Context-dependent regulation of lipid accumulation in adipocytes by a HIF1α-PPARγ feedback network

Abstract

Hypoxia-induced upregulation of HIF1α triggers adipose tissue dysfunction and insulin resistance in obese patients. HIF1α closely interacts with PPARγ, the master regulator of adipocyte differentiation and lipid accumulation, but there are conflicting results regarding how this interaction controls the excessive lipid accumulation that drives adipocyte dysfunction. To directly address these conflicts, we established a differentiation system that recapitulated prior seemingly opposing observations made across different experimental settings. Using single-cell imaging and coarse-grained mathematical modeling, we show how HIF1α can both promote and repress lipid accumulation during adipogenesis. Our model predicted and our experiments confirmed that the opposing roles of HIF1α are isolated from each other by the positive-feedback-mediated upregulation of PPARγ that drives adipocyte differentiation. Finally, we identify three factors: strength of the differentiation cue, timing of hypoxic perturbation, and strength of HIF1α expression changes that, when considered together, provide an explanation for many of the previous conflicting reports.

Keywords: HIF1A; HIF1α; PPARG; PPARγ; adipocyte; adipose tissue; bistable switch; cell differentiation; hypoxia; temporal network isolation.

Figure 1.. High levels of HIF1α can both paradoxically promote or inhibit lipid accumulation in cells in the same population.

(A-F) OP9 preadipocyte cells were differentiated by addition of rosiglitazone and then fixed and stained at Day 4 when cells were fully differentiated. LipidTOX was used to quantify lipids. (A) HIF1α knockdown reduced the lipid level. Cells were differentiated by addition of 1 µM rosiglitazone for 4 days. Each dot represents an independent biological replicate of aggregated population median intensities (n=53,730, 20,831, and 43,201 cells for each condition). Bar plot is presented as means ± SD (* p < 0.05, using a two-sided independent t test). (B) HIF1α-overexpressing cells showed reduced lipid accumulation. Cells were differentiated by addition of 1 µM rosiglitazone for 4 days. Each dot represents an independent biological replicate of aggregated population median intensities (n= 53,740; 17,429; and 17,728 cells for each condition). Bar plot is presented as means ± SD (* p < 0.05, using a two-sided independent t test). (C) Wildtype OP9 preadipocyte cells were differentiated and treated with 50 µM PX-478 or 50 µM FG-4592 throughout the differentiation for HIF1α inhibition and activation, respectively. Both chemical inhibition and activation of HIF1α resulted in reduced lipid levels at Day 4. Each dot represents an independent biological replicate of aggregated population median intensities (n= 45,220; 10,688; 18,009; 13,559; and 33,502 cells for each condition). Bar plot is presented as means ± SD (* p < 0.05, using a two-sided independent t test). (D-E) The normalized lipid intensity in each of the thousands of individual cells per condition were plotted in violin plots. The plotted shape width represents the probability density of the data at the different lipid intensity values. (D) HIF1α overexpression in cells differentiated for 4 days with 0.25 µM rosiglitazone resulted in a long-tail distribution of lipid levels. The dotted line is the 99-percentile of the control condition, and the values represent the fraction of the population above the dotted line. Cells over replicates were pooled together (N=2,960; 1,893; and 1,155 cells), and the top and bottom 3 percentile were removed for visualization. (E) Cells differentiated with 1 µM rosiglitazone for 6 days under hypoxic conditions also showed a long-tailed distribution of lipid levels. Cells were differentiated for 6 days either in 21% O2 (gray, normoxia) or 1% O2 (orange, hypoxia). The dotted line is the 99-percentile of the control condition, and the values represent the fraction of the population above the dotted line. Cells over replicates were pooled together (N=7,980; 22,873; and 7,440 cells), and the top and bottom 3 percentile were removed for visualization. (F) Representative images of cells differentiated under hypoxic conditions in (B). Nucleus marked by Hoechst (gray) and lipids marked by LipidTOX Deep Red (cyan). (G) HIF1α overexpression reduced PPARγ level in a doxycycline concentration-dependent manner. Cells were differentiated with 1 µM rosiglitazone for 4 days together with doxycycline or H₂O(control). Each dot represents one replicate averaging several hundred individual cells, and the line plot is a fitted curve of the log-logistic function (* p < 0.05, one-way ANOVA and post-hoc Tukey tests) (STAR Methods).

Figure 2.. HIF1α has two roles during adipogenesis that can be temporally separated.

(A) OP9 cells were induced to differentiate by adding 100 nM rosiglitazone and then fixed and immunocytochemistry was carried out. PPARγ levels increased immediately, but HIF1α increased acutely only two days later. Each dot represents the mean of several hundred individual cells. Results are normalized to the basal, non-perturbed condition. (B) Induction of HIF1α expression early in adipogenesis reduced the lipid level, while the late overexpression increased the lipid level measured at day 6. OP9^treHIF cells were differentiated with rosiglitazone (100 nM for grey bar, 0 nM for white bar) and fixed at day 6 for staining. stblHIF expression was induced by 0.4 µg/ml doxycycline at different timepoints and durations. (C) Early exposure to hypoxia (1% O2) reduced the lipid level, while the late exposure increased the lipid level measured at day 6. OP9 cells were differentiated with rosiglitazone (0.25 µM for grey bar, 0 nM for white bar) and fixed at day 6. Cells were exposed to 1% O2 at different timepoints and durations, where the shaded region represents hypoxia. (B-C) Results are normalized to the basal, non-perturbed condition, which is also indicated as a dashed line. Each dot represents a replicate of aggregated population mean intensities and bar plot is presented as mean ± SD. (D) Scheme of the mathematical model. (E) The simulation reproduced the timing dependency of the system. Mean lipid levels were calculated from 5000 cells differentiated at 0.25 µM R for six days. HIF1α overexpression (${\text{HIF}}_{\text{over}}\text{=1}$) was applied at 0, 48, or 96 h (shaded pattern on the left panels). (F) Through the randomized parameter search, the fraction of parameter sets that could recapitulate the timing-dependent paradoxical regulation was obtained for each model structure. Multiple simulation runs (30,000 searches for four times) are presented as mean ± SD. The models lacking either the positive or negative regulation from HIF1α were unable to reproduce the results observed experimentally in (B) and (C).

Figure 3.. PPARγ positive feedback functionally isolates the inhibitory and activating roles of HIF1α.

(A) Cartoon of the perturbation scheme used to compromise the PPARγ positive feedback and induce HIF1α activation. (B) Model simulation disrupting the positive feedback loop with dnCEBP nonlinearly substantiated the negative effect on lipid levels by HIF1α overexpression. The presence of R, HIFover , dnCEBP indicates $\text{R=}0.25$, HIFover =0.5, Ci =0.33 in the model. The box with a dashed line assumed the linear effect of HIF1α overexpression and dnCEBP expression. (C) The experimental results reproduced the model simulation in (B). Cells were differentiated for 4 days by addition of 1 µM rosiglitazone together with either 12.5 µM FG-4592 or 0.4 µg/ml doxycycline, or both. Each dot represents an independent biological replicate of aggregated population median intensities (N= 45,220; 10,688; 18,009; 13,559; and 33,502 cells for each condition), and the bar plot is presented as means ± SD. (D) Computationally-simulated relationship between PPARγ at 48h and 96 h predicts that HIF1α overexpression affects only cells with intermediate PPARγ before overexpression. Hexagons are colored according to cell density. Darker colors of hexagons mark more cells with the respective 48h and 96h PPARγ levels. The difference in mean PPARγ level at 96h between control (blue) and HIF1α overexpression (orange) conditions is plotted (bottom). (E) Schematics of the live-cell experiment. PPARγ-mCitrine (green) and H2B-Turquoise (red) fluorescence intensities were imaged for 96 h after adding 0.25 µM or 1 µM rosiglitazone to induce differentiation. The orange arrow indicates a representative cell tracked over time. Data are combined from two rosiglitazone concentrations for the following analysis. (F) FG-4592 treatment (25 µM) at 48 h after differentiation mildly inhibited PPAR accumulation. Cells were treated with rosiglitazone (blue, n=3,335 cells), rosiglitazone and FG-4592 (orange, n=4,351 cells) or without rosiglitazone (gray, n=4,179 cells). The bold lines represents population means, and the shaded areas represents values between the 25th and 75th percentile. (G) Experimentally-measured relationship between PPARγ at 48 h and 96 h. Cells were treated with DMSO (control, blue) or FG-4592 (orange) at 48 h. Hexagons are colored according to cell density. Box outlines highlight cells with intermediate PPARγ levels that are most affected by FG-4592. The difference in mean PPARγ level at 96 h between DMSO and FG-4592-treated conditions is plotted (bottom). The shaded area represents standard error of the mean estimated from 100 bootstrapping steps, sampling a half population each time.

Figure 4.. HIF1α perturbation revealed the paradoxical regulatory landscape of adipogenesis.

(A) (left heatmap) The simulated landscape varying R and HIF1α overexpression levels. Color gradients indicate the mean lipid concentration calculated from 5,000 cells at day six. The colors of dashed lines correspond to the following panels. (right panels) The role of HIF1α overexpression is dependent on the differentiation cue. For each different level of R, the mean lipid level was simulated for the cells that were treated with (orange) or without HIF1α overexpression (blue) (left panel). The experiment reproduced the same trend (right panel), where OP9^treHIF cells were differentiated with different levels of rosiglitazone (0.001, 0.004, 0.015, 0.063, 0.25, 1 µM), with or without HIF1α induction through doxycycline. Each experimental and simulation result was normalized to its maximum value. Each dot represents one replicate, and the line plot shows the mean of replicates. (B) (left heatmap) The simulated landscape varying R and HIF1α overexpression levels. The colors of dashed lines correspond to the following panels. (right panels) The emergence of increasing, decreasing and biphasic trends by changing the differentiation cue. Mean lipid levels with varying HIFover were simulated at three different levels of R (0.2 for cyan, 0.4 for red, 0.6 for purple line). For the experimental validation, OP9^treHIF was differentiated by rosiglitazone (0.015 µM for cyan, 0.063 µM for red and 1 µM for purple circles) with different concentration of doxycycline to overexpress HIF1α (0.03, 0.06, 0.13, 0.25, 0.50, 1.00 µM). Each experimental and simulation result was normalized to its maximum value. (C) The 3D landscape of lipid production is dependent on the direction of HIF1α perturbation (inhibition vs activation), the timing of perturbation (early vs late), and the strength of differentiation cue (low vs high R). (D) The effect of HIF1α knockdown and overexpression at different timing on mean lipid production at two different levels of R (0.01 for left, 0.4 for the right panel). Each heatmap color is scaled within each level of R for visualization and the color bar indicates lipid levels (μM). The left half of each panel (corresponding to −1 to 0) indicates ${\text{HIF}}_{\text{kd}}$ values, where −1 indicates the complete inhibition. The right half of each panel (corresponding to 0 to 2) indicates ${\text{HIF}}_{\text{over}}$ values, where 1 indicates the two-fold expression relative to the steady-state HIF1α level (0.025 µM) at R=1. Note that 0 on the x-axis represents endogenous HIF1 levels without any over or under expression. The blue lines on the color bars indicate the lipid level with no perturbation. Y-axis (early to late) indicates the timing of perturbation (days). (E) Perturbation landscape explains the past observations. The landscapes were first binarized to represent an increase or decrease in lipid accumulation relative to the no perturbation (top panels). Then they are reduced to the coarsely divided 2x4 grid (bottom panels: the column positions correspond to ${\text{HIF}}_{\text{kd}}=-1$, ${\text{HIF}}_{\text{kd}}=-0\text{.5}$, ${\text{HIF}}_{\text{over}}\text{=1}$, ${\text{HIF}}_{\text{over}}\text{=2}$; and the row positions correspond to day 0 and day 5 to represent the early and late timing, respectively).