Characterizing a Common CERS2 Polymorphism in a Mouse Model of Metabolic Disease and in Subjects from the Utah CAD Study

Abstract

Context

Genome-wide association studies have identified associations between a common single nucleotide polymorphism (SNP; rs267738) in CERS2, a gene that encodes a (dihydro)ceramide synthase that is involved in the biosynthesis of very-long-chain sphingolipids (eg, C20-C26) and indices of metabolic dysfunction (eg, impaired glucose homeostasis). However, the biological consequences of this mutation on enzyme activity and its causal roles in metabolic disease are unresolved.

Objective

The studies described herein aimed to characterize the effects of rs267738 on CERS2 enzyme activity, sphingolipid profiles, and metabolic outcomes.

Design

We performed in-depth lipidomic and metabolic characterization of a novel CRISPR knock-in mouse modeling the rs267738 variant. In parallel, we conducted mass spectrometry-based, targeted lipidomics on 567 serum samples collected through the Utah Coronary Artery Disease study, which included 185 patients harboring 1 (n = 163) or both (n = 22) rs267738 alleles.

Results

In-silico analysis of the amino acid substitution within CERS2 caused by the rs267738 mutation suggested that rs267738 is deleterious for enzyme function. Homozygous knock-in mice had reduced liver CERS2 activity and enhanced diet-induced glucose intolerance and hepatic steatosis. However, human serum sphingolipids and a ceramide-based cardiac event risk test 1 score of cardiovascular disease were not significantly affected by rs267738 allele count.

Conclusions

The rs267738 SNP leads to a partial loss-of-function of CERS2, which worsened metabolic parameters in knock-in mice. However, rs267738 was insufficient to effect changes in serum sphingolipid profiles in subjects from the Utah Coronary Artery Disease Study.

Cardiometabolic disease is the leading cause of death worldwide (1). Its associated disorders and risk factors include elevated fasting blood sugar, insulin resistance, dyslipidemia, hypertension, and abdominal obesity, which predispose individuals to diabetes and its complications, heart failure, myocardial infarction, and stroke. Recent advances toward identifying genetic variations associated with disease traits have improved with the use of genome-wide association studies (GWAS). The challenge arises with bridging correlations of genetic variants and the mechanistic understanding of underlying biological processes to better assess disease risk and provide targeted and timely interventions (2).

The presence of a common single nucleotide polymorphism (SNP; rs267738) is associated with elevated glycated hemoglobin levels and an impaired glomerular filtration rate (3–8). The variant is particularly prevalent (20–25%) in individuals of northern European descent, and homozygosity for the mutant allele in this population is as common as 3% to 5% (9). The missense mutation resides in the gene coding for (dihydro)ceramide synthase 2 (CERS2), which functions alongside 5 other CERS enzymes (CERS 1–6) in the de novo sphingolipid synthesis pathway to attach variable-length acyl groups to sphinganine, producing dihydroceramides. CERS2 is highly expressed in the liver and kidney and preferentially facilitates the synthesis of very-long-chain (VLC) dihydroceramides, with variable acyl groups spanning 20–26 carbons in length (Fig. 1) (10).

Figure 1.

Schematic of de novo ceramide synthesis. Palmitate and serine are the preferred substrates for de novo ceramide synthesis, which undergo a condensation reaction in the first of four pathway steps. Next, 3-keto-sphinganine is quickly converted to sphinganine, which is n-acylated by 6 CERS enzymes to produce dihydroceramide. Lastly, DES 1 and 2 introduce a double bond at the delta 4 position of the sphingoid backbone to form ceramide.

Abbreviations: CERS, (dihydro) ceramide synthase; DES, dihydroceramide desaturase; KDSR, 3-keto-sphinganine reductase; SPT, serine palmitoyltransferase.

Ceramide and related complex sphingolipid species are potent mediators of cardiometabolic disease (11–16), and recent reports suggest that the length of the variable acyl chain incorporated by CERS enzymes determines a sphingolipid’s pathophysiological role. In general, long-chain (LC) C16 and C18 ceramides are thought to drive metabolic dysfunction by impairing mitochondrial bioenergetics, inhibiting insulin signaling and inducing apoptosis. By contrast, the VLC C > 20 sphingolipids are likely benign or beneficial (14, 17–19). As such, deleterious outcomes driven by alterations in the LC/VLC sphingolipid ratio can exist without changes to total sphingolipid concentrations (20). Indeed, ratios of LC and other ceramides versus the VLC C24 species strongly correlate with a number of cardiometabolic disease endpoints, including cardiovascular death, showing strong predictive value in longitudinal studies (21–29). Because of these associations, clinics have started measuring these ceramide ratios as markers of disease risk (30).

The function of CERS2 has been evaluated in genetically engineered mice. Homozygous Cers2 knockout mice present with numerous liver deformities and abnormalities, contributing to impaired glycogen storage, reduced electron transport chain (ETC) activity, and hepatic insulin resistance (31–33). Mice that are heterozygous for Cers2 appear healthy when maintained on a normal chow diet, but show enhanced hepatic steatosis and insulin resistance when challenged with a high-fat diet (HFD) (19, 20). These various Cers2 knockout strains show alterations in sphingolipid profiles, including a decline in VLC ceramides and a compensatory increase in C16 ceramide. Importantly, this phenomenon drives metabolic dysfunction in Cers2 heterozygous mice without changing total ceramide levels (20).

The purpose of our study was to investigate the functional outcomes of the CERS2 rs267738 SNP on sphinganine acylation and metabolic function. Herein, we demonstrate that CRISPR knock-in mice homozygous for the rs267738 SNP have impaired CERS2 activity and worsened glucose tolerance and liver steatosis. Next, we characterize circulating sphingolipid profiles in humans harboring the mutation and determine that, in the present study, the rs267738 SNP is not associated with alterations in serum sphingolipid acylation patterns.

Materials and Methods

Human subjects

Clinical data and serum samples were collected from a subset of the Utah Coronary Artery Disease (CAD) case-control study (34, 35). Whole exome sequencing was performed by the Broad Institute Genomic Services to identify individuals harboring 1 or 2 copies of the rs267738 missense mutation. A final cohort was built based on available biospecimens, which integrated clinical, genetic, and sphingolipidomic data for 567 subjects. Our study sample included 185 rs267738 carriers, with 163 individuals harboring a single copy and 22 harboring both mutant alleles. Comparison of participant characteristics of the analytic cohort with the original Utah CAD cohort is presented in Supplemental Table 1 (36).

Clinical and demographic characteristics were obtained from participants, along with an extensive medical and family history, by trained interviewers (34). Blood samples were collected following a 12- to 16-hour overnight fast, and specimens were prepared according to the guidelines of the Lipid Research Clinic’s Program Manual of Laboratory Operations. Lipoprotein concentrations were detected using a microscale ultracentrifugation method (37, 38). Serum was collected and stored at -80°C, and samples were recovered to perform targeted liquid chromatography tandem mass spectrometry (LS-MS/MS) for measurement of serum sphingolipids.

Animals

All animal procedures were performed in compliance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Utah and adhered to National Institutes of Health (NIH) standards. Male (n = 10–12) and female (n = 10–12) mice were maintained under standard laboratory conditions at a temperature of 22°C to 24°C, in groups of 2 to 5 mice, with a 12-hour light/dark cycle. Mice were allowed ad libitum access to food and water at all times, unless fasting conditions were required for experimental procedures. Animals were fed a normal chow diet from the age of 4 weeks and transitioned to a high-fat diet (60% total energy; D12492; Research Diets Inc, New Brunswick, NJ, USA) at the age of 15 weeks.

Generation of Cers2 E115A mice

CRISPR Cas9 constructs and reagents were designed by the University of Utah Mutation Generation and Detection Core to introduce the human single-base-pair missense mutation resulting in the E115A substitution within CERS2. A guide ribonucleic acid (gRNA) was selected to direct Cas9-mediated cutting near the target nucleotide. A single-stranded oligo deoxyribonucleic acid (DNA) nucleotide (ssODN) donor was generated to introduce the knock-in mutation, which also contained stabilizing 5’ and 3’ phosphorothioate modifications and a mutated protospacer adjacent motif sequence, creating a unique restriction enzyme cut site. Embryos were collected 0.5 days after fertilization by the Transgenic and Gene Targeting Core at the University of Utah. A ribonucleoprotein complex of gRNA and Cas9 protein was co-microinjected with the ssODN donor into embryo pronuclei. Embryos were rinsed and surgically implanted into oviducts of 0.5-day pseudopregnant females. Implanted females were housed 2 per cage and allowed to sire and nurture pups naturally. Eight optimal founders were mated to C57Bl6/J wildtype mice to generate N1 mice. Two N1 mice were found to have the correct gene sequence and were used to generate a mutant colony for research studies. Genotypes of subsequent mice were confirmed by polymerase chain reaction (PCR) amplification of restriction enzyme digests and Sanger sequencing performed by the University of Utah DNA Sequencing Core facility. All protocols were in compliance with AALAC procedures and approved by the IACUC committee.

CERS2 activity assay

CERS2 activity was assessed as described previously (39). Briefly, liver homogenate was incubated with 100 μM of defatted BSA, 1 mM of NBD-labeled sphinganine, and 5 mM of 24:1 fatty acyl-CoA (640012; Avanti Polar Lipids) in a 15.5-μL reaction volume. The reaction was terminated after 30 minutes, with the addition of methanol containing 1% formic acid. Samples were loaded onto an SPE column (8E-S001-BGB; Phenomenex) and washed with water containing 1% formic acid. Residual NBD-sphinganine was eluted in a 30:14:6:1 solution of methanol:water:chloroform:formic acid containing 10 mM of ammonium acetate. NBD-ceramide was eluted and collected with a 30:14:6:1 solution of methanol:chloroform:water:formic acid containing 10 mM of ammonium acetate. Sample fluorescence intensity was measured using a multiwell plate reader (NBD λex = 465 nm, λem = 53 5nm). NBD-ceramide was quantified utilizing a standard curve. All samples were run in triplicate and averaged prior to statistical analysis.

Western blot analysis

Protein from liver homogenate was resolved by SDS-PAGE and transferred to nitrocellulose membranes. CERS2 protein was detected using rabbit polyclonal anti-CERS2 antibody (Cat# HPA027262; Sigma, Missouri, USA; RRID:AB_1852717). Equal loading was confirmed using rabbit monoclonal anti-beta actin antibody (Cat# 4970l; Cell Signaling, Massachusetts, USA; RRID:AB_2223172). Donkey anti-rabbit horseradish peroxidase (Cat# NA9340; Cytiva, Massachusetts, USA; RRID:AB_772191) was used for secondary antibody. Target proteins were detected using ECL developing reagents (#RPN2235; Cytiva).

Glucose and insulin tolerance tests

Glucose tolerance tests were performed in 21-week-old mice after a 5-hour fast. Basal glucose was determined via a tail nick 30 minutes prior to intraperitoneal (i.p.) administration of glucose (20% solution, 10mL/kg body weight). Subsequent blood glucose values were measured after 0, 15, 30, 60, and 120 minutes with a glucometer (Contour Next EZ, Contour Next). Insulin tolerance tests (ITTs) were performed in 22-week-old mice after a 5-hour fast. After determining basal blood glucose concentrations, mice received i.p. injections of insulin (0.75 IU/kg body weight; Actrapid; Novo Nordisk, New Jersey, USA). Glucose concentrations were measured via tail nick after 15, 30, 45, and 60 minutes.

Histology

Liver tissue was fixed in 10% buffered formaldehyde, embedded in paraffin, and sectioned at 10 μm by the Research Histology Lab through ARUP Laboratories at the Huntsman Cancer Institute, Utah, USA. Sections were stained with hematoxylin and eosin and imaged at 20X with an Axio Scan.Z1 slide scanner maintained by the Cell Imaging Core at the University of Utah. Quantification of lipid droplets was performed using ImageJ software (NIH, http://imagej.nih.gov/ij).

Lipid extraction and quantification

Lipids from human serum were extracted, as described previously (35). Mouse tissue samples were homogenized in 225 μL of lipid internal standard (IS) and 200 μL of cold PBS. Mouse serum samples were prepared with the same volume of IS and phosphate buffered saline (PBS) added to 25 μL of thawed serum. Process blanks were labeled for each tissue, which contained equivalent volumes of IS and PBS, as listed in the samples above. Lipids were extracted with the addition of 750 μL of methyl tert-butyl ether. Serum samples were sonicated for 1 minute and incubated on ice for 15 minutes, with brief vortexing every 5 minutes. Samples were centrifuged for 10 minutes at 15 000 × g, and the organic top layer containing lipid was transferred and evaporated to dryness using a miVac vacuum pump (SP Scientific, Pennsylvania, USA). Samples were resuspended in 100 μL of methanol, vortexed and centrifuged for 5 minutes at 15 000 × g. The supernatant (80 μL) was transferred to a glass vial (5182-0554, 5183-2086; Agilent, California, USA) and (10 μL) pooled quality control and stored at -20°C prior to LC-MS/MS analysis. LC-MS/MS lipid identification and quantification were performed, as previously described in detail (40).

Lipid standards

Sphingolipid standard stock solutions were obtained from Avanti Polar Lipids. The IS for human serum samples were prepared, as described previously (35). Standards for mouse tissue and serum were prepared in methanol with the following lipid species and concentrations: sphingomyelin (d18:1/18:0[d9]) (50 pmol/sample tissue, 100 pmol/sample serum); dihydroceramide (d18:0/18:1[9Z]) (5 pmol/sample tissue, 2.5 pmol/sample serum); glucosylceramide (d18:1/17:0) (50 pmol/sample tissue, 12.5 pmol/sample serum); d7-ceramide (d18:1-d7/16:0) (60 pmol/sample tissue, 12.0 pmol/sample serum); d7-ceramide (d18:1-d7/18:0) (35 pmol/sample tissue, 7.0 pmol/sample serum); d7-ceramide (d18:1-d7/24:0) (150 pmol/sample tissue, 30 pmol/sample serum), d7-ceramide (d18:1-d7/24:1) (312 pmol/sample tissue, 62.4 pmol/sample serum); TAG (15:0_18:1[d7]_15:0) (492 pmol/sample lipid and serum); and DAG (15:0_18:1[d7]) (500 pmol/sample tissue and serum).

Statistical analysis

Human participant characteristics were summarized as the mean ± standard deviation for continuous variables and n (%) for categorical variables, and compared between the Utah CAD study and our analytic dataset using t-tests for continuous variables and Chi-squared tests for categorical variables. Differences in clinical and demographic variables for individuals harboring 0 (TT, n = 382), 1 (TG, n = 163), or both (GG, n = 22) copies of the rs267738 allele were compared with one-way analysis of variance (ANOVA) for continuous variables and Chi-squared tests for categorical variables. We conducted multiple imputation for variables with missing data, which did not change the results, using the Mice package in R.

Human serum lipid concentrations were log₁₀ transformed for analysis, due to nonparametric distribution. Sums and ratios of sphingolipid species were computed prior to log transformation. We conducted multivariable linear regression of lipid concentrations (dependent variable) on additive rs267738 allele count (main predictor), adjusting for age, sex, and body mass index (BMI; continuous variable), and CAD status (yes/no). We tested for effect modification by BMI by including a BMI*rs267738 allele count interaction term and comparing models with and without the interaction term using Wald tests.

To determine individuals’ cardiac event risk test 1 (CERT1) score (0–12), we calculated 16:0, 18:0, and 24:1 ceramide concentrations and their ratios to 24:0 ceramide. For each lipid or ratio, participants were stratified into quartiles and scored, such that individuals received 2 points for variables within the 4^th quartile, 1 point for the 3^rd quartile, and 0 points for the bottom 2 quartiles (29).

Data for mouse lipidomics and tolerance tests were plotted as mean ± standard error of the mean (SEM) and analyzed with 2-tailed t-tests or 2-way ANOVA. All analyses were carried out using Prism (GraphPad Prism, California, USA) and R 4.0.2 (41). Statistical differences were deemed significant at P < 0.05.

Results

In-silico analysis of rs267738 SNP

The rs267738 SNP results in a glutamine-to-alanine substitution at position 115 (E115A) of the CERS2 enzyme. The amino acid switch occurs in a highly conserved locus within the hox-like (HOX) domain proximal to the enzyme’s catalytic Tram-Lab CLN8 (TLC) domain (42). The CERS2 crystal structure has not been solved, although the enzyme is thought to contain several transmembrane domains, primarily residing in the membrane of the endoplasmic reticulum (43). We predicted that the substitution of a medium-sized polar residue with a small nonpolar moiety would elicit deleterious structural or functional consequences. We investigated the potential effects of the rs267738 missense variant with available predictive software designed to estimate functional consequences of amino acid substitutions. The SNP received “damaging” and “possibly damaging” predictive scores from Sorting IntolerantFrom Tolerant (SIFT) and Polyphen-2 (scores 0.04 and 0.863, respectively) (44, 45). Additionally, MutationAssessor predicted that the mutant CERS2 protein would have “medium” functionality (score 2.605) (46).

Generation of a CERS2 E115A mouse line

Since the rs267738 SNP is predicted to impair CERS2 enzyme activity, we were eager to model the functional consequences of the variant. The CERS2 genomic sequence is highly conserved between mammalian species. Mouse and human CERS2 sequences are 92% identical, with perfectly matched sequences surrounding the mutated base pair of rs267738 (amino acids 102–141). Thus, we generated a knock-in mouse harboring the rs267738 mutation with CRISPR/Cas9, as shown in Fig. 2A. Two founders harboring the mutation were crossed onto a C57Bl6/J background to generate a colony of homozygous mutant mice (Cers2^mt/mt; Fig. 2B). Genotypes were confirmed by PCR amplification of a NarI DNA digest or Sanger sequencing (Fig. 2C). All analyses were performed with wildtype C57Bl6/J littermates as controls.

Figure 2.

Generation of a mouse line harboring the E115A mutation. a: CRISPR reagents were designed to introduce the human single-base-pair switch (pink) resulting in the missense mutation within exon 3 of the Cers2 gene. The single-strand oligo DNA nucleotide (ssODN) donor also contained a mutated PAM sequence introducing a unique NarI restriction site (blue). b: Embryos were collected 0.5 days after fertilization, and pronuclei were microinjected with CRISPR constructs. Embryos were implanted into pseudopregnant female mice. Founders were mated to a C57Bl6/J background to generate a mutant colony. c: Sanger sequencing was utilized to confirm the presence of the SNP. Abbreviations: gRNA, guide RNA; RNP, ribonucleoprotein; S, denotes heterozygotes; ssODN, single-stranded donor oligonucleotide; WT, wildtype.

Assessment of CERS2 E115A enzyme activity

CERS2 is the predominant CERS enzyme expressed in the liver (10). To directly investigate the effect of the rs267738 mutation on enzyme function, we compared CERS2 activity in liver tissue of Cers2^mt/mt mice with wildtype controls. Homozygosity for the SNP led to a 42% reduction in liver CERS2 activity (P = 0.0049; Fig. 3A), which was not associated with a change in CERS2 protein (Fig. 3B).

Figure 3.

Cers2 ^mt/mt mice have decreased CERS2 activity and are predisposed to diet-induced glucose intolerance and hepatic steatosis. a–e: Male and female mice (n = 12 per genotype) were placed on a HFD at 15 weeks of age. A GTT was performed at 21 weeks of age. CERS2 activity (a) and protein (b) were measured from liver homogenate. c: Body weight development over 11 weeks of HFD feeding. d: After a 5-hour fast, blood glucose levels were measured at indicated time points following i.p. administration of a 20% glucose solution (t = 0, 10 mL/kg body weight). f: Representative images were captured from fixed liver sections (n = 4–6 mice per genotype) stained with hematoxylin and eosin. g: Lipid droplet area was quantified with ImageJ. Data are expressed as mean ± SEM. *P < 0.05, **P < 0.01.

Abbreviations: AUC, area under the curve; HFD, high-fat diet; i.p., intraperitoneal; GTT, glucose tolerance test; SEM, standard error of the mean.

Metabolic phenotyping of Cers2^mt/mt mice

Previously, we demonstrated that mice heterozygous for Cers2 were indistinguishable from wildtype controls when fed a chow diet but were predisposed to insulin resistance and hepatic steatosis when fed an HFD (20). Therefore, we coupled the genetic insult of the rs267738 mutation with the environmental stressor of an HFD (60% total energy) for the present study. Diet-induced weight gain was similar across genotypes (Fig. 3C). Cers2^mt/mt mice demonstrated impaired glucose tolerance, as assessed with a GTT (Fig. 3D) and increased hepatic steatosis (Fig. 4E and 4F) compared with wildtype controls. We also compared insulin tolerance in these animals (Supplemental Figure 1A) and evaluated the correlation of CERS2 enzymatic activity with glucose tolerance (Supplemental Figure 1B), although neither measure was significant (36).

Figure 4.

Cers2 mutation elicits an increase in C16 sphingolipids in SWAT. Liver (a), serum (b), and SWAT sphingolipid species (c) totaled across dihydroceramide, ceramide, glucosylceramide, dihydroSM, and SM by variable chain length. N = 10 mice per genotype. *P < 0.05, **P < 0.01.

Abbreviations: dihydroSM, dihydrosphingomyelin; SM, sphingomyelin; SWAT, subcutaneous white adipose tissue.

Profiling of Cers2^mt/mt tissue and serum sphingolipids

We quantified mouse serum and tissue sphingolipids with targeted LC-MS/MS lipidomics (Supplemental Figures 2–6) (36), anticipating that a partial loss-of-function in the CERS2 enzyme would translate to decreased VLC sphingolipid synthesis. Cers2^mt/mt mice had increased levels of pathogenic C16 sphingolipids in subcutaneous white adipose tissue (SWAT) (Fig. 4C). This is interesting, as adipose tissue is a major site of ceramide action that shows dramatic changes in CERS6 expression in human and mouse obesity (47). However, we observed no significant decreases in VLC sphingolipids when totaled across dihydroceramide, ceramide, glucosylceramide, dihydrosphingomyelin (dihydroSM), and sphingomyelin sphingomyelin (SM) in liver or serum (Fig. 4A and 4B).

Utah CAD case-control participant characteristics

We next turned to human data to probe for an effect of the rs267738 SNP on sphingolipid acylation and corresponding disease risk. Clinical and demographic data were collected from a subset (n = 567) of patients included in the Utah coronary artery disease (CAD) case-control study (34). Selection of the current study population was determined by available biospecimens; as a result, the current study features individuals with a higher average BMI, higher proportion of males, and higher rates of diabetes, CAD, and hypertension compared with the original Utah CAD cohort (Supplemental Table 1) (36).

Whole-exome sequencing identified individuals with the rs267738 SNP. Clinical and demographic characteristics were compared between individuals harboring 0 (TT, n = 382), 1 (TG, n = 163), or both (GG, n = 22) rs267738 alleles (Table 1). We observed no differences in weight, gender, frequency of diabetes, hypertension, or CAD between genotype groups.

Table 1.

Baseline characteristics of participants from the Utah CAD case-control study

	TT	TG	GG	P-value
	n (%) or mean ± SD
No. of subjects	382	163	22
Male	268 (70%)	110 (67%)	14 (64%)	0.7023
Age (yr)	55.0 ± 7.5	55.2 ± 7.2	52.9 ± 8.9	0.5640
BMI	29.0 ± 5.3	28.7 ± 5.5	28.3 ± 4.7	0.4000
NA	9 (2%)	4 (2%)	2 (9%)
Diabetes
Yes	61 (16%)	22 (13%)	5 (23%)	0.4870
Hypertension
Yes	191 (50%)	69 (42%)	8 (36%)	0.1505
CAD
Yes	265 (69%)	114 (70%)	17 (77%)	0.7343
Total-C (mg/dL)	202.8 ± 43.1	200.5 ± 44.7	213.0 ± 58.8	0.8290
HDL (mg/dL)	42.0 ± 12.5	43.0 ± 12.2	45.9 ± 15.3	0.1520
LDL (mg/dL)	120.3 ± 39.2	118.8 ± 42.1	123.0 ± 57.9	0.9010
VLDL (mg/dL)	39.4 ± 28.6	38.1 ± 27.6	43.2 ± 40.3	0.9890
Serum TG (mg/dL)	192.9 ± 112.0	198.6 ± 147.8	229.3 ± 253.3	0.2720
Glucose (mg/dL)	97.8 ± 36.3	100.3 ± 48.6	99.3 ± 47.5	0.5600
AST (IU/L)	24.5 ± 7.7	26.7 ± 15.1	24.9 ± 8.8	0.0807
NA	2 (<1%)
ALT (IU/L)	29.2 ± 14.8	33.1 ± 19.8	29.2 ± 13.3	0.1150
NA	110 (29%)	43 (26%)	5 (23%)

Clinical characteristics of 567 participants included in the Utah CAD study with 2 copies of the predominant allele (TT, n = 382) or with 1 (TG, n = 382) or 2 (GG, n = 22) copies of the rs267738 allele. Differences between genotypes were assessed using one-way ANOVA for continuous variables and a χ2 test for categorical variables.

Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; CAD, coronary artery disease; HDL, high density lipoprotein; IQR, interquartile range; LDL, low density lipoprotein; NA, data not available; SD, standard deviation; TG, triglyceride; Total-C, total cholesterol; VLDL, very low density lipoprotein; yr, years.

Investigation of human rs267738 serum sphingolipids

We utilized targeted LC-MS/MS to quantify serum sphingolipid species in humans. Lipid species were compared across gentoypes within an additive genetic model (Table 2), which controlled for participants’ age, sex, BMI, and CAD status. We did not observe a significant difference in circulating sphingolipids according to rs267738 SNP allele count when comparing individual lipids or lipids summed by species or variable acyl chain length (Table 2). We also calculated the ceramide CERT1 for each individual, which is a ceramide-based validated measure of cardiometabolic disease risk currently in clinical use (29, 30, 35, 48). As shown in Fig. 5, we detected no differences in CERT1 scores across genotypes (P = 0.356).

Table 2.

Means and interquartile tanges for LC-MS/MS measured sphingolipids in human participants from the Utah CAD study

	TT	TG	GG	P-value
	Median (IQR)
No. of subjects	382	163	22
Ceramide
d18:1/16:0	237.8 (128.2–293.9)	214.8 (114.0–281.0)	202.8 (154.5–266.1)	0.2642
d18:1/18:0	96.7 (42.0–119.2)	84.7 (39.4–107.6)	83.9 (57.4–98.0)	0.3205
d18:1/20:0	75.8 (37.7–87.7)	68.0 (35.0–89.9)	64.6 (43.0–83.2)	0.3222
d18:1/22:0	456.2 (231.3–574.5)	419.2 (212.8–514.0)	396.2 (305.4–488.1)	0.4579
d18:1/24:0	170.5 (90.4–217.6)	157.1 (79.9–195.1)	148.9 (113.1–182.8)	0.3076
d18:1/24:1	480.0 (251.1–580.0)	439.0 (225.7–525.8)	436.9 (313.9–565.1)	0.3348
Dihydroceramide
d18:0/16:0	0.22 (0.12–0.26)	0.20 (0.12–0.27)	0.18 (0.12–0.24)	0.1691
d18:0/18:0	0.16 (0.08–0.19)	0.14 (0.07–0.16)	0.15 (0.07–0.19)	0.1113
d18:0/20:0	0.08 (0.04–0.10)	0.07 (0.04–0.08)	0.07 (0.04–0.11)	0.2497
d18:0/22:0	0.37 (0.15–0.43)	0.35 (0.17–0.40)	0.33 (0.18–0.44)	0.8350
d18:0/24:0	0.82 (0.35–0.98)	0.78 (0.37–0.96)	0.73 (0.37–1.03)	0.6060
d18:0/24:1	0.46 (0.19–0.55)	0.43 (0.20–0.51)	0.44 (0.21–0.62)	0.7313
Glucosylceramide
d18:1/16:0	391.7 (293.9–449.5)	417.7 (299.4–466.4)	375.2 (280.5–416.4)	0.4119
d18:1/18:0	75.7 (49.8–91.4)	77.9 (52.8–91.3)	79.7 (57.1–100.4)	0.3247
d18:1/20:0	97.0 (56.0–119.3)	101.3 (54.9–120.1)	97.0 (68.7–116.5)	0.7999
d18:1/22:0	742.7 (423.2–901.1)	741.1 (411.5–948.7)	604.4 (414.8–709.5)	0.3619
d18:1/24:0	616.2 (369.1–785.6)	630.8 (352.1–759.2)	547.4 (400.3–636.2)	0.6314
d18:1/24:1	617.3 (365.3–778.8)	594.3 (335.9–721.3)	555.1 (374.0–619.7)	0.2429
DihydroSM
d18:0/16:0	68.7 (39.3–80.4)	61.1 (38.6–74.5)	60.0 (43.1–70.7)	0.1495
d18:0/18:0	38.2 (11.6–41.6)	30.6 (10.6–39.9)	29.2 (15.3–41.8)	0.4904
d18:0/20:0	72.1 (20.4–73.7)	54.1 (10.6–66.6)	46.1 (27.4–60.2)	0.2433
d18:0/22:0	17.9 (4.2–16.5)	13.2 (4.1–16.1)	10.5 (5.3–12.3)	0.2030
d18:0/24:0	2.3 (0.7–2.3)	1.8 (0.7–2.2)	1.5 (0.8–2.0)	0.2511
d18:0/24:1	80.3 (19.9–93.8)	57.7 (17.9–64.2)	43.0 (22.6–50.8)	0.1280
SM
d18:1/16:0	853.8 (541.1–1014.6)	767.6 (520.4–943.9)	746.8 (592.9–903.8)	0.0919
d18:1/18:0	211.8 (125.8–247.8)	184.7 (1240-237.7)	190.1 (160.0–208.7)	0.1833
d18:1/20:0	553.2 (144.2–578.4)	410.3 (127.9–497.7)	336.8 (176.0–429.1)	0.2395
d18:1/22:0	1206.7 (283.5–1295.4)	880.3 (250.5–1078.5)	696.5 (384.6–815.0)	0.1807
d18:1/24:0	528.4 (137.9–679.7)	383.5 (114.0–420.6)	287.2 (183.6–348.7)	0.1187
d18:1/24:1	1628.4 (364.1–1701.6)	1137.4 (324.5–1187.3)	967.0 (449.7–1275.4)	0.1415
Sphinganine	0.05 (0.03–0.06)	0.04 (0.03–0.05)	0.04 (0.03–0.05)	0.2700
Sphingosine	0.36 (0.07–0.23)	0.23 (0.07–0.19)	0.15 (0.07–0.20)	0.0166
Sphingolipid totals
Ceramide	1517.2 (795.3–1852.1)	1383.0 (739.0–1712.1)	1333.4 (1054.7–1633.0)	0.3439
Dihydroceramide	2.10 (0.99–2.54)	1.96 (1.05–2.28)	1.89 (1.01–2.67)	0.5394
Glucosylceramide	2540.6 (1653.4–3105.2)	2563.1 (1559.1–3013.4)	2258.8 (1675.0–2477.6)	0.4622
DihydroSM	279.4 (102.9–299.5)	218.6 (97.3–271.0)	190.3 (110.2–230.9)	0.1599
SM	4982.2 (1631.8–5059.2)	3763.8 (1532.2–4309.9)	3224.4 (1973.0–3902.2)	0.1238
Acyl Chain Totals
16:0	1552.2 (1122.3–1774.1)	1461.4 (1041.7–1748.2)	1385.0 (1071.8–1574.7)	0.1100
18:0	422.6 (254.1–495.1)	378.1 (243.3–461.3)	383.1 (313.5–432.8)	0.2825
20:0	798.1 (288.4–838.9)	633.8 (267.8–753.7)	544.7 (324.1–692.1)	0.1975
22:0	2424.0 (1066.7–2700.4)	2054.2 (931.5–2496.7)	1707.8 (1207.4–1928.2)	0.1589
24:0	1318.2 (652.3–1462.1)	1174.0 (574.8–1475.7)	985.7 (698.6–1084.8)	0.1380
24:1	2806.5 (1048.3–2893.9)	2228.8 (969.4–2424.3)	2002.5 (1245.4–2502.1)	0.0925

Data are presented as mean (IQR) for individuals harboring zero (TT, n = 382), 1 (TG, n = 163), or 2 (GG, n = 22) rs278839 alleles. Raw P-values are reported from 2-tailed t-tests of multiple linear regression coefficients according to an additive genetic model adjusted for age, sex, BMI, and CAD. Sphingolipid concentrations were log-transformed prior to analysis due to nonparametric distribution. Sphingolipid species and acyl chains were summed prior to log transformation. After FDR adjustment, all P-values were insignificant. Units are concentration as pmol lipid/mL serum. Abbreviations: IQR, interquartile range; dihydroSM, dihydrosphingomyelin; SM, sphingomyelin; BMI, body mass index; CAD, coronary artery disease; LC-MS/MS, liquid chromatography tandem mass spectrometry; FDR, false discovery rate.

Figure 5.

CERT1 risk score profiles in humans from the Utah CAD study. Violin plots displaying median and IQR for CERT1 scores calculated for individuals with 0 (TT, n = 382), 1 (TG, n = 163), or both (GG, n = 22) rs267738 alleles.

Abbreviations: CAD, coronary artery disease; CERT1, cardiac event risk test; IQR, interquartile range.

Discussion

The purpose of this study was to determine if the rs278839 variant alters CERS2-mediated acylation of sphinganine, and thus contributes to the metabolic pathologies identified in GWAS studies (3–7). Indeed, we are the first to investigate the consequences of the rs267738 SNP in relation to sphingolipid metabolism. We anticipated that the CERS2 E115A substitution would impair enzyme activity, resulting in decreased incorporation of VLC (C20–26) acyl groups onto the sphinganine backbone.

Using the E115A knock-in mouse model, we confirmed a mild inhibitory effect on CERS2 function by directly analyzing enzyme activity from Cers2^mt/mt livers. Our findings revealed that the missense mutation has a modest impact on enzyme activity (down 42%) which was associated with exacerbated diet-induced glucose intolerance and hepatic steatosis in Cers2^mt/mt animals. This is consistent with findings from studies involving Cers2 knockout or knockdown mice, and human GWAS studies, which have identified associations of impaired glycemic control with the rs267738 mutation (3, 20, 31, 33). We attribute this phenotype in our mice to the increase in C16 sphingolipids in Cers2^mt/mt SWAT, as associations between adipose C16 ceramide and metabolic dysfunction have been reported (47, 49).

Contrary to previous studies in mice manipulating Cers2, we did not observe remarkable changes in liver sphingolipids (20, 31, 50), despite Cers2 being highly expressed in hepatic tissue; however, Cers2^mt/mt mice appeared to have increased hepatic lipid deposition compared with controls. Xia et al revealed a “cross-talk” system with liver and adipose ceramides in an inducible acid ceramidase (AC) mouse model (49). Degradation of ceramide in mice challenged with an HFD via either liver- or adipose-specific AC overexpression resulted in sphingolipid equilibration between the 2 tissues, along with decreased hepatic steatosis and improvements in whole-body glucose metabolism. In the setting of a common variant with only partial loss of enzyme function, it may be especially difficult to attribute tissue-specific phenotypes to a particular sphingolipid depot, particularly if the liver, adipose tissue, and plasma share a somewhat common pool of circulating sphingolipid.

The partial loss of CERS2 function was not reflected in serum sphingolipid acylation patterns in humans or mice. We acknowledge that, despite a potential loss in CERS2 activity within the ceramide synthesis pathway, sphingolipid acylation may have remained largely unchanged due to redundancy in CERS substrate specificity for VLC acyl groups (ie, CERS3 and CERS4; Fig. 1). Furthermore, de novo ceramide synthesis may not require complete CERS2 function unless flux through the pathway is significantly upregulated. Nevertheless, our findings raise the potential for alternative mechanisms driving rs267738 associations with diabetes and kidney disease. For example, the rs266738 variant has been associated with altered expression of nonsphingolipid related genes (ie, ANXA9 and CTSS). Furthermore, Sociale et al demonstrated activity of mammalian CERS2 as a transcriptional regulator of lipases, which was dependent on the integrity of a nuclear localization site 2 located at positions 121–124 of the CERS2 protein, just downstream of the E115A substitution (51). Similar to the lipid-sensing transcription factors SREBP and PPAR, CERS2 DNA binding and lipase transcriptional regulation were sensitive to fatty acid levels.

Several limitations to the current study exist that should be considered alongside the findings presented here. First, we acknowledge the difficulty and uncertainty of translating GWAS associations of common variants to causal mechanisms. Typically, common SNPs have mild phenotypes, which may require additional epigenetic or environmental insults to present clinically. As such, our study may be insufficiently powered to explore such nuances with a sample size of 567 individuals and the absence of a replication cohort. Additionally, an alternative SNP rs267734 occurs in a noncoding region of CERS2 and is associated with the progression of albuminuria in diabetic patients (52). We believe that the more likely functional variant is the one we studied, rs267738, as this mutation results in an amino acid substitution. Nevertheless, the two variants are in almost perfect linkage disequilibrium (r² = 0.84); therefore, most of our human subjects harboring the rs267738 mutation likely also contain the rs267734 SNP. Second, due to the origin of our human study sample, our cohort is enriched with individuals suffering from CAD. We highlight our previous work demonstrating that CAD status alone significantly alters circulating sphingolipids (35) but assert that CAD cases were similarly represented in our genotype strata (Table 1) and controlled for as a covariate in our analyses. Moreover, participants were predominantly white and male, thus the findings are not broadly generalizable. Third, CERS2 expression is also especially high in the kidney, and the majority of GWAS studies reporting on rs267738 have noted significant negative correlations between the SNP and glomerular filtration rates (4–7). Unfortunately, our investigations in Cers2^mt/mt knock-in mice did not characterize functional changes in the kidney, such as urinary albumin excretion or estimated glomerular filtration rate.

In summary, we provide the preliminary insight that the missense mutation rs267738 negatively impacts CERS2 catalytic activity. Additionally, our data from knock-in mice parallel a mild phenotype of altered glucose metabolism identified in GWAS associations. Nevertheless, at present we cannot determine whether the effects of rs267738 on CERS2 activity translate to broader changes in sphingolipid acylation and meaningful disease risk. Further investigation is required to rule out alternative mechanisms and to delineate the clinical significance of the mutant allele in larger human cohorts.

Additional Information

Disclosure Summary: SAS is a co-founder and consultant with Centaurus Therapeutics (San Francisco, CA, USA).

Data Availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request. Supplemental data is available at https://doi.org/10.6084/m9.figshare.13876148.v1.