1. Valerie Z Wall

Abstract

Individuals who are born small for gestational age (SGA) have a risk to develop various metabolic diseases during their life course. The biological memory of the prenatal state of growth restricted individuals may be reflected in epigenetic alterations in stem cell populations. Mesenchymal stem cells (MSCs) from the Wharton's jelly of umbilical cord tissue are multipotent, and we generated primary umbilical cord MSC isolates from SGA and normal neonates, which were subsequently differentiated into adipocytes. We established chromatin state maps for histone marks H3K27 acetylation and H3K27 trimethylation and tested whether enrichment of these marks was associated with gene expression changes. After validating gene expression levels for 10 significant chromatin immunoprecipitation sequencing candidate genes, we selected acyl-coenzyme A synthetase 1 (ACSL1) for further investigations due to its key roles in lipid metabolism. The ACSL1 gene was found to be highly associated with histone acetylation in adipocytes differentiated from MSCs with SGA background. In SGA-derived adipocytes, the ACSL1 expression level was also found to be associated with increased lipid loading as well as higher insulin sensitivity. ACSL1 depletion led to changes in expression of candidate genes such as proinflammatory chemokines and down-regulated both, the amount of cellular lipids and glucose uptake. Increased ACSL1, as well as modulated downstream candidate gene expression, may reflect the obese state, as detected in mice fed a high-fat diet. In summary, we believe that ACSL1 is a programmable mediator of insulin sensitivity and cellular lipid content and adipocytes differentiated from Wharton's jelly MSCs recapitulate important physiological characteristics of SGA individuals.

The risk of metabolic syndrome comprising obesity, cardiovascular disease and type 2 diabetes mellitus is enhanced by aberrant developmental fetal programming (1, 2). This encompasses a range of cellular and physiological mechanisms, which can influence tissue function depending on the in utero environment (3). There are multiple pathways to the developmental induction of an increased metabolic risk, but less than optimal fetal nutrition has been regarded as one of the most important environmental cues (4) leading to birth of small for gestational age (SGA) children (5). Fetal growth restriction with subsequent rapid postnatal catch-up growth has been strongly associated with insulin resistance and metabolic syndrome (6, 7). Several tissues contribute to fetal programming of metabolic disease involving central (8) and peripheral organs (9, 10). Adipose tissue has been perceived to be one of the major regulators of insulin sensitivity via the secretion of adipocyte-specific hormones and inflammatory cytokines (11). Subtle cues may induce a longer term fetal response, which is believed to be mediated by epigenetic pathways (12). It is therefore proposed that among other tissues, the risk for the development of metabolic diseases is programmed into the offspring's adipocytes by induced changes in the epigenetic landscape, which in turn may reflect effects on precursor stem cell populations (13, 14).

Why does such a large variation in adaptation exist among athletes that undergo the exact same training program? Why do some improve more than others? Coach Craig Pickering explores this age-old question, focusing primarily on the variation’s biological causes.

It was shown that mesenchymal stem cell (MSC)-derived adipocytes in amniotic fluid from obese women show a higher adipogenic potential (15). This suggests that there may be an existing pool of multipotent stem cells that are developmentally programmed to support the appearance of an obesogenic trait at a later stage in life. To evaluate this general hypothesis, we previously established umbilical cord (UC) MSC isolates from SGA and normal neonates and showed increased insulin sensitivity, as well as higher cell proliferation rates for the growth restricted group in the basal nondifferentiated state (16). Infants who are born growth restricted and insulin sensitive are thought to become insulin resistant in childhood (17).

In the current study, we aimed to shed light on adipocyte-specific developmental programming of metabolic disease risk. Our main hypothesis has been that specific molecular pathways predicting a trajectory towards metabolic compromise such as obesity are present and detectable in differentiated MSCs taken from SGA individuals, but not from control subjects. We further speculate that epigenetic mechanisms are involved in fetal programming of metabolic functions. In order to address these hypotheses, we have differentiated UC MSCs from SGA and control neonates into mature adipocytes and analyzed the global pattern of histone modifications regulating gene activation and repression events. One of the most significant changes in histone acetylation was observed in genomic regions of acyl-coenzyme A (CoA) synthetase 1 (ACSL1), a gene involved in fatty acid metabolism. The ACSL1 gene was significantly enriched with acetylated histones during the process of adipogenesis in adipocytes differentiated from MSCs taken from the SGA group. We functionally characterized ACSL1 in the context of adipogenesis in the in vitro MSC adipogenesis system. In addition, to position ACSL1 into a broader in vivo context of obesity, we further evaluated its expression pattern in a rodent model of nutritionally induced obesity. We further identified genes whose expression was significantly altered upon ACSL1 depletion and speculate that ACSL1 is a classical obesogenic gene, whose expression associates with high-caloric food intake and is programmable during early fetal development.

Materials and Methods

Clinical populations and sample collection

Fresh UCs taken from neonates born at the National University Hospital, Singapore, were collected under the umbrella of Internal Review Board approval by the National University Health System. Written parental consent for participation in this study was obtained and ethical approval was granted by the Domain Specific Review Board of the National University Hospital.

Assessment of fetal growth characteristics

We recently described the assessment of fetal growth for the subjects in our study in more detail (16). Briefly, SGA was diagnosed by ultrasonography as growth between the 5th and 10th percentile compared with a normal reference population. Scans were done in a standard manner by trained ultrasonographers, using ultrasound machines (Aloka SSD-4000; GE Voluson E8).

Preparation and propagation of MSCs from human UC

The preparation of MSCs is described elsewhere (16, 18). The primary cell lines were passaged up to 10 times.

Adipocyte differentiation of MSCs

The differentiation of MSCs followed standard procedures with slight modifications listed below. Briefly, cells were plated into 6 cm dishes coated with 0.2% gelatin and grown in growth medium (DMEM high glucose with 2mM L-glutamine and 20% fetal bovine serum (FBS), antibiotic/antimycotic mixture) supplemented with 1× Insulin-Transferrin-Sodium selenite, 16-ng/mL bovine Fibroblast Growth Factor (bFGF) and 0.1 mmol/L β-mercaptoethanol (Invitrogen). After 1 week after confluence growth (labeled as cycle 0), the cells were stimulated by culturing them in adipocyte induction medium (DMEM high glucose with 2 mmol/L L-glutamine, 10% FBS; HyClone), supplemented with 0.5mM isobutylmethylxanthine (Sigma I-7018), 1μM dexamethasone (Sigma D-4902), 10μM insulin (bovine; Sigma I-5500), 200μM indomethacine (Sigma I-8280), and 1μM rosiglitazone (Sigma). After 3 days, the adipocyte induction media was replaced with insulin medium (DMEM high glucose with 2 mmol/L L-glutamine, 10% FBS, supplemented with 10μM insulin and 1μM rosiglitazone). These cycles of adipocyte induction were repeated up to 5–6 times, and mature adipocytes were finally maintained in insulin medium.

Oil Red O staining

Quantification of adipogenesis was carried out by staining lipid globules with the hydrophobic dye Oil Red O. Induced adipogenic cultures in 6 cm dishes were fixed with 4 mL of 10% formaldehyde for 20 minutes at room temperature. A stock solution of Oil Red O was prepared by dissolving 0.5-g powder in 100-mL isopropanol. The Oil Red O solution was added to the fixed cells, and dishes were incubated for 1 hour at room temperature. Afterwards, the staining solution was removed, and the dishes were washed twice with distilled water.

Chromatin immunoprecipitation sequencing (ChIP-seq) of histone H3K27 acetylation (H3K27ac) and H3K27 trimethylation (H3K27me3) marks followed by bioinformatics analysis

Approximately 5 million cells from each primary MSC line were cross-linked with 1% formaldehyde for 15 minutes and excess formaldehyde quenched by addition of glycine (0.625M). Nuclei were prepared and collected under standard procedures. Protein-DNA complexes were immunoprecipitated using 3 μg of H3K27acetyl antibody (Actif Motif 39133) or 3 μg of H3K27me3 antibody (Millipore 07-449). Library preparation was done as described (19), followed by multiplexed 36-bp sequencing on an Illumina HIseq.

For mapping and peak calling, reads were mapped to the genome using BWA (20), and peaks were called using MACS (21) and GC content normalization using the Homer ChIP-seq suite (22). Browser tracks: Wiggle files were generated using DFilter in wiggle mode (23). The average log2 intensities of gene expression microarray signals were quantile normalized and signals were then split into groups of genes from the lower quartile (<25%) upper quartile (>75%), and median (25%–75%). Reads of the corresponding ChIP-seq profile (H3K27ac and H3K27me3) were counted 4 kb around the start site of each gene (22) and the average normalized read count per 100-bp bin was plotted. To perform the Pearson correlation clustering, normalized reads were counted in the union of all ChIP-seq peaks (either H3K27ac or H3K27me3) for each library and log2 transformed (22). Up-regulated peaks were defined as those that showed an SGA/control fold change of at least 1.5 at cycle 0 and were also at least 2-fold increased at cycles 3 and 6. Down-regulated peaks were defined by applying the same thresholds to control/SGA fold change. All ChIP-seq data were deposited into the Gene Expression Omnibus repository (GSE64697).

Quantitative ChIP-PCR analysis

For the data shown in Figure 3 below, ACSL1 promoter-specific primers were used to analyze the H3K27ac ChIP enrichment using the ABI 7900 system (PerkinElmer).

Discovery and validation of candidate genes. A, Box plots of acetylation peak heights. Left panel, Peaks that are increased in SGA over control, based on the cut-off (cycle 0 > 1.5-fold, cycle 3 > 2-fold, and cycle 6 > 2-fold; red box plots). Right panel, Peaks that are decreased in SGA (cycle 0 < 1.5-fold, cycle 3 < 2-fold, and cycle 6 < 2-fold). B, UCSC genome browser view of ACSL1 and TFAP2A loci for H3K27ac and methylation ChIP-seq data from MSC-01 (SGA; red) and MSC-44 (control; blue). C, H3K27ac enrichment at the ACSL1 promoter was measured by ChIP-qPCR at cycles 0, 3, and 6 of the adipocyte differentiation process in cell isolates from MSC-60 (control) and MSC-70 (SGA). Results are representative of 3 independent experiments. Values are shown as mean ± SE of the fold enrichment against total input DNA control. A two-way ANOVA indicates the significant difference in time points (P = .0426) and test groups (SGA vs control, P = .0402) for the promoter-specific association of H3K27ac between biological replicates followed by a Student's t test (*, P < .05).

Discovery and validation of candidate genes. A, Box plots of acetylation peak heights. Left panel, Peaks that are increased in SGA over control, based on the cut-off (cycle 0 > 1.5-fold, cycle 3 > 2-fold, and cycle 6 > 2-fold; red box plots). Right panel, Peaks that are decreased in SGA (cycle 0 < 1.5-fold, cycle 3 < 2-fold, and cycle 6 < 2-fold). B, UCSC genome browser view of ACSL1 and TFAP2A loci for H3K27ac and methylation ChIP-seq data from MSC-01 (SGA; red) and MSC-44 (control; blue). C, H3K27ac enrichment at the ACSL1 promoter was measured by ChIP-qPCR at cycles 0, 3, and 6 of the adipocyte differentiation process in cell isolates from MSC-60 (control) and MSC-70 (SGA). Results are representative of 3 independent experiments. Values are shown as mean ± SE of the fold enrichment against total input DNA control. A two-way ANOVA indicates the significant difference in time points (P = .0426) and test groups (SGA vs control, P = .0402) for the promoter-specific association of H3K27ac between biological replicates followed by a Student's t test (*, P < .05).

RNA extraction

MSCs were pelleted in 1.7-mL tubes and then resuspended in 0.5 mL of TRIzol Reagent (Life Technologies 15596-026). All samples were immediately snap frozen on dry ice and kept at −80°C until further use. RNA was prepared and purified using the RNeasy Mini kit (QIAGEN 74106) following the manufacturer's instructions. RNA integrity was checked using Agilent 2100 Bioanalyzer and RNA 6000 Nano Labchips (Agilent Technologies 5067-1511).

Quantitative real-time PCR

For quantitative reverse transcription polymerase chain reaction, total RNA (4 μg) was reverse transcribed using a High Capacity cDNA Reverse Transcription kit (Applied Biosystems, Inc). All cDNA samples were subjected to real-time PCR analysis (in triplicates) with gene-specific primers using an ABI 7900 HT Sequence Detection System (Applied Biosystems, Inc). Target gene expression levels were normalized to 3 endogenous control genes (RPL18, RPL19, and HSP90AB1) and presented relative to the controls.

In vitro glucose uptake assay

The uptake of glucose was measured with the QuantiChrom Glucose Assay kit (Medibena BioAssay Systems) following the manufactures instructions. For this purpose, the glucose concentration was determined in the tissue culture medium by collecting 50 μL from each sample after each adipogenesis induction cycle. The concentration of glucose was determined by directly employing a proprietary o-Toluidine method using a specific colorimetric reaction. Absorbance was measured at 630 nm and was directly proportional to the glucose concentration in the medium as validated via a standard curve against fixed glucose concentrations. The glucose concentration in the adipogenesis induction medium was 4.5-g/L D-glucose. Sample values were first normalized against total protein content (Bradford assay; Bio-Rad 500-0205) of respective samples and then glucose consumption for each sample was calculated relative to control cycle 0 samples.

Quantification of triglyceride content

Total soluble lipids were extracted using the buffer provided with the commercial Adipogenesis Detection kit (Abcam ab102513) and samples were kept at −80°C until further analysis. The amount of total triglycerides was measured using a colorimetric assay as per the manufacturer's instructions.

Western blotting

Differentiated adipocytes were harvested in radioimmunoprecipication assay buffer (Cell Signaling 9806). The protein concentration was determined using the Bradford assay (Bio-Rad 500-0205). Membranes were probed with primary antibodies specific to ACSL1 (Cell Signaling 4047S), ERK1/2, (Cell Signaling 4696S), phospho-ERK1/2 (Cell Signaling 4370), p38 MAPK (Cell Signaling 8690S), phospho-p38 MAPK (Cell Signaling 9216S), phosphatidylinositide (PI) 3 kinase (Cell Signaling 4255), phospho-PI3 kinase (Cell Signaling 4228S), peroxysome proliferator activated receptor γ (PPARγ) (Santa Cruz Biotechnology, Inc c7196), CCAAT/enhancer binding protein α (C/EBPα) (Santa Cruz Biotechnology, Inc c61), protein kinase B (AKT) (Santa Cruz Biotechnology, Inc sc8312), phosphorylated AKT (Ser473) (Santa Cruz Biotechnology, Inc sc7985R), fatty acid binding protein 4 (FABP4) (Abcam ab66682), and the endogenous control protein β-actin (Sigma A1978). The LICOR system was used for signal detection (secondary antibodies 926-32211 and 926-68020). Densitometric band analysis was carried out via Odyssey 2.1.

ACSL1 siRNA knockdown studies

The ACSL1 gene was knocked down during each cycle of the adipocyte induction process. Briefly, we used ON-TARGETplus Human ACSL1 small interfering RNA (siRNA) (Dharmacon L-011654-00-0050) or ON-TARGETplus Nontargeting Pool siRNA (Dharmacon D-001810-10-50) at a final concentration of 50nM. Transfections were conducted using Lipofectamine RNAiMAX (Invitrogen 13778150).

Gene expression microarray

HumanHT-12 v4 Expression BeadChips (Illumina BD-103-0204) with 47231 transcript probes were used for gene expression analysis. Briefly, 500 ng of total RNA was used using the Illumina TotalPrep-96 RNA Amplification kit (Life Technologies 4393543) according to the manufacturer's instructions. The arrays were scanned on an Illumina iScan and data extracted by the Illumina GenomeBeadStudio Software for further analysis. Background subtraction data was exported, and probes were filtered for the determination of the P values of less than 0.05 and number of beads more than 3 using the R program. Data were then loaded into Arraystudio (Omicsoft) and all expression values were log2 transformed. Data were normalized using quantile normalization and all samples with Median Absolute Deviation scores of less than −5 were removed from the analysis. Data were subjected to principal component analysis (PCA) and unsupervised hierarchical clustering (Euclidean distance, ward link). Differentially expressed genes comparing ACSL1 knockdown and control groups were subjected to a de novo network analysis using GeneGo MetaCore. All gene expression microarray data were deposited into the GEO repository (comparison between SGA and control groups, GSE64699; ACSL1 siRNA depletion study, GSE64698).

Animal studies

All animal procedures were performed according to the Institutional Animal Care and Use Committee (no. 130829) of the Agency for Science, Technology and Research. All mice were male, of the C57BL/6J genetic background and housed in a 12-hour light, 12-hour dark cycle with access ad libitum to water and a normal chow diet (NCD) (Harlan 2018 Teklab Global 18% Protein Rodent Diet). For the diet-induced obesity study, mice were fed diets with 60% of the calories from fat (D12492; Research Diets) from 8–17 weeks of age, as indicated in the text. The fat and lean mass of each mouse in this study were measured weekly by an EchoMRI-100 body composition analyzer (Echo Medical Systems). Histological analysis was performed as previously described (24). Briefly, epididymal fat tissue was fixed in 10% neutral formal in buffer, embedded in paraffin, and sectioned at 4 μm. Tissue sections were processed for deparaffinization and rehydration. Sections were further stained with hematoxylin and counterstained with eosin (both reagents from Fisher Scientific). Sections were then dehydrated, mounted with Cytoseal (Thermo Scientific), and examined using a Leica CTR6500 microscope. For the oral glucose tolerance test, all mice were fasted overnight. After baseline glucose values were measured using Accu-Check Performa (Roche), the mice were given glucose at 2-mg/g body weight by oral gavage. Subsequently, the clearance of plasma glucose was monitored at 15, 30, 60, 90, and 120 minutes after glucose administration.

Statistics

Values presented are expressed as mean ± SEM from separated experiments. Statistical analysis was performed using SPSS version 16.0 (IBM, SPSS Statistics). All normal comparisons between groups were assessed using a Student's t test. Experiments involving group comparisons and time courses such as the multicycle adipogenesis differentiation assays were analyzed by employing repeated measures ANOVA, which probes for potential interactions between “time” and “SGA vs control” on the outcomes tested for. In case of a statistically significant interaction, differences in gene expression between SGA and control group were tested at each time point. P < .05 was considered significant. For the animal studies, a two-way ANOVA was employed with 4 mice per each group of NCD and high-fat diet (HFD) using the Graph Prism software.

Results

Establishment of MSC cultures and adipocyte differentiation from SGA and normal neonates

We previously reported that primary human MSC lines established from growth restricted neonates showed increase in both, proliferation rates, and insulin sensitivity (16). Here, we expanded the assessment of biological characteristics of these cells by differentiating them into adipocytes comparing MSCs derived from 10 SGA individuals and 7 matching controls (Supplemental Table 1), which were differentiated over 6 cycles of a standard adipocyte induction treatment. Oil droplets appeared from cycle 2 onwards and a representative experiment is shown in Figure 1. Consistent with the morphological changes at macroscopic (Figure 1A) and microscopic (Figure 1B) levels, the expression of the adipocyte-specific protein FABP4 was significantly induced at later stages of adipogenesis (Figure 1C).

Adipogenic differentiation of human UC-derived MSCs. A, Triglyceride accumulation was detected by Oil Red O staining of MSC-01 cells at different stages of the differentiation process, and culture dishes were photographed. B, Pre- and mature adipocytes were stained with Oil Red O, visualized under a light microscope, and photographed (magnification, ×400). C, The expression of the adipose-specific gene FABP4 was significantly induced throughout the adipocyte differentiation process of MSCs as determined by Western blotting.

Adipogenic differentiation of human UC-derived MSCs. A, Triglyceride accumulation was detected by Oil Red O staining of MSC-01 cells at different stages of the differentiation process, and culture dishes were photographed. B, Pre- and mature adipocytes were stained with Oil Red O, visualized under a light microscope, and photographed (magnification, ×400). C, The expression of the adipose-specific gene FABP4 was significantly induced throughout the adipocyte differentiation process of MSCs as determined by Western blotting.

Histone ChIP-seq studies revealed significantly different enrichment of H3K27ac sites in human MSC adipocytes from SGA background

Adipogenesis is governed by epigenetic processes, which may be perturbed at least experimentally in the context of fetal growth restriction (25). We conducted ChIP linked to massive parallel sequencing (ChIP-seq) using antibodies directed against histone H3K27ac and histone H3K27me3, which are known gene activating and repressive marks, respectively (26, 27). We analyzed preadipocytes (cycle 0), intermediate (cycle 3), and mature differentiated adipocytes (cycle 6) from 2 representative cell lines of SGA (MSC-01) and control group neonates (MSC-44). On average, we generated 11–20 million uniquely mapped reads per sample for H3K27ac and 32–54 million reads for H3K27me3 (Supplemental Table 2). First, key markers of adipogenesis and MSCs were validated (adiponectin receptor [ADIPOR] and thymocyte antigen 1 [THY1], respectively) (Figure 2A). Histone acetylation surrounding the THY1 promoter and nearby enhancers was diminished at cycles 3 and 6, suggesting a progressive down-regulation of this marker. The converse was true for ADIPOR, which showed increasing promoter and enhancer acetylation in cycle 3 and 6. Similarly, the adipocyte marker gene FABP4 showed a steady increase in promoter acetylation during differentiation (Figure 2B), in agreement with our finding that protein levels are up-regulated (Figure 1C).

Epigenetic analysis of SGA and control adipocyte differentiation. A, UCSC genome browser view of SGA (red) and control (blue) H3K27ac and H3K27me3 ChIP-seq signals around THY1 and ADIPOR gene loci. Shown are wiggle tracks of normalized read counts. The arrow displays the gene start, length, and direction. B, UCSC browser snapshot of the FABP4 gene locus (red, SGA; blue, control) with the promoter peak being highlighted. C, PCA. Scatterplots of PC1 and PC2 for SGA (red) and control (black) for H3K27ac and H3K27me3. D, Average distribution of H3K27ac signals relative to the center expressed genes divided in to lower quartile (<25%; red), upper quartile (>75; blue), and median (25%–75%; green). On the x-axis, 0 indicates the TSS start site of the expressed genes. The y-axis displays the sequence depth normalized read densities. E, Genomic annotation of histone H3K27ac and histone H3K27me3 peaks (pie charts, upper panel). The union of peaks from all 6 ChIP-seq samples was associated to the corresponding genomic annotation (RefSeq genes). The bar plot (lower panel) shows the percentage of read counts found in each annotation for each of the samples.

Epigenetic analysis of SGA and control adipocyte differentiation. A, UCSC genome browser view of SGA (red) and control (blue) H3K27ac and H3K27me3 ChIP-seq signals around THY1 and ADIPOR gene loci. Shown are wiggle tracks of normalized read counts. The arrow displays the gene start, length, and direction. B, UCSC browser snapshot of the FABP4 gene locus (red, SGA; blue, control) with the promoter peak being highlighted. C, PCA. Scatterplots of PC1 and PC2 for SGA (red) and control (black) for H3K27ac and H3K27me3. D, Average distribution of H3K27ac signals relative to the center expressed genes divided in to lower quartile (<25%; red), upper quartile (>75; blue), and median (25%–75%; green). On the x-axis, 0 indicates the TSS start site of the expressed genes. The y-axis displays the sequence depth normalized read densities. E, Genomic annotation of histone H3K27ac and histone H3K27me3 peaks (pie charts, upper panel). The union of peaks from all 6 ChIP-seq samples was associated to the corresponding genomic annotation (RefSeq genes). The bar plot (lower panel) shows the percentage of read counts found in each annotation for each of the samples.

PCA of ChIP-seq peak heights revealed that adipocyte differentiation stage and fetal growth status were both associated with global shifts in histone modification profiles (Figure 2C). In the case of H3K27ac, the first principal component (x-axis) aligned with differentiation, whereas the second principal component (y-axis) separated SGA and control samples. Next, we performed a gene expression microarray analysis at the same differentiation stages and compared the expression data to ChIP-seq signal intensity at the corresponding transcription start sites (Figure 2D). Expression levels at cycle 0 were binned into 3 groups (upper quartile, lower quartile, and the medium 25%–75%). H3K27ac reads at the corresponding transcription start sites were quantified and in both cell lines, the pattern of acetylation correlated well with gene expression. Our data showed that the vast majority of acetylation and methylation peaks were located in intergenic or intronic regions, reflecting their status as distal regulatory elements (Figure 2E).

To specifically uncover SGA-associated epigenetic perturbations, we evaluated regulatory regions that were differentially acetylated between SGA and control cell lines, and showed either increased (Figure 3A, left panel) or decreased fold change during adipogenesis (Figure 3A, right panel). H3K27ac peaks detected were mapped to their closest flanking genes (Supplemental Table 3). One of the SGA up-regulated peaks was located at the 3′ end of ACSL1. The ACSL1 locus also contained 2 additional histone acetylation peaks that did not fit the original criterion for up-regulation but were nevertheless more strongly acetylated in SGA at cycles 0 and 3 (Figure 3B, left panel). Shown in comparison is the locus for TFAP2A as an example based on consistently lower histone acetylation during adipocyte differentiation, both at the promoter and at an intronic regulatory element (Figure 3B, right panel). We confirmed enriched H3K27ac at the ACSL1 promoter in one additional primary cell line pair (control MSC-60 and SGA MSC-70) via chromatin immunoprecipitation-quantitative PCR (ChIP-qPCR) (Figure 3C). In addition, we tested for regions with differential H3K27me3 in SGA and control cell lines (Supplemental Figure 1; full list provided in Supplemental Table 3). For the current study, we decided to focus on the characterization of the ACSL1 gene, a key mediator of fatty acid metabolism, because we hypothesized that its aberrant expression in the context of fetal growth restriction may be a contributing factor to the metabolic phenotypes observed during the life course of SGA individuals.

ACSL1 expression is significantly increased in adipocytes differentiated from SGA MSCs concomitant with enhanced glucose uptake and increased total lipid content

We studied the expression pattern of ACSL1 and found that it was highly consistent with the ChIP-seq datasets (Figure 4A for cell lines MSC-01 and MSC-44). Expression analysis in the remaining MSC lines (Supplemental Table 1) confirmed up-regulation of ACSL1 in the SGA group on both, mRNA and protein levels (Figure 4, B–D). ACSL1 expression levels were consistently increased during the adipogenesis process with a maximum expression in the final differentiated adipocyte state (cycle 6). We further found that the basal glucose uptake rate was increased in preadipocytes from SGA neonates. The magnitude of this difference was relatively higher at later cycles of adipogenesis (Figure 4E). In addition, we observed that mature adipocytes differentiated from SGA derived MSCs had a significantly higher content of total triglycerides (Figure 4F).

ACSL1 is overexpressed in mature adipocytes from SGA derived MSCs. A, Quantitative PCR experiments showing higher mRNA expression of ACSL1 in adipocytes from MSC-01 (SGA) compared with MSC-44 (control). Results are representative of 3 independent experiments. Values are shown as mean ± SE for ACSL1 expression relative to that of endogenous control genes. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 8.9E-16) and test groups (P = 6.1E-12) for gene expression between biological replicates followed by a Student's t test (*, P < .05; **, P < .01; ***, P < .001). B, ACSL1 gene expression in the whole panel of SGA and control cell isolates. Quantitative PCR analysis was performed in duplicate RNA samples isolated from differentiated adipocytes at cycle 0, 3, and 6. Values are shown as mean ± SE for ACSL1 expression relative to endogenous control genes. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 2E-16) and test groups (P = 9.6E-05) for gene expression between cell isolates followed by a Student's t test (*, P < .05). C, ACSL1 protein expression was significantly higher in mature adipocytes from SGA-derived adipocytes. D, Densitometry analysis was performed for quantification of the Western blotting results of cycle 6 shown in C. Values were normalized against endogenous control β-actin expression. The P value was determined using a Student's t test. E, Glucose uptake assay of adipocytes from SGA vs control panel cell isolates. SGA and control MSC lines (see C) were induced for adipocyte differentiation and glucose uptake measured as described. Values are shown as mean ± SE for glucose consumption relative to control cycle 0 samples. A two-way ANOVA was performed to detect significant difference in effect of time points (P = .1970) and test groups (P = 6.89E-04) over glucose consumption between 2 groups followed by a Student's t test (*, P < .05; **, P < .01). F, The amount of total triglycerides in duplicate samples of each cell sample was analyzed, and values (mean ± SE) were normalized against total protein content of cells. A two-way ANOVA was performed to show significant difference in effect of time (P = 9.49E-06) points and between groups of cell isolates (P = .141) over total triglyceride content between the 2 groups followed by a Student's t test (**, P < .01).

ACSL1 is overexpressed in mature adipocytes from SGA derived MSCs. A, Quantitative PCR experiments showing higher mRNA expression of ACSL1 in adipocytes from MSC-01 (SGA) compared with MSC-44 (control). Results are representative of 3 independent experiments. Values are shown as mean ± SE for ACSL1 expression relative to that of endogenous control genes. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 8.9E-16) and test groups (P = 6.1E-12) for gene expression between biological replicates followed by a Student's t test (*, P < .05; **, P < .01; ***, P < .001). B, ACSL1 gene expression in the whole panel of SGA and control cell isolates. Quantitative PCR analysis was performed in duplicate RNA samples isolated from differentiated adipocytes at cycle 0, 3, and 6. Values are shown as mean ± SE for ACSL1 expression relative to endogenous control genes. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 2E-16) and test groups (P = 9.6E-05) for gene expression between cell isolates followed by a Student's t test (*, P < .05). C, ACSL1 protein expression was significantly higher in mature adipocytes from SGA-derived adipocytes. D, Densitometry analysis was performed for quantification of the Western blotting results of cycle 6 shown in C. Values were normalized against endogenous control β-actin expression. The P value was determined using a Student's t test. E, Glucose uptake assay of adipocytes from SGA vs control panel cell isolates. SGA and control MSC lines (see C) were induced for adipocyte differentiation and glucose uptake measured as described. Values are shown as mean ± SE for glucose consumption relative to control cycle 0 samples. A two-way ANOVA was performed to detect significant difference in effect of time points (P = .1970) and test groups (P = 6.89E-04) over glucose consumption between 2 groups followed by a Student's t test (*, P < .05; **, P < .01). F, The amount of total triglycerides in duplicate samples of each cell sample was analyzed, and values (mean ± SE) were normalized against total protein content of cells. A two-way ANOVA was performed to show significant difference in effect of time (P = 9.49E-06) points and between groups of cell isolates (P = .141) over total triglyceride content between the 2 groups followed by a Student's t test (**, P < .01).

ACSL1 deficiency impedes lipid loading and insulin sensitivity in adipocytes differentiated from MSCs

ACSL1-specific and control siRNAs were transfected into MSCs at all cycles of adipogenic differentiation and knockdown efficiency was monitored on both, the mRNA and protein levels (Figure 5A, left and right panel, respectively). Upon transfection of ACSL1-specific siRNAs, markers for insulin sensitivity such as phosphorylated AKT, PI3 kinase, ERK1/2, and MAPK were found to be down-regulated (Figure 5B), concomitant with a decrease in glucose uptake (Figure 5C).

ACSL1 deficiency reduces glucose uptake in differentiated adipocytes. A, After siRNA mediated knockdown, ACSL1 mRNA expression was analyzed by RT-PCR, and relative expression values are shown (left panel). Protein expression was assayed by Western blotting and densitometry analysis (shown is a representative result from 3 independent experiments, right panel). ACSL1 RT-PCR expression data are plotted as mean ± SE of gene expression relative to that of endogenous control genes. P value significance was demonstrated by a Student's t test (*, P < .05; **, P < .01). B, Protein expression of key regulators involved in insulin signaling in mature MSC-01 differentiated adipocytes treated with either negative control or ACSL1-specific siRNAs. The data are representative for 2 independent experiments. C, Inhibition of glucose uptake in MSC-01-derived adipocytes transfected with ACSL1-specific siRNAs. Values are shown as fold change (mean ± SE; n = 3) for glucose consumption in each group relative to cycle 0. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 2.2E-11) and test groups (P = 1.1394E-09) over glucose consumption between 2 groups followed by a Student's t test (*, P < .05; **, P < .01).

ACSL1 deficiency reduces glucose uptake in differentiated adipocytes. A, After siRNA mediated knockdown, ACSL1 mRNA expression was analyzed by RT-PCR, and relative expression values are shown (left panel). Protein expression was assayed by Western blotting and densitometry analysis (shown is a representative result from 3 independent experiments, right panel). ACSL1 RT-PCR expression data are plotted as mean ± SE of gene expression relative to that of endogenous control genes. P value significance was demonstrated by a Student's t test (*, P < .05; **, P < .01). B, Protein expression of key regulators involved in insulin signaling in mature MSC-01 differentiated adipocytes treated with either negative control or ACSL1-specific siRNAs. The data are representative for 2 independent experiments. C, Inhibition of glucose uptake in MSC-01-derived adipocytes transfected with ACSL1-specific siRNAs. Values are shown as fold change (mean ± SE; n = 3) for glucose consumption in each group relative to cycle 0. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 2.2E-11) and test groups (P = 1.1394E-09) over glucose consumption between 2 groups followed by a Student's t test (*, P < .05; **, P < .01).

The knockdown of ACSL1 led to a significant decrease in total lipid (Figure 6A) and triglyceride contents (Figure 6B). This reduction of lipids was not due to changes in the adipocyte differentiation program per se, as the ACSL1 knockdown did not alter the expression of key modulators of adipogenesis such as PPARγ, C/EBPα, and FABP4 (Figure 6C).

ACSL1 deficiency reduces total lipid accumulation without affecting the adipogenesis differentiation process. A, Total lipid content of cells was detected by Oil Red O staining at different cycles. Values (mean ± SE; n = 3) are shown as fold change relative to preadipocytes or cycle 0 samples. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 1.1E-15) and test groups (P = 2.96E-10) for total lipid content between 2 groups followed by a Student's t test (*, P < .01; **, P < .001). B, Corresponding total triglyceride content of ACSL1 knockdown vs control adipocytes. The values (mean ± SE; n = 3) were normalized against total protein content of cells and represented as fold change with respect to cycle 0. Two-way ANOVA indicated significant difference in effect of time points (P = 1.02E-08) and test groups (P = 1.73E-03) between the 2 groups followed by a Student's t test (*, P < .01). C, Western blotting showing that ACSL1 depletion in differentiating adipocytes did not influence the expression of key adipogenesis mediators (PPARγ, C/EBPα, and FABP4).

ACSL1 deficiency reduces total lipid accumulation without affecting the adipogenesis differentiation process. A, Total lipid content of cells was detected by Oil Red O staining at different cycles. Values (mean ± SE; n = 3) are shown as fold change relative to preadipocytes or cycle 0 samples. A two-way ANOVA was performed to show the significant difference in effect of time points (P = 1.1E-15) and test groups (P = 2.96E-10) for total lipid content between 2 groups followed by a Student's t test (*, P < .01; **, P < .001). B, Corresponding total triglyceride content of ACSL1 knockdown vs control adipocytes. The values (mean ± SE; n = 3) were normalized against total protein content of cells and represented as fold change with respect to cycle 0. Two-way ANOVA indicated significant difference in effect of time points (P = 1.02E-08) and test groups (P = 1.73E-03) between the 2 groups followed by a Student's t test (*, P < .01). C, Western blotting showing that ACSL1 depletion in differentiating adipocytes did not influence the expression of key adipogenesis mediators (PPARγ, C/EBPα, and FABP4).

Gene expression microarray experiments reveal cytokines as novel ACSL1 downstream target genes

In order to shed light on putative genes and pathways downstream of ACSL1, we conducted gene expression microarray experiments using the same ACSL1 knockdown or control specimens as shown in Figure 6. We identified 309 differentially expressed probes between the knockdown and control groups with P < .05 and FC > 1.2 at cycle 6 of the adipogenesis process and 42 differentially expressed probes between cycles 3 and 6 with P < .05 and FC > 1.2 (heat map in Supplemental Figure 2). Significant genes up-regulated by the ACSL1 knockdown were chemokine (C-C motif) ligand 2 (CCL2) and chemokine (C-C motif) ligand 5 (CCL5) and the macrophage associated surface antigen cluster of differentiation 14 (CD14). In contrast, myosin regulatory light chain interacting protein (MYLIP), a gene involved in regulating the degradation of the low density lipoprotein receptor, was found to be down-regulated in ACSL1 knockdown cells. Each of the above genes was validated by RT-PCR (Figure 7, A–D). In summary, the results from our ACSL1 depletion study indicate that differentiated adipocytes show expression changes in inflammatory genes concomitant with reduced lipid accumulation, as well as decreased glucose uptake.

ACSL1 knockdown gene expression microarray study and qPCR validation of ACSL1 downstream genes. A–D, Gene expression was analyzed by RT-PCR, and relative expression values are shown for selected downstream genes CCL2, CCL5, CD14, and MYLIP. All experiments were performed using RNA samples isolated from 3 replicates, and the data are shown as mean ± SE of gene expression relative to that of endogenous control genes. P value significance was demonstrated by a Student's t test (*, P < .05; **, P < .01).

ACSL1 knockdown gene expression microarray study and qPCR validation of ACSL1 downstream genes. A–D, Gene expression was analyzed by RT-PCR, and relative expression values are shown for selected downstream genes CCL2, CCL5, CD14, and MYLIP. All experiments were performed using RNA samples isolated from 3 replicates, and the data are shown as mean ± SE of gene expression relative to that of endogenous control genes. P value significance was demonstrated by a Student's t test (*, P < .05; **, P < .01).

Obese mice fed a HFD show increased ACSL1 expression in white adipose tissue

In order to interpret our data within a broader in vivo context of metabolic dysfunction and to assess whether ACSL1 associates with the obese state, we used mice with acquired obesity in order to measure the expression of ACSL1 in relation to its putative downstream target genes. Mice were fed with fat-rich (HFD) or control diets, and body weights, as well as additional body composition data, were collected (Figure 8, A–D). Mice fed with a HFD were insulin resistant compared with the controls (Figure 8E), and their corresponding adipocytes were found to be larger in size (Figure 8F).

ACSL1 is overexpressed in obese mice. A–D, Body weight and body composition data of mice (n = 4) fed a HFD or NCD were obtained during the course of 7 weeks as described in Materials and Methods. Statistical analyses for all time points were carried out using a two-way ANOVA (*, P < .05; **, P < .01; ***, P < .001; ****, P < .0001). E, Oral glucose tolerance test was performed at 6 weeks. F, H&E-stained epididymal white adipose tissue sections of representative mice described in A. G, At week 7, RNA samples were isolated from epididymal WAT tissues from all mice, and expression levels of ACSL1 and putative downstream genes were analyzed using RT-PCR. The P value significance was determined by a Student's t test (*, P < .05; **, P < .01).

ACSL1 is overexpressed in obese mice. A–D, Body weight and body composition data of mice (n = 4) fed a HFD or NCD were obtained during the course of 7 weeks as described in Materials and Methods. Statistical analyses for all time points were carried out using a two-way ANOVA (*, P < .05; **, P < .01; ***, P < .001; ****, P < .0001). E, Oral glucose tolerance test was performed at 6 weeks. F, H&E-stained epididymal white adipose tissue sections of representative mice described in A. G, At week 7, RNA samples were isolated from epididymal WAT tissues from all mice, and expression levels of ACSL1 and putative downstream genes were analyzed using RT-PCR. The P value significance was determined by a Student's t test (*, P < .05; **, P < .01).

We further observed that ACSL1 mRNA levels were markedly increased in epididymal white adipose tissue from obese HFD mice and confirmed the same gene expression pattern for the candidate genes CCL5, MYLIP, and CD14 as identified in the ACSL1 depletion study using human MSCs (shown in Figure 7). However, in contrast to our human MSC data, we observed an increase in CCL2 expression in mice exposed to a HFD (Figure 8G).

Discussion

Fetal growth restriction has longer-term consequences that can be interpreted in an adaptive and life course context, first growth restricted infants develop insulin sensitivity in the infant period, which is then replaced in childhood by insulin resistance (28, 29). How this switch occurs is not fully understood and relevant biological test systems derived from human tissues of growth restricted individuals are difficult to obtain. Few studies have addressed this and a recent example is a report describing that adipose precursor cells from adult individuals born with low birth weight showed different DNA methylation and expression levels of the leptin gene (30). The findings indicate that important epigenetic modifications in metabolic signaling pathways are already present at the early postnatal state. Here, we describe for the first time the use of adipocytes differentiated from Wharton's jelly MSCs isolated from growth restricted neonates to study fetal programming. The current study extends our previous report demonstrating that physiological properties of SGA individuals are imprinted in UC-derived MSCs (16). By assessing 2 basal histone modifications, we identified a key player in fatty acid metabolism, ACSL1, as being associated with increased histone H3K27ac marks in differentiated adipocytes from Wharton's jelly derived MSCs obtained from SGA children. This chromatin mark correlated well with corresponding gene expression data suggesting an important functional role for ACSL1 such as mediating increased lipogenesis in times of high demand. ACSL1 belongs to the class of acyl-CoA synthetases that are crucial in fatty acid activation, transport, and degradation, as well as lipid synthesis (31). Due to its central function in lipid metabolism, it is a prime candidate for regulating energy storage with multiple implications in metabolic disorders (32) and in adipocytes, ACSL1 is one of the most highly expressed acyl-CoA synthetases involved in the uptake of fatty acids (33). It has also been demonstrated in mouse 3T3-L1 adipocytes that ACSL1 participates in the reesterification of fatty acids, which are hydrolyzed from intracellular lipid droplets (34, 35). Our results confirm these published data showing that ACSL1 depletion in differentiated adipocytes decreases both, cellular triglyceride and glucose levels. The data stand in contrast to published in vivo findings, demonstrating that adipose tissue-specific ACSL1 knockout mice are obese and cold intolerant, however, with lower plasma glucose levels (36). It is possible that adipose tissue-specific ACSL1 function may favor fatty acid reesterification and subsequent efflux, which could be a contributing factor in the development of insulin resistance as postulated by Lobo et al (34).

We show that MSCs isolated from subjects with fetal growth restriction exert biochemical and metabolic signatures of insulin sensitive cells programmed for enhanced anabolism, evidenced by increased glucose uptake and lipid synthesis. This confirms findings obtained from studies with growth restricted rodents (37). Infants who are growth restricted are generally markedly insulin hypersensitive (38) but later in life become insulin resistant (17). The up-regulation of ACSL1 expression in SGA individuals could therefore function as a double-edged sword, because it may promote rapid catch-up growth but also enhance the release of esterified fatty acids, eventually leading to insulin resistance. Alternatively, one can speculate that in the context of human metabolism, the ACSL1-driven triglyceride biosynthesis enhances the onset of obesity, which creates a platform for the development of metabolic disorders. High-calorie intake accelerates this ACSL1-dependent pathway by promoting ACSL1 expression itself and acting on downstream genes, as shown by our in vivo data from mice fed a HFD. The data show a significant increase in ACSL1 expression compared with chow-fed controls, with concomitant increase in plasma glucose levels. It is therefore conceivable that the elevated ACSL1 levels in individuals with SGA background and a high-caloric diet can act synergistically in inducing obesity and related comorbidities. In future experiments, this hypothesis needs to be addressed in more mechanistic detail using rodents with artificially induced fetal growth restriction in combination with high-calorie and normal diets. Insulin resistance may also be caused by additional pathways downstream of ACSL1 involving chemokines such as CCL2 and CCL5, which have been reported to be expressed in pre- as well as mature adipocytes with both being down-regulated during the course of adipogenesis (39). Our data support this finding as the expression of these genes was found to be significantly altered upon the knockdown of ACSL1. Another gene found to be downstream of ACSL1 encodes the membrane bound protein CD14, involved in toll like receptor interactions important in stress response pathways and previously linked with the onset of insulin resistance (40, 41). ACSL1 modulates the expression of MYLIP, an ubiquitin ligase, in SGA-derived adipocytes. This protein has been shown to down-regulate the receptors for the lipoproteins LDL and VLDL (LDLR and VLDLR), which are well-known atherogenic targets, distributed ubiquitously in all tissues, including adipocytes, and contributing to tissue inflammation (42). Additional inflammatory cytokines may be expressed that in time will affect insulin sensitivity. In line with this, we have recently reported increased expression of the cytokine gene CXCL14 in neonates born with low birth weight and macaques with nutritionally induced growth restriction (43). Developmental programming also affects appetite control (8) and thus exposure to the postnatal obesogenic environment may involve both central and peripheral factors that lead to later obesity and then insulin resistance.

We conclude that ACSL1 found to be overexpressed in adipocytes from Wharton's jelly derived MSCs from SGA subjects may be a key player in mediating fetal programming of human metabolic disease and future experiments will need to address in more detail the underlying mechanisms leading to the up-regulation and metabolic function of ACSL1 in the context of SGA.

Acknowledgments

We thank the expert technical assistance of Jun Hao Tan, Wei Ling Beh, Yhee Cheng Chng, and Maggie Lim.

This work was supported by the Singapore National Research Foundation under its Translational and Clinical Research (TCR) Flagship Programme and administered by the Singapore Ministry of Health's National Medical Research Council (NMRC), Singapore-NMRC/TCR/004-NUS/2008 and NMRC/TCR/012-NUHS/2014. Singapore Institute for Clinical Sciences Investigators are supported through Agency for Science, Technology and Research funding.

Disclosure Summary: The authors have nothing to disclose.

Abbreviations

• acyl-CoA synthetase 1

• adiponectin receptor

• protein kinase B

• CCAAT/enhancer binding protein α

• chemokine (C-C motif) ligand

• cluster of differentiation

• chromatin immunoprecipitation sequencing

• coenzyme A

• fatty acid binding protein 4

• fetal bovine serum

• high-fat diet

• H3K27 acetylation

• H3K27 trimethylation

• mesenchymal stem cell

• normal chow diet

• principal component analysis

• Phosphatidylinositide

• peroxysome proliferator activated receptor γ

• small for gestational age

• small interfering RNA

• thymocyte antigen 1

• umbilical cord.

References

, . . . ;:–., . . . ;():–., . . . ;:–., , , . . . ;:–., , , . . . ;:–., , , , . . . ;:–.. . . ;:–., . . . ;:–., , , . . . ;:–., , , , . . . ;:., . . . ;:–., , . . . ;:–., . . . ;:–., , . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;:–., . . . ;:–., , , et al. . . . ;:., , , et al. . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;():–., , , et al. . . . ;:–.. . . ;:–., , , et al. . . . ;:–.,

Valerie Z Wall

. . ;:–., , . . . ;:–., , , et al. . . . ;:–., . . . ;:–., . . . ;:–., , , . . . ;:., , . . . ;:–., , . . . ;:–., , , et al. . . . ;:–., , , et al. . . . ;:., , , et al. . . . ;:–., , . . . ;:–., . . . ;:., , , et al. . . . ;:., , , . . . ;:–., , , et al. . . . ;:–.

Supplementary data