/mmol creatinine) (Supplemental Figure 7).

Figure 3:

Association of urine albumin creatinine ratio with HDL lipids (Footnote: uACR – urine albumin creatinine ratio; FDR – false discovery rate; BMI – body mass index; DM – diabetes mellitus; HTN – hypertension; MI – myocardial infarction; HDL-C – high-density lipoprotein cholesterol; LDL-C; low-density lipoprotein cholesterol).

Sensitivity Analyses

Compared to those not taking statins, participants taking statins had significantly lower relative abundance of hexosylceramide and several phosphatidylcholines with 1–2 double bonds, and higher relative abundance of several phosphatidylcholines with 4–5 double bonds, after adjusting for multiple potential confounders (Supplemental Table 3). Those taking statins had non-significantly lower relative abundance of ceramides present in HDL (mean difference in concentration −0.64, 95% CI −1.11, −0.17; adjusted p-value 0.064) (Supplemental Table 4).

DISCUSSION

In this cross-sectional analysis of data from a well-characterized CKD cohort, the urine albumin-creatinine ratio was directly correlated with the relative abundance of HDL ceramides, moderate-long R-chain sphingomyelins, ceramides 22:0 and 24:1, hexosylceramide 16:0, sphingomyelin 16:0, and phosphatidylcholines 29:0, 30:1, and 38:2. It is possible that alterations in HDL lipid composition may be associated with some of the functional HDL alterations observed in patients with CKD.

HDL is a family of the highest-density lipoproteins; the lipid portion of HDL includes cargo lipids and surface lipids in the lipid monolayer, including phosphatidylcholines and lipid raft components, mostly sphingolipids such as ceramides, glycoceramides, and sphingomyelins.(1, 2, 18) Sphingolipids are a large class of bioactive lipids with regulatory and signaling functions; central to their metabolism are ceramides, which may be converted to sphingomyelins, hexosylceramides, and lactosylceramides.(19) In this study, we noted associations between greater albuminuria and higher relative HDL concentrations of several sphingolipids.

Previous studies of ceramides in patients with CKD utilized whole-serum lipidomics rather than evaluating HDL specifically; though it is unknown if HDL and plasma lipidomics correlate, these studies still provide useful data for comparisons. For instance, Mantovani et al recently demonstrated that participants with an eGFR <60 mL/min/1.73 m² had higher plasma levels of ceramides 16:0, 18:0, 20:0, and 24:1 versus participants with greater eGFR.(13) Furthermore, compared with patients without albuminuria, patients with albuminuria (≥30 mg/g [≥3.4 mg/mmol] urine creatinine) had higher levels of ceramides 16:0, 18:0, 20:0, 22:0, and 24:0. This study was limited by a relatively low number of subjects with abnormal kidney function (65 with eGFR <60 mL/min/1.73 m², 44 with albuminuria ≥30 mg/g [≥3.4 mg/mmol] urine creatinine), and use of dichotomous albuminuria variables. Similarly, in a study of 93 children with CKD (eGFR 30–90 mL/min/1.73 m²) and 24 children without, participants with CKD had significantly higher levels of ceramides 24:0 and 24:1 and lactosylceramide 16:0.(20) In contrast, we did not find associations between eGFR and the relative abundance of ceramides, though as above, it is not clear if lipid measures in HDL and plasma are directly comparable. Higher albuminuria in patients with diabetes has been associated with greater serum levels of ceramides and sphingomyelins, specifically ceramide 16:0, glucosylceramide 18:0, and sphingomyelin 18:1.(21) While we did not measure these specific sphingolipids, we observed significant associations between albuminuria and relative HDL concentrations of ceramide 24:1, hexosylceramide 16:0, and sphingomyelin 16:0.

Levels of sphingomyelins and phosphatidylcholines have not been as well-evaluated as ceramides in patients with CKD. We found associations between greater proteinuria and higher relative HDL abundance of sphingomyelin 16:0 and phosphatidylcholines 29:0, 30:1, and 38:2. Albuminuria has previously been associated with elevated sphingomyelin levels in patients with type 1 diabetes when adjusting for eGFR.(22) Similarly, Afshinnia et al noted increased concentrations of long polyunsaturated sphingomyelins in participants with stage 5 versus less severe CKD.(9) In the same study, the authors noted a significant trend towards greater amounts of long, polyunsaturated phosphatidylcholines in participants with lower eGFR. While we failed to demonstrate similar associations between eGFR and phosphatidylcholine levels in HDL, albuminuria was positively associated with concentrations of phosphatidylcholines 29:0, 30:1, and 38:2 in HDL. As above, however, our study evaluated HDL specifically rather than whole plasma; it is unclear whether these results are comparable.

The potential biologic mechanisms underlying the associations between albuminuria and HDL concentrations of sphingolipids and phosphatidylcholines have been incompletely described, though lesser eGFR has been associated with impaired beta-oxidation, a key mechanism of fatty acid metabolism.(9, 23) Impaired beta-oxidation may cause intracellular accumulation of fatty acids, which has been associated with increased biosynthesis of long-chain lipids; both mechanisms maybe associated with greater plasma levels of long-chain polyunsaturated fatty acids.(24–26) In this study, lower eGFR was not associated with greater relative HDL concentrations of any phosphatidylcholines, which may be at odds with the mechanisms above; however as the prior studies evaluated whole plasma, it is unclear if these mechanisms may be generalized to HDL specifically. Proteinuria has been associated with increased toll-like receptor 4 (TLR-4) expression in diabetes and IgA nephropathy.(27, 28) Furthermore, albuminuria has been associated with increased tumor growth factor beta (TGF-β).(29, 30) Elevated TLR-4 may increase circulating ceramide levels via upregulation of various ceramide synthesis enzymes, and may also increase glycosphingolipid synthesis.(31–33) TGF-β has also been associated with ceramide biosynthesis in-vitro.(34) Further studies assessing specific biological pathways which are upregulated in patients with albuminuria would lend additional mechanistic insight. Additional work describing the functional implications of these specific HDL alterations may also be helpful.

Though determination of causation is not possible given our cross-sectional, observational study design, elevations in albuminuria, more than decrements in eGFR, may be associated with specific alterations in HDL lipid composition. This is consistent with literature suggesting that albuminuria is more strongly associated with cardiovascular disease than reductions in eGFR in patients with CKD.(35, 36) Given increasing evidence in the general population that higher plasma ceramide levels are associated with adverse cardiovascular events, this may represent an important target for reducing the disproportionately high rates of cardiovascular disease seen in patients with CKD.(37, 38) It is also possible that this underscores the importance of aggressive statin treatment in these patients. For instance, in a sensitivity analysis of the above data we found that the use of statins was significantly associated with lower relative HDL concentrations of hexosylceramides, lower levels of five different phosphatidylcholines with 1–2 double bonds, and higher levels of three phosphatidylcholines with 4–5 double bonds. Though these analyses were exploratory in nature, they appear to be in alignment with prior work demonstrating widespread decreases in levels of hexosylceramides and lactosylceramides in response to statin administration, which the authors hypothesized was due to inhibition of Rab prenylation-dependent synthesis mechanisms at the Golgi apparatus.(39) Similarly, prior in vitro studies have suggested increases in polyunsaturated fatty acids in HDL in response to statin treatment.(40)

Our study has several key strengths, including the use of a well-characterized CKD cohort and the targeted approach to lipid quantification with internal standards. However, there are several notable limitations. First, given the cross-sectional nature of the study, we are unable to make causal inferences or determine directionality of associations; it is possible that existing disordered HDL lipid composition may increase risk of albuminuria. Second, we measured only a limited number of lipid species, and were unable to assess the full complexity of the HDL lipidome; it is further possible that the HDL lipidome may not reflect the whole serum lipidome. We measured the concentrations of lipids relative to the concentrations of an internal calibrator, making the validity of our results contingent upon the inter-run consistency of the amount of calibrator present. This approach makes interpretation of our findings more difficult, as we lack definitive units for any of the lipids. Further, our use of LC-MS/MS means we are unable to state with certainty the number of carbons in a particular R chain. For species with 2 or more R chains (sphingolipids) we assumed the presence of the most common d(18:1) backbone, however, different backbone lengths may have been present; the functional significance of this limitation is unclear at this time. All comorbidities were by self-report, and it is possible that some were under-ascertained as a result. We lacked data on the etiology of CKD, as biopsies were not obtained as part of the study protocol. We were unable to normalize lipid levels to HDL particle numbers as these data were not available. Finally, the cohort was comprised of research volunteers from the Seattle area who were followed in a subspecialty clinic, which may introduce selection bias and limit generalizability to other CKD populations.

In summary, we found significant associations between albuminuria, but not eGFR, and the lipid composition of HDL in participants with CKD. The functional significance of these alterations and mechanisms underlying these relationships may warrant further investigation.

List of Abbreviations:

CKD

Chronic kidney disease

HDL

High-density lipoprotein

HDL-C

High-density lipoprotein cholesterol

LDL

Low-density lipoprotein

LDL-C

Low-density lipoprotein cholesterol

eGFR

Estimated glomerular filtration rate

LC-MS/MS

Liquid chromatography-tandem mass spectrometry

TLR-4

Toll-like receptor 4

TGF-β

Tumor growth factor beta

uACR

Urine albumin creatinine ratio

FDR

False discovery rate