Trim28 Haploinsufficiency Triggers Bi-stable Epigenetic Obesity

Abstract

More than one-half billion people are obese, and despite progress in genetic research, much of the heritability of obesity remains enigmatic. Here, we identify a Trim28-dependent network capable of triggering obesity in a non-Mendelian, "on/off" manner. Trim28(+/D9) mutant mice exhibit a bi-modal body-weight distribution, with isogenic animals randomly emerging as either normal or obese and few intermediates. We find that the obese-"on" state is characterized by reduced expression of an imprinted gene network including Nnat, Peg3, Cdkn1c, and Plagl1 and that independent targeting of these alleles recapitulates the stochastic bi-stable disease phenotype. Adipose tissue transcriptome analyses in children indicate that humans too cluster into distinct sub-populations, stratifying according to Trim28 expression, transcriptome organization, and obesity-associated imprinted gene dysregulation. These data provide evidence of discrete polyphenism in mouse and man and thus carry important implications for complex trait genetics, evolution, and medicine.

Graphical abstract

Figure 1

Trim28 Haploinsufficiency Induces Stochastic, Bi-stable Obesity in the Mouse (A and B) Body mass of wild-type littermates and Trim28^+/D9 mice at 14–18 weeks of age (A) and frequency distribution of body weight normalized to wild-type littermates (B). (C) Obese-Trim28^+/D9 mice are heavier and longer relative to wild-type and lean-Trim28^+/D9 mice. (D and E) Increased mass in obese-Trim28^+/D9 results from expansion of fat depots (eWAT, scWAT, and BAT) and not organomegaly. (F) Trim28^+/D9 mice are lighter at weaning relative to wild-type littermates; body-weight differences between lean- and obese-Trim28^+/D9 mice emerge near adulthood. (G) The Trim28^+/D9 colony body-mass distributions at three different sites (MPI-IE, Germany; IMBA, Austria; QIMR, Australia) with a variable frequency ranging from 10%–50%. (H and I) H&E staining of epididymal adipose (scale bar: 200 μm) shows no sign of adipocyte hypertrophy in Trim28^+/D9 mice. (J) Tissue-specific knockout of Trim28 in liver (Alb-Cre), muscle (McK-Cre), adipose (Adipoq-Cre), POMC (POMC-Cre), or AgRP (AgRP-Cre) neurons does not impact body mass. Data are mean ± SEM (^∗p < 0.05). See also Figure S1.

Figure 2

Non-classical Imprinted Gene Dysregulation Specifies the Obesity “On” State (A–C) Poor correlation between measures of Trim28^+/D9-sensitized obesity and diet-induced obesity (high-fat diet and low-fat diet; HFD/LFD; GSE38337). Anti-correlated genesets from the transcriptome comparison (A) underwent MsigDb pathway enrichment analysis to reveal downregulated (B) or upregulated (C) pathway enrichment specific to obese-Trim28^+/D9 adipose. (D) GSEA analysis reveals marked PEGs downregulation specifically in obese-Trim28^+/D9 mice. (E) Heatmap visualization of the same RNA-seq data reveals that a subset of expressed PEGs (FPKM > 0.3) is downregulated in obese-Trim28^+/D9. Genes marked with an asterisk belong to IGN1. Columns represent sequenced biological replicates. (F) GSEA enrichment of imprinted gene networks (IGN1-3; Al Adhami et al., 2015). (G–I) Imprinted genes are dysregulated non-classically. (G) No changes in DNA methylation at germline DMRs measured by quantitative bisulfite and (H) no changes at the gene body or (I) promoters of imprinted genes as measured by RRBS in mature adipocytes. Imprinted genes are shown in red. RRBS data represent the mean of two independent replicates per group. Data are mean ± SEM (^∗p < 0.05). See also Figure S2 and Tables S1, S2, S3, and S4.

Figure 3

Nnat and Peg3 Knockout Mice Exhibit Bi-stable Obesity (A) Targeting approach for a “knockout first” Nnat deletion allele. (B) mRNA and (C) protein-level expression in E17.5 embryo heads of paternal Nnat null (Nnat^+/−p) mutants and their wild-type littermate controls (+/+). (D) Hypervariable body mass at 14–18 weeks of age observed upon deletion of the paternal and not maternal (Nnat^+/−m) allele. (E) Body-mass distribution is bi-modal (gray). The single-Gaussian sub-distributions of the double-Gaussian fit (gray) are shown in green. Inset highlights that body-mass distribution upon maternal transmission (Nnat^+/−m) of the Nnat null allele follows a single-Gaussian distribution. (F) Hypervariable body-mass distributions observed at two different sites (MPI-IE, Germany; MRC, UK). Shown are male progeny of ∼10 litters at each site. (G) Epididymal adipose tissue mass from lean- and obese-Nnat^+/−p mice and their littermate controls. (H) Body-fat distribution of Peg3^+/−p mice is bi-modal (gray). The single-Gaussian sub-distributions of the double-Gaussian fit (gray) are shown in blue (re-graphed from Curley et al., 2005). (I) Individual replicates for total fat mass and body weight for Peg3^+/−p mice (re-graphed from Curley et al., 2005). (J and K) Obese-Trim28^+/D9 and obese-Nnat^+/−p animals exhibit decreased mRNA expression of most of the recruitment factors (Zfp57, Hp1α, and Hp1γ) concomitant with increased expression of silencing factors SetDb1, Dnmt’s 1, 3a, and 3L relative to their lean siblings. Data are median with interquartile range (boxplots) or mean ± SEM (^∗p < 0.05). See also Figure S3.

Figure 4

Trim28-Low Human Children Are Obesity Susceptible and Exhibit a Distinct Transcriptome Landscape (A) Taqman qPCR of TRIM28 mRNA expression in sub-cutaneous adipose from human pre-pubescent children. (B) Stratification by BMI indicates that the obese sub-group is enriched for Trim28-low individuals. (C) Taqman qPCR shows that obese, Trim28-low individuals specifically exhibit reduced expression of IGN1-imprinted genes. (D) Correlation of TRIM28 versus IGN1 member gene expression. (E) PCA of adipose tissue RNA-seq from the same individuals reveals Tim28-low versus -high individuals to be substantially different. Inset highlights the same to be true when analyzing only IGN1 qPCR data. (F) Heatmap visualization of hierarchical clustering of the most variable 6,000 genes expressed in adipose from the cohort. Vertical lines are for visualization purposes only. (G) TRIM28 and (H) IGN1 pathway expression are selectively decreased in obese co-twins in a cohort of 13 discordant monozygotic twin pairs (Pietiläinen et al., 2008). Data are mean ± SEM (^∗p < 0.05) or min-to-max whiskers. See also Figure S4.

Figure 5

BMI of the General Population Is Consistent with a Bi-modal Distribution (A) BMI distribution of 6- to 11-year-old non-hispanic white males from the continuous NHANES 1999–2012 survey (CDC, 2012). Data are fit to a single Gaussian (gray) and a double Gaussian (blue). (B) Individual Gaussian components of the double Gaussian from (A). (C) Near-perfect double-Gaussian fit is observed across children of five major ethnicity classes, as well as in adult cohorts from (D) continuous NHANES 1999–2012 (CDC, 2012) and (E) Han chinese populations. (D and E) Shown are age-normalized BMI distributions for females aged 25–50. (F) Comparison of similar fitting of childhood data from continuous NHANES 1999–2012 (CDC, 2012) and prior NHANES/NHES surveys (1963–1994) (CDC, 1994) shows a marked shift in recent decades where the heavy sub-population triples in size (pie charts). See also Figure S5.

Figure S1

Obese-Trim28^+/D9 Animals Are Metabolically “Healthy,” Related to Figure 1 (A) Trim28 is reduced and equally expressed in lean- and obese-Trim28^+/D9 animals relative to wild-type littermates. (B) Comparison of body mass, coefficient of variation, and standard deviation against large, independently obtained, non-early-life-controlled cohorts confirmed the heightened variation specifically of the Trim28^+/D9 heterozygotes. (C–H) Assessment of metabolic health by OGTT (C), shows that both lean- and obese-Trim28^+/D9 exhibit unaltered glucose homeostasis and plasma levels of (D) FFA and (E) triglycerides. Obese-Trim28^+/D9 animals showed (F) slightly higher blood glucose levels after 6 hr fast and (G) elevated serum levels of insulin. (H) Plasma levels of adipokines known to be associated with metabolically diseased obesity were unremarkable. (I) H&E staining of liver sections (scale bar: 200μm) revealed no sign of hepatosteatosis in Trim28 haploinsufficient animals consistent with being metabolic healthy. (J and K) Trim28^+/D9-induced obesity was associated with moderate expression changes with both (J) positive and (K) negative metabolic outcomes. (L–O) Indirect calorimetry in obese- and lean-Trim28^+/D9 animals show changes in (L) V_O2, V_CO2 and (M) energy expenditure (n = 3–5). The changes are associated with trend toward reduced (N) activity and (O) unaltered food intake. Data are mean ± SEM (^∗p < 0.05).

Figure S2

Non-classical Imprinted Gene Dysregulation in Trim28^+/D9-Induced Obesity, Related to Figure 2 (A and B) Correlation of RNA-seq data of epididymal adipose tissue from obese-Trim28^+/D9 animals with either (A) wild-type or (B) lean-Trim28^+/D9 littermates. (C) “Imprinted genes” is the 3^rd most discordant gene set when comparing high-fat diet induced and Trim28^+/D9-sensitized obesity. (D) Heatmap of gene expression changes to maternally expressed imprinted genes. (E) Reduced expression of Nnat, Plagl1, and Peg3 in obese-Trim28^+/D9 animals was confirmed by qPCR. (F) Nnat was also found reduced at the protein level when assessed by immunofluorescence. DAPI is shown in blue, Nnat in red and 488-AF (autofluorescence) in green. (G) GSEA of diet-induced obesity adipose tissue mRNA expression data reveal, if anything, an opposite regulation of IGN1. (H) Bisulfite pyrosequencing of germline DMR’s in stromal vascular adipocyte progenitor preparations from lean- and obese-Trim28^+/D9 showed no differences. Data are mean ± SEM (^∗p < 0.05).

Figure S3

Increased Adiposity in Obese-Nnat^+/−p Mice, Related to Figure 3 Dual-emission X-ray absorption analysis for lean and fat mass in heavy and light paternal Nnat deletion mutants. Means of each group are shown as squares. Data are mean ± SEM (^∗p < 0.05).

Figure S4

PCA of Human Childhood Adipose Tissue Transcriptome Data, Related to Figure 4 (A and B) PCA of adipose tissue RNA-seq from children of the Leipzig AT cohort revealed Tim28-low versus -high individuals to be substantially different. Visualization of (A) males versus females, or (B) obese versus lean individuals does not yield the same group segregation. Insets highlight the same to be true when analyzing on IGN1 qPCR expression data. (C) Hierarchical clustering of the heatmap visualization of the 6,000 most variable, expressed genes reveals the relative genic sub-structure of the major clusters including the major Trim28-low and -high stratified clusters. Vertical lines are for visualization purposes only.

Figure S5

NHANES Population BMI Distributions, Related to Figure 5 (A) BMI distribution of 6- to 11-year-old non-hispanic white male participants from the continuous NHANES 1999–2012 survey (CDC, 2012). Data are fit to a single Gaussian (gray) and a double Gaussian (blue) and visualized on a log-scale to highlight data from the more rare, morbidly obese individuals that avoid the fit. (B) Individual Gaussian components of the double Gaussian from (A). (C) FTO expression is significantly reduced in obese-Trim28^+/D9 animals. Data are mean ± SEM.