Associations of plasma sphingolipids with measures of insulin sensitivity, β-cell function, and incident diabetes in Japanese Americans

Abstract

Aims:

To prospectively investigate associations of plasma sphingolipids with insulin sensitivity, β-cell function, and incident diabetes in the Japanese American Community Diabetes Study.

Methods and Results:

Baseline plasma samples from adults without diabetes (n=349; mean age 56.7 years, 51% men) were assayed for circulating ceramide and sphingomyelin species. Adjusted regression models examined cross-sectional and longitudinal associations with insulin sensitivity (HOMA2-%S), β-cell function (oral disposition index: DIo) and with incident diabetes over 5 years follow-up. Concentrations of four species (Ceramide C16:0, C18:0, C20:0, and C22:0) were inversely associated with HOMA2-%S at baseline (all P values <0.05, Q values <0.05) and change in HOMA2-%S over 5 years (all P values <0.05, Q values <0.05). No sphingolipids were associated with baseline or change in DIo. Of the four species associated with HOMA2-%S, only Ceramide C18:0 was significantly and positively associated with incident diabetes (RR/1SD 1.44, 95% CI 1.10–1.80, P=0.006, Q=0.024). The association of plasma Ceramide C18:0 with the risk of diabetes was partially mediated by change in HOMA2-%S between baseline and 5 years (mediation proportion: 61.5%, 95% CI 21.1%–212.5%).

Conclusion:

Plasma Ceramide C18:0 was associated with higher risk of incident diabetes which was partially mediated through a decrease in insulin sensitivity between baseline and five years. Circulating Ceramide C18:0 could be a potential biomarker for identifying those at risk of developing diabetes.

Introduction

Diabetes is a common chronic disease associated with significant morbidity and mortality [1]. The increasing prevalence of type 2 diabetes (T2D) observed worldwide is a heavy burden for both individual patients and society [2]. Thus, prevention is critical to reduce this burden. Prevention of T2D starts with identifying individuals at risk for the disease [3]. Efforts to identify novel risk factors or biomarkers related to development of disease may contribute to diabetes prevention.

Sphingolipids, including ceramides, sphingomyelins, and cerebrosides, are a class of lipids composed of a sphingoid backbone and a fatty acid. They have recently been suggested to act as modifiable risk factors that may influence the pathophysiology of diabetes or predict disease [4, 5]. In humans, plasma levels of several sphingolipid species are associated with fasting measures of insulin resistance and β-cell function [5–7], as well as with incident T2D, suggesting their possible role as biomarkers [5, 8–11]. In cultured cells, endogenous ceramides inhibit insulin-stimulated glucose uptake and glycogen synthesis by blocking activation of Akt/PKB [12–14]. Various animal and experimental studies have demonstrated that sphingolipid metabolites modulate signaling pathways implicated in β-cell dysfunction [4]. Studies in rodent models revealed that inhibition of ceramide biosynthesis improved insulin sensitivity and glucose tolerance [12, 15–17].

The current study utilized stored samples and data from the prospective longitudinal Japanese American Community Diabetes Study (JACDS), a community-based study of diabetes risk factors. The aims were to a) investigate the association of circulating ceramides and sphingomyelin species with insulin sensitivity and β-cell function assessed by an oral glucose tolerance test (OGTT), b) determine if certain ceramides or sphingomyelin species were associated with incident diabetes and c) examine mediators of the relationship between sphingolipids and incident diabetes. We hypothesized that plasma sphingolipid species associated with insulin sensitivity or β-cell function would be prospectively associated with incident diabetes and that any relationship between sphingolipids and incident diabetes would be mediated by insulin sensitivity or β-cell function. To our knowledge, this is the first study to investigate associations of plasma sphingolipids with insulin sensitivity, β-cell function, and incident diabetes in an Asian American population. Also, unlike other studies we used a measure of β-cell function derived from an OGTT to provide a more reliable measure of β-cell function than the fasting plasma insulin-based marker HOMA-B [18, 19].

Research Methods

Study Design

The study was a prospective, longitudinal, observational study of Japanese American adults without diabetes at baseline. We measured baseline concentrations of plasma sphingolipid species and examined associations with insulin sensitivity and β-cell function both cross-sectionally and longitudinally. For those species that were associated with insulin sensitivity or β-cell function in our analysis, we examined longitudinal associations between baseline plasma concentration and the risk of incident diabetes and performed mediation analysis. The study was approved by the University of Washington Human Subjects Division, and written informed consent was obtained from all participants at enrollment.

Study Population

Participants were taken from the parent JACDS study, a prospective community-based cohort of second- and third- generation Japanese Americans of 100% Japanese ancestry. Details of JACDS have been described previously [20, 21]. In brief, JACDS was designed to investigate risk factors for and prevalence of T2D and related conditions in Japanese Americans. Participants were enrolled through a community-wide recruitment strategy using a comprehensive mailing list and telephone directory that included nearly 95% of the Japanese American population of King County, Washington, between 1983 and 1991 (total n = 658). Follow-up visits were conducted at 5–6 years and 10–11 years after enrollment. For the current analysis, as no stored plasma samples were available from the original (entry) visit, the cohort comprises all participants with available fasting plasma samples (n=383) at the 5-year follow-up visit (n = 574), termed the JACDS Biomarker Discovery Project (BDP). For the purposes of the current analysis, the original JACDS 5-year visit is termed the BDP baseline visit, and the JACDS 10-year visit is termed the BDP follow-up visit. Of 383 individuals with available plasma samples at the 5-year visit, only those without diabetes at this time point (n=349) are included in this analysis. All of them also completed the 10-year follow-up evaluation thus providing 5 years of follow-up (Fig.1).

Figure 1.

Flow of participants across the study. BDP, Biomarker Discovery Project

Evaluations were performed at the General Clinical Research Center at the University of Washington, Seattle. At the BDP baseline and follow-up visits, a 75-gram OGTT was performed after a 10-h overnight fast and samples were collected just before and at 30, 60, and 120 min after ingestion of the oral glucose load by trained phlebotomists. Plasma was isolated and stored at −80° C. Plasma glucose was measured by the glucose oxidase method. Plasma insulin was measured using a modified double-antibody radioimmunoassay as described previously [22]. Diabetes diagnosis was based on fasting plasma glucose ≥126 mg/dl, 2-h plasma glucose ≥200 mg/dL [23] or self-report of use of glucose-lowering medications. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. The physical activity index was applied to responses to the Paffenbarger Physical Activity Questionnaire to estimate physical activity in kilocalories/week [24].

Calculations

Insulin sensitivity was estimated by homeostasis model assessment for insulin sensitivity (HOMA2-%S) and the Matsuda index. The HOMA2-%S was calculated using the Web-based HOMA calculator (http://www.dtu.ox.ac.uk/homacalculator) [25] and the Matsuda index was calculated from plasma glucose and insulin values in the fasting state and during the OGTT as follows: Matsuda index=10,000/(fasting plasma glucose × fasting plasma insulin × mean OGTT glucose concentration × mean OGTT insulin concentration)^0.5 [26]. As a composite measure of β-cell function, the oral disposition index (DIo) was calculated using the following formula: DIo=insulinogenic index × HOMA2-%S [22] where the insulinogenic index represents the ratio of increments in plasma insulin to increments in plasma glucose between basal and 30 minutes (ΔI_0–30/ΔG_0–30) from an OGTT [27].

Measurement of Sphingolipids

Sphingolipids were measured using stored plasma at the baseline BDP visit only. A targeted sphingolipid panel was assayed using liquid chromatography-electrospray ionization tandem mass spectroscopy as previously described [28]. Within this panel assay, 37 lipids were detected in the JACDS cohort, including ceramides, glucosylceramides, lactosylceramides, dihydroceramides, and gangliosides. Also, some sphingolipids, specifically sphingomyelins, were analyzed with an untargeted approach using a shotgun lipidomics assay. For the current report, we restricted analyses to the pre-specified sphingolipid species carrying saturated fatty acids that have been reported to be associated with insulin resistance or β-cell function in previous studies [5, 6, 11, 29]. This included 4 ceramide species, 3 sphingomyelins, 2 glucosylceramides, and 1 lactosylceramide containing distinct saturated fatty acids acylated to d18:1 sphingoid base backbone (i.e., Ceramide C16:0, Ceramide C18:0, Ceramide C20:0, Ceramide C22:0, Sphingomyelin C16:0, Sphingomyelin C18:0, Sphingomyelin C20:0, Glucosylceramide C22:0, Glucosylceramide C24:0 and Lactosylceramide C16:0). All sphingolipids measured in the targeted lipidomics assay were expressed in ng/mL and relative sphingomyelin concentrations (sphingomyelin C16:0, C18:0, and C20:0) measured in the shotgun lipidomics assay were expressed as a ratio to the area of an internal standard consisting of a non-naturally occurring sphingomyelin.

Plasma Lipidomics Analysis

Liquid Chromatography-Electrospray Ionization-Tandem Mass Spectrometry (LC-ESI-MS/MS) analysis of plasma of most sphingolipids was performed quantitatively using an AB Sciex quadrupole mass spectrometer 6500 (Sciex, Framingham, MA) equipped with an ESI probe and interfaced with the Agilent 1290 infinity LC system (Agilent, Palo Alto, CA). Sphingolipids were separated with a Poroshell 120 EC- C8 column, 2.1×50 mm, 2.7 μm (Agilent, Palo Alto, CA). Lipids from plasma were extracted using 1-phase extraction (methanol-dichloromethane), after internal standard addition. Quantification was performed using the ratio of analyte to internal standards relative to sphingolipid calibration curves.

Within the shotgun lipidomics assay plasma lipids, including sphingomyelins, triglycerides, cholesterol esters, phospholipids, and lysophospholipids, were extracted via a modified Folch extraction after internal standard addition. Analysis was conducted via flow injection ESI-MS/MS into a 5600 TripleTOF mass spectrometer (Sciex, Framingham, MA). For sphingomyelin analysis, the TOF analyzed was scanned with no fragmentation from 100 to 1200 Da. The high resolution of the TOF analyzer was used to separate sphingomyelin species from other isobaric analytes. Results are reported as ratios of analyte area to internal standard area.

Statistical analysis

To approximate a normal distribution and to facilitate comparisons with previous studies, sphingolipid species and outcomes were natural log(ln)-transformed. We fit multivariate linear regression models to examine the association of each sphingolipid species with baseline HOMA2-%S and change in HOMA2-%S between baseline and follow-up examinations. The associations between each sphingolipid species and DIo were tested in the same manner. For species that were associated with HOMA2-%S or DIo, we fit relative risk regression using a log-link binominal generalized linear model to examine the associations of sphingolipid species with the risk of incident diabetes. All models were adjusted for age, sex, BMI, low density lipoprotein (LDL) cholesterol, smoking status (i.e., current, former, and never) and physical activity.

Because we hypothesized that insulin sensitivity or β-cell function might mediate the association of sphingolipids with incident diabetes, we used the R mediation package to examine mediation effects in a secondary analysis for those sphingolipid species that were associated with incident diabetes. We estimated the total effect of a one standard-deviation increase in sphingolipid concentration on risk of incident diabetes, which can be decomposed into two components: the average direct effect from the sphingolipid (ADE) and the average causal mediation effect (ACME) that is due to the mediating variable of interest. These quantities are expressed as risk differences. When applicable, we also reported the proportion mediated, which can be conceptualized as ACME/total effect. To identify statistically significant mediating effects, we tested the null hypothesis that ACME=0 and estimated 95% CIs using nonparametic bootstrap confidence intervals with the percentile method (1000 replicates). P-values were adjusted for multiple comparison testing using Benjamini–Hochberg procedure with a false discovery rate (FDR) of 0.05, in which sphingomyelin species were examined separately. FDR adjusted P value (i.e., Q value) <0.05 was considered statistically significant. All analyses were performed using the Stata program, version 15.1 (Stata Corp., College Station, TX, USA) and R version 4.0.2.

Results

Participants at baseline were 56.7 years old on average with a mean BMI of 24.6 kg/m² and 49% were women (Table 1). We observed differences in metabolic parameters and sphingolipid concentrations by sex (Table 1) and BMI level (Supplemental Table 1). Women had higher HOMA2-%S and DIo at baseline compared to men. The values of all sphingolipid species except Ceramide C22:0 and Glucosylceramide C24:0 were higher in women than in men. Participants with BMI≥25 kg/m² had lower HOMA2-%S, Matsuda index, and DIo at baseline compared to those with BMI<25 kg/m². Baseline plasma concentrations of Ceramide C18:0, C20:0, and C22:0 were higher in participants with BMI≥25 kg/m² compared to those with BMI<25 kg/m².

Table 1.

Baseline characteristics of Japanese American Community Diabetes Study Biomarker Discovery Project participants, stratified by sex.

	All (n=349)	Female (n=171)	Male (n=178)	P value
Age, (years)	56.7±11.1	56.8±11.5	56.2±10.8	0.607*
BMI (kg/m2)	24.6±3.5	23.7±3.6	25.5±3.1	<0.001*
HOMA2-%S	58.4 (42.7, 76.0)	63.1 (45.0, 82.5)	52.1 (40.5, 69.3)	0.004†
Matsuda index	2.95 (2.18, 4.03)	3.12 (2.25, 4.17)	2.83 (2.16, 3.84)	0.091†
IGI (pmol/mmol)	105.0 (73.5, 163.8)	105.3 (75.4, 169.9)	104.7 (70.4, 158.5)	0.358†
DIo (IGI × HOMA2-%S)	5904 (4138, 9318)	6582 (4591, 10404)	5609 (3876, 8414)	0.027†
LDL-cholesterol (mg/dl)	130.4±35.1	123.1±34.3	137.3±34.5	<0.001*
Physical activity (kcal/week)	2672±1811	2081±1440	3240±1948	<0.001*
Smoking status
Current smoker	38 (10.9)	17 (9.9)	21 (11.8)
Former smoker	141 (40.4)	41 (24.0)	100 (56.2)	<0.001**
Never smoker	170 (48.7)	113 (66.1)	57 (32.0)
Sphingolipid concentration
Ceramide 16:0 (ng/ml)	124.5±33.8	129.6±34.3	119.7±32.2	0.005*
Ceramide 18:0 (ng/ml)	82.8±36.0	88.1±37.2	77.5±34.5	0.006*
Ceramide 20:0 (ng/ml)	243.6±90.7	256.9±103.0	230.9±75.2	0.008*
Ceramide 22:0 (ng/ml)	795.4±295.3	791.6±295.9	790.2±296.2	0.965*
GluCer 22:0 (ng/ml)	119.8±47.5	127.1±51.0	112.7±42.8	0.005*
GluCer 24:0 (ng/ml)	270.1±104.5	278.6±114.5	262.0±93.4	0.140*
LacCer 16:0 (ng/ml)	3566±1398	3869±1534	3276±1188	<0.001*
Sphingomyelin 16:0	9.54±2.78	10.02±2.87	9.07±2.61	0.001*
Sphingomyelin 18:0	0.95±0.26	1.03±0.24	0.87±0.24	<0.001*
Sphingomyelin 20:0	0.49±0.12	0.51±0.12	0.47±0.12	0.002*

Data are mean ± SD, median (IQR), or n (%). BMI, body mass index; HOMA2-S, homeostasis model assessment of insulin sensitivity; IGI, insulinogenic index; DIo, oral disposition index; LDL, low density lipoprotein; GluCer, glucosylceramide; LacCer, lactosylceramide; SD, standard deviation.

^*By t test.

^†By Wilcoxon rank sum test.

^**By chi square test.

In multiple linear regression models with adjustment for baseline age, sex, BMI, LDL-cholesterol, smoking status, and physical activity, higher plasma concentrations of lnCeramide-C16:0, lnCeramide-C18:0, lnCeramide-C20:0, and lnCeramide-C22:0 were significantly associated with lower values of baseline lnHOMA2-%S and lnMatsuda-index (Table 2 and supplemental Table 2). Inverse associations were also found between plasma ceramide species lnCeramide-C16:0, lnCeramide-C18:0, lnCeramide-C20:0, lnCeramide-C22:0 and log-transformed change in HOMA2-%S from baseline to 5-year follow-up (Table 2). In the relationship with the change in Matsuda index, only Ceramide C18:0 showed an inverse association (Supplemental Table 2). Except for sphingomyelin C16:0 and C20:0, other plasma sphingomyelin species and ceramide derivatives were not associated with baseline or change in HOMA2-%S and Matsuda index. Sphingomyelin C16:0 showed a significant positive association with baseline Matsuda index, whereas Sphingomyelin C20:0 was inversely associated. However, these two species of sphingomyelin had no association with change in Matsuda index and HOMA2-%S. There was no association between any plasma sphingolipid species and DIo at baseline or change in DIo over 5 years (Table 2 and supplemental Table 2). During the 5 years of follow-up, 39 of the 349 participants (11.2%) progressed to diabetes. Of these, 37 were newly diagnosed based on the results of the follow-up OGTT and 2 were diagnosed during the follow-up period based on report of taking a glucose-lowering drug. As shown in Table 3, participants who developed diabetes were more likely to be older and current smokers, to have higher BMI, and to have lower HOMA2-%S and DIo at baseline.

Table 2.

Association of plasma sphingolipids with HOMA2-%S and oral disposition index

	ln (HOMA2-%S) at baseline			ln (DIo) at baseline
	β-coefficient (95% CI)	P value	Q value†	β-coefficient (95% CI)	P value	Q value†
lnCer-C16:0	−0.201 (−0.362, −0.041)	0.014	0.035	−0.301 (−0.582, −0.020)	0.036	0.133
lnCer-C18:0	−0.144 (−0.245, −0.043)	0.005	0.017	−0.038 (−0.216, 0.197)	0.670	0.931
lnCer-C20:0	−0.208 (−0.326, −0.091)	0.001	0.010	−0.198 (−0.408, 0.011)	0.063	0.157
lnCer-C22:0	−0.194 (−0.315, −0.074)	0.002	0.010	−0.222 (−0.434, −0.011)	0.040	0.133
lnGluCer-C22:0	−0.042 (−0.153, 0.069)	0.463	0.514	0.103 (−0.091, 0.299)	0.297	0.594
lnGluCer-C24:0	−0.072 (−0.189, 0.038)	0.215	0.275	−0.004 (−0.202, 0.194)	0.969	0.969
lnLacCer-C16:0	0.076 (−0.045, 0.197)	0.220	0.275	0.018 (−0.194, 0.231)	0.868	0.964
lnSM-C16:0	0.196 (0.018, 0.374)	0.031	0.052	−0.051 (−0.365, 0.261)	0.745	0.931
lnSM-C18:0	−0.051 (−0.245, 0.143)	0.607	0.607	−0.120 (−0.460, 0.219)	0.487	0.811
lnSM-C20:0	−0.226 (−0.441, −0.023)	0.029	0.052	−0.401 (−0.766, −0.032)	0.031	0.133
	ln (Change of HOMA2-%S)*			ln (Change of DIo)**
	β-coefficient (95% CI)	P value	Q value†	β-coefficient (95% CI)	P value	Q value†
lnCer-C16:0	−0.189 (−0.318, −0.061)	0.004	0.010	−0.028 (−0.085, 0.029)	0.314	0.628
lnCer-C18:0	−0.188 (−0.267, −0.108)	<0.001	<0.003	−0.021 (−0.057, 0.015)	0.260	0.628
lnCer-C20:0	−0.160 (−0.257, −0.064)	0.001	0.003	−0.025 (−0.068, 0.017)	0.241	0.628
lnCer-C22:0	−0.173 (−0.270, −0.076)	<0.001	<0.003	−0.013 (−0.055, 0.031)	0.567	0.652
lnGluCer-C22:0	−0.068 (−0.156, 0.021)	0.136	0.194	0.014 (−0.025, 0.054)	0.446	0.650
lnGluCer-C24:0	−0.047 (−0.138, 0.044)	0.309	0.354	0.006 (−0.034, 0.046)	0.778	0.778
lnLacCer-C16:0	−0.027 (−0.124, 0.071)	0.591	0.591	0.033 (−0.010, 0.076)	0.135	0.628
lnSM-C16:0	0.147 (0.003, 0.290)	0.045	0.075	0.070 (0.007, 0.134)	0.030	0.300
lnSM-C18:0	−0.167 (−0.322, −0.012)	0.035	0.070	−0.020 (−0.089, 0.050)	0.587	0.652
lnSM-C20:0	−0.086 (−0.256, 0.084)	0.319	0.354	−0.029 (−0.104, 0.046)	0.455	0.650

Model was adjusted for age, sex, BMI, LDL-C, physical activity and smoking status.

^{*, **}Additionally adjusted for baseline *HOMA2-S or **DIo.

^†Corrected for multiple comparison with Benjamini–Hochberg method (with FDR of 0.05).

HOMA2-S, homeostasis model assessment of insulin sensitivity; DIo, oral disposition index; BMI, body mass index; LDL-C, low density lipoprotein-cholesterol; FDR, false discovery rate; Cer, ceramide; GluCer, glucosylceramide; LacCer, lactosylceramide; SM, sphingomyelin.

Table 3.

Baseline characteristics by development of diabetes

	No incident DM (n=310)	Incident DM (n=39)	P value
Age, (years)	56.2±11.0	60.8±11.1	0.015*
Male (%)	157 (50.6)	21 (53.8)	0.695**
BMI (kg/m2)	24.5±3.4	26.1±3.9	0.003*
HOMA2-%S	59.8 (43.9, 76.4)	45.9 (29.3, 61.1)	<0.001†
Matsuda index	3.07 (2.29, 4.13)	2.32 (1.28, 3.14)	<0.001†
IGI (pmol/mmol)	107.7 (75.2, 163.9)	91.1 (45.5, 146.7)	0.171†
DIo (IGI × HOMA2-%S)	6104 (4421, 9562)	3554 (2208, 7627)	<0.001†
LDL-cholesterol (mg/dl)	129.2±35.3	139.8±31.6	0.074*
Physical activity (kcal/week)	2693±1842	2511±1556	0.556*
Smoking status
Current smoker	29 (9.4)	9 (23.1)
Former smoker	129 (41.6)	12 (30.7)	0.029**
Never smoker	152 (49.0)	18 (46.2)
Sphingolipid concentration
Ceramide C16:0 (ng/ml)	124.3±32.6	129.7±40.6	0.341*
Ceramide C18:0 (ng/ml)	80.9±33.1	97.5±51.9	0.006*
Ceramide C20:0 (ng/ml)	241.8±88.3	265.0±103.8	0.126*
Ceramide C22:0 (ng/ml)	787.9±294.6	854.7±306.1	0.179*
Glucosylceramide C22:0 (ng/ml)	121.2±47.9	108.4±42.3	0.087*
Glucosylceramide C24:0 (ng/ml)	274.7±106.7	278.1±158.6	0.854*
Lactosylceramide C16:0 (ng/ml)	3546±1343	3699±1733	0.514*
Sphingomyelin C16:0	9.70±2.88	9.31±2.60	0.201*
Sphingomyelin C18:0	0.95±0.24	1.01±0.34	0.273*
Sphingomyelin C20:0	0.49±0.12	0.50±0.12	0.613*

Data are mean ± SD, median (IQR), or n (%). BMI, body mass index; HOMA2-S, homeostasis model assessment of insulin sensitivity; IGI, insulinogenic index; DIo, oral disposition index; LDL, low density lipoprotein; SD, standard deviation.

^*By t test.

^†By Wilcoxon rank sum test.

^**By chi square test.

Of the four sphingolipid species that were associated with HOMA2-%S (Ceramide C16:0, Ceramide C18:0, Ceramide C20:0, and Ceramide C22:0), only Ceramide C18:0 was associated with a higher risk of incident diabetes after adjustment for baseline age, sex, BMI, LDL-cholesterol, smoking status, and physical activity, showing an adjusted risk ratio (RR) per 1SD increment of 1.44 (95% CI 1.10–1.80) (Table 4). There was no evidence of first-order multiplicative interactions by sex or BMI (data not shown). In mediation analysis, the association of plasma Ceramide C18:0 with the risk of incident diabetes was partially mediated by change in HOMA2-%S between baseline and five years (proportion mediated: 61.5%, 95% CI 21.1–212.5%) (Table 5).

Table 4.

Risk ratio for incident diabetes by sphingolipid species

	Risk ratio (95% CI) (per SD increment)
	Adjusted for age and sex	P value	Q value*	Adjusted for age, sex, BMI, LDL-C, PA, and smoking status	P value	Q value*
Ceramide C16:0	1.17 (0.87–1.56)	0.295	0.295	1.17 (0.90–1.54)	0.242	0.323
Ceramide C18:0	1.59 (1.25–2.01)	<0.001	<0.004	1.44 (1.10–1.80)	0.006	0.024
Ceramide C20:0	1.31 (0.99–1.73)	0.056	0.112	1.07 (0.80–1.42)	0.647	0.647
Ceramide C22:0	1.25 (0.96–1.63)	0.095	0.126	1.22 (0.91–1.64)	0.176	0.323

Data represent odds ratio and 95% CI for incident diabetes.

^*Corrected for multiple comparison with Benjamini–Hochberg (with FDR of 0.05).

SD, standard deviation; BMI, body mass index; LDL-C, low density lipoprotein-cholesterol; PA, physical activity; FDR, false discovery rate.

Table 5.

Mediation model of insulin sensitivity in the association between plasma Ceramide C18:0 and incident diabetes

	Risk differences *
	Total effect	Average causal mediation effect	Average direct effect	Proportion mediated
Mediator	% (95% CI)	% (95% CI)	% (95% CI)	% (95% CI)
Baseline HOMA2-%S	2.1 (−2.2, 2.6)	0.1 (0–2.0)	2.0 (−4.1, 2.4)	2.9 (−3.8, 594.9)
Change in HOMA2-%S†	2.4 (0.9–3.5)	1.6 (0.5–2.9)	0.8 (−1.4, 2.1)	61.5 (21.1–212.5)

Data are the expected probability of incident diabetes due to the total effect from a 1SD increase in Ceramide C18:0, the effect mediated through change in HOMA2-%S, and direct effect due to Ceramide C18:0.

^*Models are adjusted for age, sex, body mass index, physical activity, and smoking status.

^†Additionally adjusted for baseline HOMA2-S. Model was run using 1000 simulations with percentile CIs.

HOMA2-%S, homeostasis model assessment of insulin sensitivity; SD, standard deviation.

Discussion

In a community-based cohort of second- and third-generation Japanese Americans, fasting concentrations of plasma ceramide of varying acyl-chain lengths (i.e., Ceramide C16:0, Ceramide C18:0, Ceramide C20:0, Ceramide C22:0) were inversely associated with insulin sensitivity (i.e., positively associated with insulin resistance). On the other hand, there were no significant associations of plasma ceramides or related sphingolipids with β-cell function. Of the four ceramide species that were associated with insulin sensitivity, only Ceramide C18:0 was associated with the risk of incident diabetes. This relationship was partially mediated through change in insulin sensitivity between baseline and follow-up.

While several observational studies have evaluated associations of plasma sphingolipid species with measures of insulin resistance, β-cell function, or incident diabetes [5–11], this study adds to our knowledge by examining a unique population of second- and third-generation Japanese Americans, utilizing an OGTT-based measure of β-cell function and insulin sensitivity, and performing mediation analysis. Each study had ethnically different backgrounds and was conducted with different analytical methodologies. Nonetheless, there are some consistent findings among these studies, which are also similar to the findings in the current report. That is, long- and very long-chain plasma ceramides are associated with insulin resistance or incident diabetes [5–10].

In the Strong Heart Family Study, a cohort of Americans Indians (n=2,086), higher levels of plasma Ceramide C16:0, Ceramide C18:0, Ceramide C20:0, and Ceramide C22:0 were cross-sectionally associated with higher HOMA-IR [6]. Clinical observations from a multi-ethnic cohort study (n=1,557) revealed that Ceramide C16:0 and Ceramide C18:0 in plasma were also positively associated with HOMA-IR in cross-sectional analyses [7]. In ethnic Chinese in Singapore (n=2,302), the ceramides with the strongest correlations were those containing a C18:0, C20:0, or C22:0 N-linked fatty acid [5]. In our study, baseline levels of Ceramide C16:0, C18:0, C20:0, and C22:0 were associated with both baseline HOMA-%S and Matsuda index, and these findings increase our confidence for their role in insulin sensitivity.

Unlike change in HOMA-2S%, change in the Matsuda index was not associated with Ceramide C16:0, C20:0 or C22:0. HOMA-2S% and Matsuda index are measuring slightly different things – HOMA based on fasting values measuring primarily hepatic insulin sensitivity while the Matsuda index measuring both hepatic and peripheral tissue (primarily muscle) insulin sensitivity [26, 30]. Some individuals with type 2 diabetes have a predominant defect in hepatic insulin sensitivity and present with fasting hyperglycemia, whereas others have a more pronounced disturbance in peripheral (primarily muscle) sensitivity to insulin and present with post-meal glucose intolerance [26]. Of the four ceramide species, only Ceramide C18:0 was associated with both change in HOMA-2S% and Matsuda Index and incident diabetes, suggesting that this ceramide is probably the most predictive or the most pathogenic sphingolipid in the pathophysiology to focus research on.

Others have also found associations between ceramide species and incident diabetes. In a recent meta-analysis that included 2 studies with participants from 12 American Indian communities (total n=2,337), higher concentrations of circulating Ceramide C18:0, C20:0, and C22:0 were each associated with a higher risk of incident diabetes [8]. Likewise, high levels of plasma Ceramide C16:0, C18:0, C20:0, and C22:0 were associated with a higher risk of diabetes in the Cardiovascular Health Study of 3,645 White and Black adults [10]. Similar to our findings, Ceramide C18:0 showed the strongest association with incident diabetes among the ceramide species in both a prospective cohort in Finland (n=8,045) [9] and the study of Chinese Singaporeans mentioned above [5]. Mediation analysis in our study related this association to change in insulin sensitivity. More than 60% of the risk associated with circulating Ceramide C18:0 could be accounted for by changes in HOMA2-%S between baseline and follow-up. These findings suggest that this relationship is mediated in part by decreasing insulin sensitivity. To our knowledge, this study is the first to show the relationship of ceramide levels, insulin sensitivity, and incident diabetes using mediation methods.

Regarding the association between sphingolipids and β-cell function, the findings are inconsistent among studies. The Strong Heart Family Study showed that higher levels of plasma Ceramide C20:0 and Ceramide C22:0 were associated with higher baseline HOMA-B [6]. On the other hand, in a population-based cohort study in China (n=1,974), four ceramide species (i.e., Ceramide C18:1, C20:0, C20:1, and C22:1) were inversely associated with baseline HOMA-B, and mediation analysis revealed a significant association of these ceramide species with incident diabetes that was partially mediated by HOMA-B (mediation proportion:11–42%), rather than HOMA-IR [11]. In our study, there was no association of plasma ceramide species with β-cell function assessed by the oral disposition index, which, to the best of our knowledge, has not been previously employed in this research.

Ceramide synthases (CerS) are central enzymes required for the synthesis of ceramides. Mammalian CerS exists in six isoforms (CerS1–6), each of which generate different proportions of ceramides with distinct acyl-chain lengths [31]. CerSs vary in tissue distribution, with each tissue having a unique profile of CerS expression [32]. Especially, the synthesis of Ceramide C18:0 is primarily involved by CerS1, which is the most abundant isoform in skeletal muscle [31, 32]. Considering that skeletal muscle is the predominant site of insulin-mediated glucose uptake and a major tissue of insulin resistance [33], it can be inferred that Ceramide C18:0 may be a predominant species of ceramide in the development of insulin resistance. Indeed, mice lacking muscle-specific CerS1 exhibited reduced muscle Ceramide C18:0 contents and improvement in systemic insulin resistance [34]. In insulin-resistant human subjects, concentration of Ceramide C18:0 in plasma and muscle correlated positively [35].

While our findings emphasize the usefulness and potential of circulating ceramides as biomarkers for insulin resistance and incident diabetes, this does not imply that circulating ceramides play a causative role. This is because there is still a lack of knowledge of how circulating sphingolipids are transported from cell to cell or organ to organ within the body, exactly where they originate from and are being directed to, and how they adapt during metabolic disease [29]. However, evidence continues to emerge suggesting the possibility that circulating ceramides may play a causative role in metabolic diseases. In a rodent model, overexpression of ceramidase in the liver improved adipose insulin sensitivity, inflammation, and fibrosis. Conversely, overexpression of ceramidase within adipose tissue improved hepatic lipid accumulation. When ceramide species were acutely broken down by ceramidase in liver and adipose tissue, there was also a corresponding significant lowering of plasma ceramides [35]. In other studies, infusing LDL-ceramide 24:0 into lean mice caused whole-body insulin resistance, an effect mediated by decreased insulin action in skeletal muscle [36].

Strengths and limitations of the study

The strengths of the study include the use of an OGTT to assess β-cell function and Matsuda index, the study of a unique cohort of second- and third- generation Japanese Americans and the prospective, longitudinal follow-up and evaluation for incident diabetes using OGTT. Also, this is the first lipidomic study to be conducted in an Asian-American population. There are also some important limitations. Given the sample size, we were underpowered to detect an exhaustive list of plasma sphingolipid species associated with β-cell function. We cannot rule out the possibility of residual confounding, given the observational nature of the study. Our study cohort included a single ethnicity that limits external generalizability of our findings. Further, this population, which is of 100% Japanese ancestry but exposed to a western lifestyle, may have risk factors for diabetes that are not present in other populations. However, the study of a single population group reduces the potential for unmeasured confounding. The number of incident diabetes was small and is also a limitation.

In summary, our analysis found that four plasma ceramide species (i.e., Ceramide C16:0, C18:0, C20:0, and C22:0) were inversely associated with insulin sensitivity at baseline as well as change over time. On the other hand, we did not observe any association of plasma ceramides and related sphingolipids with measures of β-cell function derived from an OGTT. Of the four ceramide species, only plasma Ceramide C18:0 was associated with the risk of incident diabetes and this relationship was partially mediated through a decrease in insulin sensitivity between baseline and follow-up. Our findings suggest that circulating Ceramide C18:0 could be a potential biomarker for identifying individuals at risk of developing type 2 diabetes.

Highlights.

1. Plasma Ceramides C16:0, C18:0, C20:0, and C22:0 were inversely associated with insulin sensitivity.

2. Plasma Ceramide C18:0 was associated with the risk of incident diabetes and this was partially mediated through a decrease in insulin sensitivity.

3. Plasma Ceramide C18:0 could be a potential biomarker for identifying individuals at risk of developing diabetes.