Fig 2: Overexpression of miR-148b Rescues Cytokine-Induced EndMTHUVECs were transfected with miR-148b mimic, SMAD2 siRNA, or control and treated with TNF-α/IL-1β for 6 days. (A) Representative western blot analysis of TGFB2, SMAD2, p-SMAD2 CD31, VE-cadherin, SLUG, FSP-1, and COL1A1; HDAC3 is used as a loading control (n = 3). (B) Immunofluorescence images showing localization of CD31, VE-cadherin, COL1A1, and FSP-1 in miR-148b- or control-transfected HUVECS and/or treated with TNF-α/IL-1β; scale bars, 50 μm (magnification 400×) (n = 3). (C) Representative western blot analysis of SMAD2, CD31, VE-cadherin, SLUG, COL1A1, and FSP-1; HDAC3 is used as a loading control (n = 3). (D) Immunofluorescence images showing localization of CD31, VE-cadherin, COL1A1, and FSP-1 in SMAD2 siRNA or control-transfected HUVECS and/or treated with TNF-α/IL-1β; scale bars, 50 μm (magnification 400×) (n = 3).

Fig 3: The HDAC3–PROX1 module suppresses steatosis by controlling a hepatic lipid metabolism gene program. a Western blot of liver samples from mice treated for 3 weeks with AAV8 TBG shLuciferase (Prox1 Ctl) or shPROX1 (Prox1 KD). Densitometry of the western blot shown above (n = 3). Data are presented as mean ± s.d. b Hepatic triglyceride assay of livers infected for 3 weeks (n = 3) or 6 weeks (n = 6) with indicated virus. Data are presented as mean ± s.d. c Oil red O staining of livers infected for 6 weeks as in (b). Scale bar is 50 µm. d RNA-seq analysis of Hdac3 fl/fl mice infected for 2 weeks with AAV8 TBG Egfp (n = 2) or Cre (Hdac3 KO, n = 3) vs. wild-type mice infected with AAV8 TBG shLuciferase (n = 3) or shPROX1 (Prox1 KD, n = 3) for 3 weeks. Heatmap displays coregulated genes (1.7-fold change, P < 0.05) grouped by expression correlation (135 genes) or anti-correlation (50 genes) upon ablation of HDAC3 or PROX1. Scale bar represents log2(fold change). Inset shows the P-value and corresponding gene list for the highest ranking Reactome pathway in the co-upregulated cluster. e Heatmap displays co-upregulated lipid-related genes upon ablation of HDAC3 or PROX1 and the corresponding binding strength of adjacent HDAC3–PROX1 co-bound peaks (−50kb upstream of the transcription start site, TSS through +2 kb from the transcription end site, TES) in the HNF4α KO liver relative to control. Scale bar represents log2(fold change) for RNA-seq and fold-change for ChIP-seq. f Example ChIP-seq and GRO-seq browser tracks at the G0s2 locus. Red boxes indicate location of putative G0s2 enhancers. g Luciferase assay (n = 3) indicating transcriptional response to co-expression of HNF4α, PROX1 and C/EBPα at G0s2 enhancers as identified in (f). Data are presented as mean ± s.d. Two-tailed unpaired Student’s t-test, *P < 0.05, **P < 0.01, ***P < 0.001, ns not significant

Fig 4: Increased ABAT expression leads to reduced H3K4ac through HDAC3-mediated histone deacetylation(A and B) H3K4ac expression analysis in neural cells depicts significantly decreased levels in neurons (n = 3) compared with E12.5 NSC (n = 48), P2 cerebellar SC (n = 34), P2 hindbrain SC (n = 35), and astrocytes (n = 3). Scale bar, 50 µm; one-way ANOVA followed by Tukey’s multiple comparison test.(C and D) ABAThigh cells in D425 (n = 13), CHLA-01-MED (n = 17), and CHLA-01R-MED (n = 27) display significantly decreased H3K4ac expression compared with their ABATlow counterparts (n = 50 for all cell lines). Scale bar, 20 µm; t test.(E–H) ABAT OE cells in both in vitro (n = 50) and in vivo (n = 60) conditions have significantly lower H3K4ac expression compared with ABAT KD cells (n = 50 in vitro and n = 60 in vivo). Scale bar, 50 µm; t test.(I and J) LMs show significantly decreased H3K4ac (n = 160) compared with cerebellar tumors (n = 160). Scale bar, 50 µm; t test.(K) Histone deacetylase (HDAC) activity analysis reveals significantly increased HDAC activity in ABAT OE cells (n = 3).(L and M) H3K4ac deacetylase HDAC3 is significantly higher in ABAT OE cells (n = 50) compared with ABAT KD cells (n = 50). Scale bar, 50 µm; t test.n = nuclei (B, D, F, H, and J) or cells quantified (M) or replicates (K). Histogram data depict mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See also Figure S5.

Fig 5: The HDAC3–PROX1 module is recruited by HNF4α in liver. a Comparison of HNF4α binding (>2 rpm, filtered on the HNF4α KO, 50% minimum overlap) at overlapping and non-overlapping peaks in the HDAC3 and PROX1 cistromes from Fig. 2a. Numbers above brackets indicate P-values, χ 2 test. b HOMER motif analysis of peaks co-bound by HDAC3, PROX1, and HNF4α displaying over-represented sequences. c Representative browser tracks of HNF4α, PROX1, and HDAC3 ChIP-seq in Hnf4α fl/fl livers infected with AAV8 TBG Egfp (Ctl) or Cre (Hnf4α KO). Indicated scales are in RPTM. d Scatter plots of PROX1 (left) and HDAC3 (right) ChIP-seq in control versus Hnf4α KO. Red and blue shaded regions indicate a twofold decrease in peak intensity upon loss of HNF4α. e, f Co-occupancy of HDAC3–PROX1 and HNF4α as indicated by ChIP–reChIP (n = 3) from liver. Legend indicates reChIP antibody following primary HDAC3 or PROX1 ChIP elution. Data are presented as mean ± s.d, one-tailed unpaired Student’s t-test, *P < 0.05, **P < 0.01, ***P < 0.001, ns not significant