If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Stuart D. Horswell
Footnotes
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Affiliations
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomCancer Research UK, London Research Institute, London, United Kingdom
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Lee G.D. Fryer
Footnotes
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Affiliations
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomCentre for Endocrinology, William Harvey Research Institute, Queen Mary University of London, London, United KingdomCancer Research UK, Cambridge Institute, Cambridge, United Kingdom
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Claire E. Hutchison
Footnotes
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Affiliations
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomCentre for Endocrinology, William Harvey Research Institute, Queen Mary University of London, London, United Kingdom
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Dlear Zindrou
Footnotes
1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
Affiliations
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomBritish Heart Foundation Cardiac Surgery Unit, Imperial College London, London, United Kingdom
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomCancer Research UK, London Research Institute, London, United KingdomCentre for Endocrinology, William Harvey Research Institute, Queen Mary University of London, London, United Kingdom
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomNational Heart and Lung Institute, Imperial College London, London, United Kingdom; and
Medical Research Council, Clinical Sciences Centre, Hammersmith Hospital, London, United KingdomCancer Research UK, London Research Institute, London, United Kingdom
[S] The online version of this article (available at http://www.jlr.org) contains supplementary data in the form of three figures and six tables. 1 S. D. Horswell, L. G. D. Fryer, C. E. Hutchison, and D. Zindrou contributed equally to this work.
The purpose of this study was to determine the core biological processes perturbed in the subcutaneous adipose tissue of familial combined hyperlipidemia (FCHL) patients. Annotation of FCHL and control microarray datasets revealed a distinctive FCHL transcriptome, characterized by gene expression changes regulating five overlapping systems: the cytoskeleton, cell adhesion and extracellular matrix; vesicular trafficking; lipid homeostasis; and cell cycle and apoptosis. Expression values for the cell-cycle inhibitor CDKN2B were increased, replicating data from an independent FCHL cohort. In 3T3-L1 cells, CDKN2B knockdown induced C/EBPα expression and lipid accumulation. The minor allele at SNP site rs1063192 (C) was predicted to create a perfect seed for the human miRNA-323b-5p. A miR-323b-5p mimic significantly reduced endogenous CDKN2B protein levels and the activity of a CDKN2B 3′UTR luciferase reporter carrying the rs1063192 C allele. Although the allele displayed suggestive evidence of association with reduced CDKN2B mRNA in the MuTHER adipose tissue dataset, family studies suggest the association between increased CDKN2B expression and FCHL-lipid abnormalities is driven by factors external to this gene locus. In conclusion, from a comparative annotation analysis of two separate FCHL adipose tissue transcriptomes and a subsequent focus on CDKN2B, we propose that dysfunctional adipogenesis forms an integral part of FCHL pathogenesis.
Familial combined hyperlipidemia (FCHL) is a common, complex genetic disorder characterized by a pernicious lipoprotein profile that markedly increases the risk of premature coronary heart disease (CHD) (
TCF7L2 is associated with high serum triacylglycerol and differentially expressed in adipose tissue in families with familial combined hyperlipidaemia.
). Precise definition, however, is somewhat contentious, reflecting etiological uncertainty. Hence, slightly different inclusion and exclusion criteria are used to diagnose FCHL (
TCF7L2 is associated with high serum triacylglycerol and differentially expressed in adipose tissue in families with familial combined hyperlipidaemia.
). For example, Huertas-Vazquez and colleagues defined their cohort of Mexican patients on the basis of index patients having hypercholesterolemia and/or hypertriglyceridemia, plus premature CHD, and at least one first-degree relative with hyperlipidemia (
TCF7L2 is associated with high serum triacylglycerol and differentially expressed in adipose tissue in families with familial combined hyperlipidaemia.
). Probands also had to have a first-degree relative with hyperlipidemia, and although clinically overt CHD was not an obligatory inclusion criterion, over 50% of the index patients had a personal or family history of premature CHD.
In pursuit of etiological insight, in this study we examined whether specific biological processes are perturbed in the subcutaneous adipose tissue of FCHL patients. In this regard, tantalizing results have emerged from the radiocarbon dating study of Arner et al. (
), who found that subcutaneous adipose tissue of nonobese FCHL patients displayed a lower rate of triglyceride storage than non-FCHL individuals, plus a marked reduction in the apparent rate of adipocyte triglyceride hydrolysis. Thus, Arner and colleagues have speculated that FCHL-blood lipid abnormalities could partially stem from decreased fluxes of dietary and de novo synthesized triglycerides/fatty acids through adipose tissue depots, leading to increased delivery of these lipid moieties to the liver, thereby partly explaining the increased production of VLDL observed in this condition (
Stable isotopes show a direct relation between VLDL apoB overproduction and serum triglyceride levels and indicate a metabolically and biochemically coherent basis for familial combined hyperlipidemia.
). It is also of interest that a systems genetics approach comprising RNA microarray studies of Mexican FCHL fat samples plus a FADS3 genetic association analysis (
) indicates that dysfunctional fatty acid metabolism in adipose tissue depots may form part of the root cause of FCHL. We hypothesized that a microarray analysis of adipose tissue samples from a series of carefully selected groups of white British FCHL patients and controls would serve to home in on a core set of genes that were differentially expressed in FCHL and, more importantly, provide a reasonable launch pad for examining the direct contribution of one of this gene set to the development of adipose tissue and, thus, the risk of developing this highly atherogenic condition.
MATERIALS AND METHODS
Microarray gene expression and data annotation
Adipose tissue was obtained from white British males either at coronary artery bypass surgery (groups designated as FCHL-CHD and non-FCHL-CHD patients) or at heart valve replacement [non-FCHL, non-CHD subjects (i.e., the control subjects)]. FCHL-CHD patients had serum cholesterol and triglyceride levels equal to or greater than 90th age/sex-specific percentile values and CHD equal to or less than 65 years, as described (
). One adopted patient (44 years of age; BMI, 25.7 kg/m2) had fasting serum cholesterol and triglycerides of 8.8 mmol/l and 3.24 mmol/l, respectively. All other FCHL-CHD patients had a family history of dyslipidemia. All CHD patients had taken HMG-CoA reductase inhibitors for over 3 months but no other lipid-lowering medications. No control subject was taking lipid-lowering medications. CHD patients were taking an aspirin, β-blocker, and/or an angiotensin-converting enzyme inhibitor. Exclusion criteria for participants included type 1 and 2 diabetes, hypothyroidism, or raised serum γ-glutamyl transferase and aspartate transaminase levels. The study was approved by the Research Ethics Committee at Hammersmith Hospital, and all participants gave written, informed consent. Fasting lipid levels were determined as described (
To ensure that the biopsies came from the same anatomical site (i.e., subcutaneous upper-abdominal region), the biopsies were obtained by two surgeons. To minimize variability in sample processing and handling, they were collected at one site and immediately snap-frozen in liquid nitrogen in the operating room. All subsequent sample processing was performed by the same individual using the same standardized Affymetrix protocol, which was optimized on comparable samples prior to commencing this study. Biotinylated-labeled cRNA, prepared as described (
), and probe sets were summarized using “median polish.” Probes were labeled “absent” or “present” according to the Affymetrix Microarray Suite 5.0 algorithm. Probes returning absent calls in all cel files were removed. The coefficient of variation across probes was determined using the formula, CoV = σ / I μI, where σ = standard deviation and μ = mean. Differences in probe values were detected by the significance analysis of microarray (SAM) procedure (
). The Ingenuity pathway analysis (IPA; Ingenuity® Systems, www.ingenuity.com) score assigned to the networks represents the likelihood that the number of focus genes (i.e., differentially expressed genes) within the network (maximum 35) are found therein by chance, as assessed by a hypergeometric test, with the score equal to the negative exponent of 10 of the respective P value (e.g., score of 3 equates to P = 0.001).
RT-qPCR
Human tissue cDNAs and RNAs were obtained from Clontech and Ambion, respectively. RNA from cell lines was isolated using RNeasy Plus kits (Qiagen) and Qiashredder columns (Qiagen). RT-qPCR was performed in triplicate (with no RT control) using the Brilliant II SYBR Green QPCR Master Mix (Stratagene) or a Quantifast SYBR Green RT-PCR kit (Qiagen) and Quantitect primer assays (Qiagen). The CDKN2A assay amplifies the p16-encoding (NM_000077.4) and p14-encoding (NM_058195.3) transcripts, both of which were probed for in the Affymetrix microarray experiment. The CDKN2B primer set and microarray assay probes detect transcript 1 (NM_004936) and 2 (NM_078487.2) sequences. Data were analyzed with either SDS 2.3 (Applied Biosystems) or MxPro (Stratagene) software. Threshold cycle (Ct) values were analyzed using the relative 2-ΔΔCt method (
In vitro differentiation of 3T3-L1 murine preadipocytes
Mouse 3T3-L1 cells (Sigma-Aldrich) were cultured in DMEM (Gibco) with 10% fetal calf serum (FCS), 1 mM L-glutamine and 100 u/ml penicillin and streptomycin (Gibco) at 37°C under 5% CO2. Differentiation was induced at 48 h postconfluence (day 0) using medium supplemented with 10 µg/ml insulin, 0.5 mM isobutylmethylxanthine, and 1 µM dexamethasone, as recommended by the supplier. Where specified, differentiation medium was replenished on day 2. Otherwise, differentiation medium was replaced on day 3 with medium containing 10 µg/ml insulin, and then replenished every 48 h. Each independent experiment comprised three technical replicates.
Knockdown of CDKN2A and CDKN2B mRNA was performed with predesigned HP GenomeWide siRNA (Qiagen), using HiPerFect transfection reagent (Qiagen) according to the manufacturer's fast-forward protocol. Single knockdown was performed at final concentration of 5nM siRNA, using two siRNA for each isoform (i.e., Mm_Cdkn2a_1 and 3; Cdkn2b_2 and 4); double knockdown was performed with 5nM of each isoform combination. The scrambled siRNA sequence was 5′-UUCUCCGAACGUGUCACGUdTdt-3′ (Qiagen).
For Oil Red O staining, cells were washed in PBS, fixed with formalin for 30 min, rinsed in water, incubated with 3% Oil Red O in 70% isopropanol for 30 min, and extensively washed in water. Images were acquired using a CCD camera and MetaMorph software (Molecular Devices). For quantification, Oil Red O was extracted in isopropanol. For flow cytometry, cells were detached with TrypLE Express (Gibco) and fixed by slowly adding 1.8 ml (−20°C) 70% ethanol while vortexing. Cells were kept for over 1 h at 4°C, and then centrifuged, resuspended, and washed in PBS. Samples containing 20,000 cells were run on a FACScalibur flow cytometer (BD Biosciences) at between 100 and 500 events/second. Data were analyzed with WinMDI 2.9 software (Scripps Institute; http://facs.scripps.edu). Side scatter versus forward scatter measurements were divided into six regions of equal dimensions along the Y-axis to determine the proportion of cells with increased side scatter and, therefore, increased stored lipid content (
Western blotting was performed on cell lysates harvested in 50 mM Tris/HCl (pH 7.5), 50 mM NaF, 5 mM sodium pyrophosphate, 1 mM DTT, 10% glycerol, and 1% Triton X-100. PVDF membranes were blocked in 5% nonfat milk powder and probed with specified antibodies [p15: rabbit polyclonal to p15 INK4b (Abcam); p16: mouse monoclonal to p16 (Santa Cruz Biotechnology)], followed by an appropriate secondary antibody linked to horseradish peroxide (Bio-Rad). Immunoreactive products were visualized by enhanced chemiluminescence using a 16-bit charge-coupled, device-cooled camera (GeneGnome, Syngene, Cambridge, UK).
eQTL analyses
Tagging single nucleotide polymorphisms (SNP) were identified with the Haploview integrated program Tagger (
). Probes ILMN_1723198 and ILMN_2376723 hybridize to CDKN2B transcript 1 (NM_004936) and to CDKN2B transcript 2 (NM_078487.2). Probe ILMN_1670111 detects only CDKN2B transcript 2 species and spans three SNP sites, precluding its use. Haplotype frequencies were estimated in Haploview (
) using 1000 Genome phase 1 EUR data, downloaded using the VCF to PED converter tool in the 1000 Genomes Browser (www.broswer.1000genomes.org).
miRNA analyses
The putative miRNA recognition elements for miR-138-2-3p and miR-323b-5p on haplotype 1-derived CDKN2B mRNA [i.e., minor allele (C) at SNP site rs1063192 and major allele (G) at rs3217992 (NM_004936)] were identified through fulfilling at least two of the following three criteria: i) a total context score ± -0.25 in TargetScan 6.2 (
); iii) a minimum free energy (MFE) of hybridization with their predicted target sites ≤ −7 kcal/mol plus a miRanda score ≥ 140, as calculated in RegRNA 1.0 (
). miRNAs were isolated from human embryonic kidney 293 (HEK293) cells using a miRNeasy Plus kit, quantified with Quantitect primer assays miScript RT and SYBR green PCR kits (all from Qiagen), and normalized against RNU6B.
HEK293 cells were cultured in DMEM containing 10% heat-inactivated FCS (Biosera), 50 u/ml penicillin, and 50 μg/ml streptomycin (Sigma) at 37°C under 5% CO2. Cells were plated in 24-well plates (Greiner Bio-One) at a density of 1.2 × 105 cells per well and transfections were performed at final concentrations of 5nM miRNA or scrambled control (Qiagen) using Hiperfect transfection reagent (Qiagen) according to the manufacturer's fast forward protocol. Cells were harvested in RIPA buffer (Sigma-Aldrich) containing protease inhibitors (Complete Protease Inhibitor Cocktail Tablets, Roche). Equal amounts of protein were separated on 4-12% gradient NuPage Bis-Tris precast gels (Life Technologies). Immunoreactive products were quantified with dye-labeled secondary anti-bodies (LI-COR Biosciences) using the Odyssey imaging system (LI-COR). Molecular masses were estimated using Novex Sharp Protein Standards (Life Technologies).
Luciferase reporter assays
Nucleotides c.*1835 to c.*2948 of CDKN2B cDNA (NM_004936) were inserted into pmirGlo (Promega) downstream of the firefly luciferase open reading frame using standard techniques. All constructs were validated by sequencing. Cells were grown to ∼80-90% confluence and plated at 5 × 104 cells per well in 96-well plates (Corning). Cells were transfected with 50 μl Optimem I (Invitrogen) containing 150 ng plasmid DNA, 0.5 μl Attractene (Qiagen), and the specified miRNA (Qiagen) at a concentration of 5 nM, unless otherwise stated, according to the manufacturer's fast-forward protocol. Luciferase activities were assayed 22 h post transfection using a Dual-Glo Luciferase Assay kit (Promega). Firefly luciferase activity was normalized to the corresponding Renilla luciferase activity. Each independent experiment included six transfections per combination of miRNA and reporter construct.
). The diagnosis required probands to have a relative with primary hyperlipidemia. Exclusion criteria for probands and family members included non-Northern Europeans, other forms of hyperlipidemia (e.g., familial hypercholesterolemia) based on molecular diagnosis, standard clinical signs, or diagnostic criteria. SNPs rs3217992 and rs1063192 were genotyped using the Illumina Inc. Genotyping Service (San Diego, CA) and a Taqman SNP Genotyping Assay 7900HT Fast Real-Time PCR machine (Applied Biosystems), respectively. Genotypes calls were validated through use of duplicate samples and by Sanger sequencing. The quantitative trait analysis invoked linear mixed-effects models implemented in the package GWAF (
) for genome-wide association analyses with family data in R (version 2.11.1). The function “lmekin” was employed to account for within-family phenotype correlations, and age, sex, and BMI were included as covariates. Additive and recessive models were specified. Triglyceride levels were log-transformed.
Statistics
Statistical analyses were performed using unpaired Student t-test or one-way or two-way ANOVA, followed by Sidak's multiple comparison test or posttest for linear trend, where indicated, using GraphPad Prism 6.
RESULTS
Overrepresentation of cell-cycle genes in FCHL
Table 1 summarizes the study participants' clinical data. The FCHL-CHD patients (n = 13) had higher serum cholesterol levels than non-FCHL-CHD patients (n = 6), both before and after commencement of statin-based medications. Microarray analysis returned expression values for 27,487 probe sets. The coefficient of variation was less than 10% for 96.4% of the probes across all individuals, indicative of high-quality data (
). Moreover, a heatmap representation of the entire dataset demonstrates the approximate uniformity of the data across study participants (supplementary Fig. I), compliant with the standard assumptions of SAM (
). At a 1% FDR, 190 probe sets were identified as differentially expressed in CHD patients versus the control subjects. This number increased to 311 when only FCHL-CHD patients were compared with non-FCHL, non-CHD controls (supplementary Table I). By contrast, even after relaxing the FDR to 5%, only 29 probes were differentially expressed in the six non-FCHL-CHD patients relative to the controls (supplementary Table II). Fifteen of these were also differentially expressed in the FCHL-CHD patient samples (supplementary Table I) and were therefore removed from all subsequent analyses, reducing the “FCHL-CHD” probe set to 296 (i.e., 311 − 15), representing 284 genes (supplementary Table I). Of these, 56 (19.7%) displayed concordant differential expression in the adipose tissue of Mexican FCHL patients (Table 2).
b Functional categories were determined through an independent critique of the literature (supplementary Table I).
c Highlights genes differentially expressed in the abdominal subcutaneous adipose tissue of obese (BMI 33.5 ± 6.0 kg/m2), “normolipidemic” (total cholesterol 4.24 ± 1.13 mmol/l, triglyceride 1.50 ± 0.39 mmol/l) men compared with lean (BMI 23.9 ± 3.1 kg/m2), “normolipiemic' (total cholesterol 3.47 ± 0.64 mmol/l, triglyceride 0.89 mmol/l) controls (18).
The FCHL-CHD patients appeared to comprise two distinct subgroups: four patients whose serum triglycerides were unaltered or elevated (median 3.44 mmol/l versus 3.47 mmol/l) on statin-based medications and seven who had greater than 25% reductions (median 1.40 mmol/l versus 3.40 mmol/l) on such medications. Their mean serum cholesterol and BMI were, respectively, 5.95 ± 0.95 mmol/l and 26.83 ± 1.46 kg/m2 and 5.21 ± 1.06 mmol/l and 28.10 ± 1.67 kg/m2. Because of the triglyceride difference between the two FCHL subgroups, we analyzed their expression data separately, each with respect to the controls. At a 1% FDR, SAM returned an additional 105 differentially expressed probes, representing 103 genes that had not been identified when all 13 FCHL-CHD patients were analyzed together (supplementary Table I); of these, 17 (16.5%) displayed concordant differential expression in the Mexican FCHL samples (Table 2 and supplementary Table I). Thus, in total, we identified 387 genes displaying differential expression in FCHL-CHD adipose tissue samples, and of these, 73 (18.9%) were also concordantly differentially expressed in the periumbilical subcutaneous fat of Mexican patients (
) (supplementary Table I and Table 2). Collectively, these data suggest that the subcutaneous adipose tissue transcriptome of our male FCHL-CHD patients share more in common with that of Mexican FCHL-patients than obese, normolipidemic men.
To extract biological insights from the FCHL-CHD gene expression data, we first examined the differentially expressed gene lists for enrichment of gene ontology (GO) terms using DAVID (
). Overall, the analysis returned rather modest P values (Table 3). However, seven of the biological processes characterizing the FCHL-CHD dataset displayed stronger P values in the Mexican FCHL dataset (Table 3); these were not enriched in either the non-FCHL-CHD or the obese, normolipidemic datasets (supplementary Table III). Furthermore, IPA identified 20 highly significant networks in the FCHL-CHD dataset (Table 4). The highest ranking network (lipid metabolism, small molecule biochemistry, vitamin and mineral metabolism) comprised 27 focus genes (i.e., differentially expressed in FCHL-CHD adipose tissue samples at a FDR of 1%); specifically, the lipid-associated genes CYB5R3, LPL, PLIN, and VLDLR (
); MTHD1; and 21 genes tagged with one or more of the GO terms returned by the DAVID cluster analysis (supplementary Table IV). Similar numbers of such tagged GO genes were also present in IPA networks 2–4; e.g., 23 in network 2 and 18 in network 4 (supplementary Table IV).
TABLE 3Biological processes enriched in adipose tissue of the British FCHL-CHD differentially expressed gene list and comparison with Mexican-FCHL dataset
TABLE 4Ingenuity pathway analysis of differentially expressed genes (n = 387) in adipose tissue of FCHL-CHD patients
TABLE 4Ingenuity pathway analysis of differentially expressed genes (n = 387) in adipose tissue of FCHL-CHD patients
Bolding indicates genes differentially expressed in both white British and Mexican FCHL patients. Red indicates upregulated; green indicates downregulated.
The top five IPA networks contained 26 focus genes (e.g., CDKN2B, ITSN2, OPTN, COL4A1, LAMA4, CRYAB) that displayed concordant differential expression in both the British- and Mexican-FCHL fat samples (TABLE 2, TABLE 3, TABLE 4), and that, based on the functions of the genes within these networks, contribute to five broad cellular processes affecting i) the composition of the extracellular matrix, cell-cell contacts, and the adipocyte cytoskeleton; ii) vesicular trafficking pathways; iii) lipid-associated activities; iv) cell-cycle progression; and v) apoptosis. (See supplementary Table IV and the literature cited in supplementary Table I.)
Effect of CDKN2 isoforms on adipogenesis
We elected to follow up the differentially expressed gene CDKN2B because its mRNA levels were increased in the adipose tissue samples of white British and Mexican FCHL patients (Table 2) and because the IPA placed CDKN2B in the highest-ranking network (Table 4). Additionally, quiescent 3T3-L1 preadipocytes synthesize more Cdkn2 (both types a and b) than 3T3-L1 cells induced to differentiate (
), suggesting that the raised levels of CDKN2B mRNA in the subcutaneous adipose tissue samples of FCHL patients (Table 2) may be causally related to their lower rate of triglyceride storage in this fat depot (
As judged by RT-qPCR, human adipose tissue samples contained over 10-fold more CDKN2B mRNA than CDKN2A (supplementary Fig. II-A). Moreover, FCHL-CHD patient samples tended to contain more CDKN2B mRNA than the control samples (supplementary Fig. II-B), supporting the microarray finding of increased CDKN2B mRNA in the adipose tissues of both white British and Mexican FCHL patients (Table 2). In comparison, transcript levels for CDKN2A, the adipogenic marker C/EBPα (
Signaling pathways through which insulin regulates CCAAT/enhancer binding protein alpha (C/EBPalpha) phosphorylation and gene expression in 3T3-L1 adipocytes. Correlation with GLUT4 gene expression.
), and two control genes were not increased
(supplementary Fig. II, C–F). CDKN2B expression was also higher than that of CDKN2A in the lung; while the amount of CDKN2B mRNA in the liver was relatively low (supplementary Figs. II-A and III).
To determine the role of CDKN2 in 3T3-L1 preadipocye differentiation, the standard mitogenic and adipogenic cocktail was used (
). Induction of adipogenesis was confirmed by marked rises in C/EBPα mRNA at days 2 and 4 (Fig. 1A, left panel), following the addition of the differentiation media at day 0 and of fresh differentiation media on day 2. In comparison, both CDKN2A (Fig. 1A, middle panel) and CDKN2B (Fig. 1A, right panel) mRNA levels were significantly decreased at the days 1, 2, and 3 time points. At day 6, CDKN2A mRNA levels were still reduced, while CDKN2B mRNA had returned to uninduced levels. By day 9, ∼10% of induced cells contained lipid droplets (Fig. 1B), and Oil Red O staining was significantly higher in the induced cultures than the noninduced cells (Fig. 1C).
Fig. 1Contributions of CDKN2A and CDKN2B to differentiation of 3T3-L1 cells. (A) Mouse 3T3-L1 cells were grown to confluence and induced to differentiate (day 0) into adipocytes or maintained as a confluent monolayer in growth media (noninduced). In this experiment, induced cells were given fresh differentiation media on day 2. Transcript levels were assayed by RT-qPCR. Values (mean ± SEM) relative to noninduced are from three independent experiments. (B, C) 3T3-L1 cells were stained with Oil Red O at days 0 (48 h post confluence) and 9. Representative microscopic images are at 20× magnification, with the inset on day 9 at 40× magnification. The white arrows highlight lipid-droplets (B). Oil Red O values (mean ± SEM) are from three independent experiments (C). (D, E) Induced cells were transfected with specified siRNA at days −3, −1, 2, 5, 8. CDKN2A (D) and CDKN2B (E) mRNA levels were assayed by RT-qPCR. Values (mean ± SEM) relative to scrambled at day −1 are from three independent experiments. (F) Western blot of cell lysates from noninduced 3T3-L1 cells (lane 1) and 3T3-L1 cells transfected with the indicated siRNA on days −3 and −1 were analyzed at day 2. (G) C/EBPα mRNA in 3T3-L1 cells transfected with specified siRNAs on days −3, −1, 2, 5, 8, and 10, and induced with the mitogenic and adipogenic cocktail on day 0. Values (relative to scrambled at day −2) are from three independent experiments. (H) Individual Oil Red O staining from two independent experiments. Results are expressed as the ratio of staining in induced cells relative to noninduced cells at day 9. (I) Ratios of side scatter measurements from induced versus noninduced 3T3-L1 cells on day 9 indicate cell granularity (i.e., stored lipid content).
Isoform-specific CDKN2 siRNAs were used to define the effects of CDKN2A/B expression on C/EBPα mRNA levels and adipogenesis. These were administered to the 3T3-L1 cells on days −3, −1, 2, 5, and 8 and, where specified, on day 10. Relative to the scrambled control, the CDKN2A siRNAs significantly decreased CDKN2A mRNA on days 0 and 3 (Fig. 1D). The CDKN2B siRNAs also decreased CDKN2B mRNA on these days and on days −1 and 1 (Fig. 1E). Additionally, the siRNAs reduced CDKN2A and B protein (Fig. 1F).
CDKN2A and B knockdown increased mRNA levels of C/EBPα (Fig. 1G), an early adipogenic marker (
). Specifically, CDKN2A knockdown was associated with increased C/EBPα mRNA levels on days 1, 2, 4, and 6 (Fig. 1G, first panel), while CDKN2B knockdown led to higher C/EBPα mRNA for a rather shorter timeframe (Fig. 1G, middle panel); by day 6, levels had returned to the scrambled control values. Similar to CDKN2A knockdown, combined CDKN2A and B knockdown was associated with significantly higher C/EBPα mRNA levels at day 6 (Fig. 1G, third panel).
Corroborating the C/EBPα findings (Fig. 1G), CDKN2 knockdown increased the lipid contents of 3T3-L1 cells (Fig. 1H, I). Relative to un-induced cells, combined CDKN2A and CDKN2B knockdown increased Oil Red O staining by an average of 2.09-fold versus 1.38-fold for the control scrambled siRNA (Fig. 1H); in a separate experiment, the number of cells attaining high cytoplasmic granularity (a good marker of stored cellular fat (
)) was increased [Fig. 1I, regions (R) 3–6 versus R1 and 2]. Individual CDKN2A and CDKN2B knockdown also increased Oil Red O staining by ∼2-fold (Fig. 1H), consistent with the results of the FACS analysis [i.e., higher proportion of the scrambled siRNA control cells were retained in R1 than in R4–6 compared with the CDKN2 knockdown cells (Fig. 1I)]. Collectively, these data suggest that the enrollment of preadipocytes into an adipogenic differentiation program(s) (
) is enhanced by the lowering of CDKN2A and CDKN2B expression.
CDKN2B eQTLs in adipose tissue samples
In view of the transcriptome (Table 2 and supplementary Fig. II) and in vitro (Fig. 1) data, we checked the MuTHER dataset (856 samples) for association between adipose tissue CDKN2B mRNA levels and the six SNPs at the CDKN2B locus that capture its genetic diversity: rs2069422 and rs2069426 (CDKN2B intron 1), rs1063192 and rs3217992, which reside at nucleotide 2619bp and 2,763bp of CDKN2B‘s 3′UTR, plus rs3218018 and rs2811712 residing in CDKN2B/CDKN2A‘s intergenic region.
With two independent reporters, the minor allele (G) at SNP site rs1063192 displayed association with lower adipose tissue CDKN2B mRNA levels, albeit with rather modest P values (Table 5). Additionally, one reporter returned a P value of 0.0012 for association of the minor (T) allele at SNP site rs3217992 with higher CDKN2B mRNA levels (Table 5). By contrast, P values for the tagging SNPs rs2069422, rs2069426, rs3218018, and rs2811712 were all greater than 0.01 (Table 5); in skin and lymphoblast cell line samples, neither rs1063192 nor rs3217992 returned P values less than 0.01.
TABLE 5cis-Acting CDKN2B eQTL analysis in MuTHER adipose tissue samples
The alleles at SNP site rs1063192 and rs3217992 define four distinct haplotypes. The first comprises the minor allele at rs1063192 (G) and the major allele at rs3217992 (C), and it has an estimated frequency of 0.430 in Europeans. The second, which has a frequency of ∼0.399, bears the major allele at SNP site rs1063192 and the minor allele at site rs3217992. In comparison, haplotype 3 (major alleles at both loci) has an estimated frequency of 0.169, and the fourth, carrying both minor alleles, 0.002.
Functional analysis of CDKN2B's haplotype 1- and 2-derived transcripts
Because the rs1063192 and rs3217992 SNPs reside in the 3′UTR of CDKN2B transcripts and do not display strong linkage disequilibrium (LD) (R2 ≥ 0.8) with any other known CDKN2B SNP (supplementary Table V), we searched for miRNAs with potential to target CDKN2B's 3′UTR, as described in Materials and Methods. The analyses predicted binding between miR-323b-5p and haplotype 1-derived CDKN2B sequences spanning SNP site rs1063192, with perfect complementarity in the seed sequence (Fig. 2A, left). This perfection is lost in haplotype 2-derived CDKN2B mRNA (c.*2619C>T), reducing the estimated MFE of miR-323b-5p hybridization to this putative miRNA binding site to −25 kcal/mol from −27.7 kcal/mol. No other putative miR-323b-5p binding site was detected in CDKN2B‘s 3′ UTR. The analysis also identified miR-138-2-3p as having the potential to bind to haplotype 1-derived CDNK2B mRNA sequences encompassing SNP site rs3217992, again with perfect complementarity in the seed region (Fig. 2A, right). The MFE values for the major allele (i.e., G in mRNA, haplotype 1) and minor (A) allele were −28.1 and −22.2 kcal/mol, respectively. A second putative recognition element for miR-138-2-3p was identified (nucleotides 2549−2556 of CDKN2B‘s 3′UTR), which had a seed length of eight nucleotides and estimated MFE of hybridization of −16.9 kcal/mol.
Fig. 2Minor allele (C) of rs1063192 creates a functional miRNA recognition element for miR-323b-5p in the human CDKN2B 3′UTR. (A) Schematic representation of interactions between haplotype 1-derived CDKN2B mRNA and specified miRNAs. Seed sequences are highlighted in gray, with the minor and major alleles at SNP sites rs1063192 and rs3217992 shown in bold. (B) Relative levels of miR-323b-5p and miR-138-2-3p across indicated human pooled tissue samples and HEK293 cells quantified by RT-qPCR. Expressions values (mean ± SEM), relative to thyroid, are from three quantifications of pooled RNA samples from the stated human tissues and three independent HEK293 cell cultures. (C) HEK293 cells were transfected at a final concentration with 5 nM of specified miRNAs. CDKN2A and B mRNA were quantified by RT-qPCR. Values (mean ± SEM) relative to scrambled (scr) siRNA are from three independent experiments. (D) CDKN2A and B protein levels in HEK293 cells transfected with specified miRNAs. Values (mean ± SEM) relative to scrambled siRNA are from five independent experiments. (E–G) Effects of miR-138-2-3p and miR-323b-5p mimics on luciferase reporter activities in HEK293 cell lysates. Values (mean ± SEM) are from five independent experiments. (H) Effect of combining miR-138-2-3p and miR-323b-5p mimics on luciferase reporter activities in HEK293 cell lysates. Values (mean ± SEM) are from five independent experiments.
We examined miR-323b-5p and miR-138-2-3p transcript levels in a panel of human tissues plus HEK293 cells, which are heterozygous at both rs1063192 and rs3217992 SNP sites. Comparable amounts of miR-323b-5p were present in adipose tissue and liver (Fig. 2B), consistent with a global gene expression dataset (
) is expressed in multiple tissues, including fat. Transfection of HEK293 cells with a miR-323b-5p mimic had no significant impact on CDKN2B mRNA levels (Fig. 2C) but did reduce CDKN2B protein (Fig. 2D). By contrast, the miR-138-2-3p mimic produced a more marked impact on CDKN2B mRNA than protein (Fig. 2C, D). Thus, these data indicated that the 3′ UTR of CDKN2B may contain functional miRNA recognition elements for miR-323b-5p and miR-138-2-3p.
To validate that CDKN2B's 3′ UTR is a direct target of miR-323b-5p and miR-138-2-3p, we examined the activities of two reporter constructs in a dual-luciferase reporter assay. The first construct contained CDKN2B haplotype 1-derived 3′ UTR sequences [i.e., rs1063192 minor allele (C); rs3217992 major allele (G)] and the second, haplotype 2-derived [i.e., rs1063192 (T); rs3217992 (A)] sequences. Compared with the control scrambled mimic, the miR-323b-5p mimic significantly reduced the relative luciferase activity produced from haplotype 1-derived CDKN2B 3′ UTR sequences (P = 0.0056, Fig. 2E); whereas no such reduction was observed from the construct containing haplotype 2 sequences (P = 0.97). Furthermore, the haplotype 1 construct responded in a dose-dependent fashion (Fig. 2F). Thus, these data suggested that human CDKN2B transcripts harboring the C allele at SNP site rs1063192 are a direct target of miR-323b-5p. Moreover, combined with the data reported here (Fig. 2C, D, and Table 5) and elsewhere (
), translational repression of CDKN2B mRNA may precede its deadyenylation and decay.
The miR-138-2-3p mimic produced a small, but nonsignificant (P = 0.27) reduction in relative luciferase activity produced from the haplotype 1 construct, and an even smaller reduction from the haplotype 2 construct (Fig. 2G). Combining the miR-138-2-3p mimic with that of miR-323b-5p (2.5 nM each) reduced luciferase activity produced by the haplotype 1 construct by 17.8 ± 1.7% compared with 5.2 ± 1.6% from the haplotype 2 construct (Fig. 2H).
Finally, in view of the increased CDKN2B mRNA levels in FCHL adipose tissue (Table 2), the CDKN2B eQTL data (Table 5) and the two sets of in vitro findings (Fig. 1, Fig. 2), we felt obliged to examine for association of rs1063192 and rs3217992 with plasma triglyceride levels in our cohort of 244 FCHL families, which comprised 1,729 family members. Under a continuous trait, additive model, both SNPs returned poor P values (supplementary Table VI); while under a recessive model, the rare rs1063192 allele produced a P of 0.0029 for association with lower, not higher, plasma triglyceride. Thus, these data indicated that although raised CDKN2B mRNA levels in the adipose tissue of FCHL patients may contribute to the FCHL-triglyceride trait, factors other than alleles at SNP sites rs1063192 and rs3217992 drive this relationship.
DISCUSSION
The primary aim of this study was to identity the core biological processes perturbed in the subcutaneous adipose tissue of FCHL patients and to gain etiological insight into the triglyceride storage defect (
). An important discovery is that the adipose tissue transcriptome of FCHL patients is characterized by gene expression changes that would be expected to impair adipose tissue development and/or renewal. One specific change was the increased level of CDKN2B mRNA, replicating data from a separate FCHL cohort (
). We also found that CDKN2B knockdown in 3T3-L1 preadipocyes increased adipogenesis and that CDKN2B mRNA is targeted by miR-323b-5p, which in common with CDKN2B is expressed in multiple tissues, including fat. Thus, through the combination of results from a comparative annotation analysis of two separate FCHL adipose tissue transcriptomes, subsequent CDKN2B studies, and the radiocarbon-dating study of Arner et al. (
), we propose that dysfunctional adipogenesis forms an integral part of FCHL.
Our genome-wide microarray experiment was performed on a relatively small number of samples, which stemmed from the stringent selection and exclusion criteria employed: participants were required to be male, of North European ancestry, and undergoing coronary artery bypass surgery. Moreover, we excluded all participants with hyperlipidemia potentially attributable to a high-fat diet, obesity, an alcohol intake greater than 14 units/week or a medical condition known to raise blood lipid levels. This stringency may have precluded the identification of many genes differentially expressed in the upper abdominal subcutaneous fat of FCHL patients at modest-moderate levels, especially as we implemented a stringent FDR threshold (1%) to minimize the risk of identifying false positives. Nonetheless, it is well established that microarray studies using even smaller patient cohorts than that used in the current study have successfully identified important gene expression changes and the nature of the dysfunctional pathways underlying disease (
Identification of acquired somatic mutations in the gene encoding chromatin-remodeling factor ATRX in the alpha-thalassemia myelodysplasia syndrome (ATMDS).
). Indeed, our approach was vindicated in that we identified 73 differentially expressed FCHL genes that also exhibited altered expression in the subcutaneous periumbilical fat of rather younger (38 ± 9.3 years of age), leaner (27 ± 2.4 kg/m2), and healthier Mexican FCHL patients who had somewhat higher total cholesterol (6.8 ± 1.2 mmol/l) and triglyceride (4.7 ± 2.6 mmol/l) levels (
) than our patients (Table 1), presumably because they were not on treatment with statins. Nevertheless, this gene overlap (∼20%) with the Mexican FCHL study is rather modest, which very likely relates to a number of differences between the two studies, including our smaller sample size, different inclusion criteria (e.g., men only), removal of gene expression changes related to FCHL pathogenesis rather than etiology, potential metabolic differences between upper abdominal (current study) and periumbilical (Mexican) fat (
Synthesis of the microarray data annotated here identifies cytoskeletal and cell adhesion dysregulation as a defining feature of FCHL adipose tissue (Tables 2–4). Does such dysfunction drive the cell biological etiology of FCHL, rather than its pathogenesis and what is the evidence to suggest that it might? Noteworthy signals include those for integrin β5 (ITGB5), α-actinin (ACTN1), α-B crystallin (CRYAB), and collagen IVα1 (COL4A1), all of which were increased in the British- and Mexican-FCHL adipose tissue samples (Table 2). Integrin β5, which directly interacts with α-actinin (
), suggest that FCHL-lipid abnormalities may develop, at least in part, from dysfunction of the complex developmental pathway that produces mature adipocytes.
Interpreting the clinical significance of the increased CDKN2B mRNA levels in the subcutaneous fat of FCHL patients (Table 2) is facilitated by the studies reported here and elsewhere (
). Thus, CDKN2B-deficient mouse embryonic fibroblasts (MEF) cultured under standard 3T3 conditions exhibit a ∼2-fold higher proliferation rate than wild-type MEFs and following serum deprivation, reenter the S phase of the cell cycle more efficiently (
). Here, we show that adding a mitogenic/adipogenic cocktail to 3T3-L1 fibroblasts induced comparable drops in CDKN2B and CDKN2A (its neighbor) mRNA levels (Fig. 1A) and that relative to a scrambled control, siRNA-mediated silencing of either CDKN2A or CDKN2B mRNA increased adipogenesis (Fig. 1D–I). We also found that CDKN2B mRNA levels are much higher in human adipose tissue samples than are CDKN2A levels, implying that although either CDKN2 isoform can promote cell cycle arrest in G1 (
), this role is largely served by CDKN2B in human adipocyte precursors.
The observed increases in the adipogenic marker C/EBPα and cellular lipid content in our 3T3-L1 cells may appear relatively small. However, measurements were carried out in a mixed population of cells, in which at the time point investigated, only ∼10% of the cells had undergone differentiation. When cells were left longer, a higher proportion differentiated (results not shown), but we elected to study an early time point, as the changes in the initiation of adipogenesis resulting from decreased CDNK2 expression were expected to be more evident, pronounced, and physiologically relevant.
Germaine to the effects of miR-323b-5p and miR-138-2-3p mimics on CDKN2B mRNA and protein and on CDKN2B reporters, interesting data have emerged from a genome-wide study regarding the fate (i.e., decay versus translational inhibition) of mRNAs targeted by miR-155 (
). The group of degraded targets was enriched for those that had a seed length of eight nucleotides compared with the group of translationally inhibited targets, and the group of translationally repressed transcripts was enriched for targets that had just one putative binding site for this mRNA (74% compared with 30% in the group of degraded targets). Here, we provide evidence that the activity of miR-323b-5p toward its predicted target CDNK2B is primarily one of translational inhibition (Fig. 2D, E) rather than mRNA degradation (Fig. 2C); note that its 3′ UTR contains just one putative recognition element for this mRNA and that the seed match is seven nucleotides in length (Fig. 2A, haplotype 1-derived transcripts). By contrast, the activity of miR-138-2-3p, which is predicted to target CDNK2B mRNA through two recognition sites, both containing an eight nucleotide seed match (Fig. 2A, haplotype 1-derived sequence), is more consistent with it targeting CDKN2B mRNA for degradation (Fig. 2C, G). In summary, we are confident that haplotype 1-derived CDKN2B transcripts are targeted by miR-323b-5p, and we suggest that mRNA decay secondary to translation repression (
) may underlie the weak association between decreased CDKN2B mRNA levels in the MuTHER dataset and the rare allele at SNP site rs1063192.
Previous studies have made good progress in understanding the relationship between CDKN2B expression and that of the long, noncoding RNA called antisense non-coding RNA in the INK4 (i.e., CDKN2A/B) locus (ANRIL). For instance, it is known that ANRIL is expressed at orders of magnitude lower than CDKN2B (
). Here, we have considered the association between ANRIL-associated alleles and CDKN2B mRNA levels in human adipose tissue samples (supplementary Table V), while recognizing that cis-acting effects explain a smaller proportion of the overall variance in CDKN2B mRNA than ANRIL transcript levels in the human tissue samples examined to date (
) had an effect on adipose tissue CDKN2B mRNA levels comparable to the rare allele at the CDKN2B SNP site rs1063192 (supplementary Table V and Table 5). Moreover, all alleles in LD with this 3′UTR allele displayed less robust association with adipose tissue CDKN2B mRNA levels than rs1063192 (supplementary Table V). Thus, while we recognize that an ANRIL-associated allele(s) in LD with the rare allele at SNP site rs1063192 may increase ANRIL transcript levels and in turn mediate polycomb-mediated repression of CDKN2B expression, previous data (
) and the findings reported here (Fig. 2) indicate that, in adipose tissue, miR-323b-5p contributes directly to the activity of haplotype 1-derived CDKN2B mRNA.
With regard to the finding of increased CDKN2B mRNA in the adipose tissue samples of FCHL patients (Table 2), first it is important to appreciate that suppressing the commitment of precursor cells to the adipogenic lineage can raise fasting serum triglyceride (
). Second, the consequence of impaired adipogenesis and/or dysfunctional adipocyte death on serum triglyceride levels is likely to be exacerbated in FCHL patients because of the increased production of VLDL that occurs in this condition (
Stable isotopes show a direct relation between VLDL apoB overproduction and serum triglyceride levels and indicate a metabolically and biochemically coherent basis for familial combined hyperlipidemia.
)] and hence that determinants of CDKN2B activity may contribute to the processes regulating cell death, as well as adipogenesis. However, despite these reasons for believing that increased adipose tissue CDKN2B expression could contribute directly to the development of the FCHL-triglyceride trait, the results from our association analysis (supplementary Table VI) do not support the hypothesis that CDKN2B's 3′UTR variants (rs1063192 and rs3217992) contribute to the etiology of the FCHL-triglyceride trait, consistent with the results from genome-wide association studies, which failed to identify significant lipid associations with the CDKN2B/ANRIL locus in the population at large (
). Our results underscore the importance of future studies to establish the identities of the trans-acting factors and signaling molecules that help maintain the high level of CDKN2B expression in the adipose tissue of FCHL patients (and quiescent adipocytes) and to clarify how these relate to the gene expression changes that characterize the FCHL adipose tissue transcriptome and in turn the development of FCHL-lipid abnormalities.
Acknowledgments
The authors are indebted to all study participants, to Dr. Andrew F. Dean for helpful discussion, and to Ms. Sophie C. Dean for help in preparing the manuscript.
TCF7L2 is associated with high serum triacylglycerol and differentially expressed in adipose tissue in families with familial combined hyperlipidaemia.
Stable isotopes show a direct relation between VLDL apoB overproduction and serum triglyceride levels and indicate a metabolically and biochemically coherent basis for familial combined hyperlipidemia.
Signaling pathways through which insulin regulates CCAAT/enhancer binding protein alpha (C/EBPalpha) phosphorylation and gene expression in 3T3-L1 adipocytes. Correlation with GLUT4 gene expression.
Identification of acquired somatic mutations in the gene encoding chromatin-remodeling factor ATRX in the alpha-thalassemia myelodysplasia syndrome (ATMDS).
This work was supported by the British Medical Research Council and by an unrestricted educational research grant from Merck Sharp & Dohme / Schering Plough.
35700-X&id=gr1.jpg){kind=link}
35700-X&id=gr2.jpg){kind=link}