TBX3 Mutations Linked to Hepatic Clonal Expansion

Somatic mutations in TBX3 are observed in human livers. The MOSAICS platform permits the modeling of liver somatic mosaicism in different disease contexts, resulting in unique patterns of hepatocyte expansion and depletion (Figure 1A). Using MOSAICS, we previously observed that sgRNAs targeting the transcription factor Tbx3 were among the most statistically significantly enriched under MASLD conditions (Figure 1B) (11), suggesting that Tbx3 loss confers a selective advantage to hepatocyte clones during MASLD development.

Somatic mutations identified in TBX3 in liver tissues from MASLD patients. (A) Graphic of the MOSAICS screen workflow. (B) β-Score showing sgTbx3 enrichment in WD- versus NC-fed mice after 6 months of selection. (C) Somatic mutations in TBX3 identified from the livers of patients with MASLD. The mutation in blue represents a synonymous mutation. (D) AlphaFold-predicted TBX3 structure (gray) showing somatic mutations (orange) that do not fall within the DNA-binding domain. (E) X-ray crystal structure (PDB ID: 1H6F) of the TBX3 T-box (gray) bound to DNA (teal) showing somatic mutations (orange) with their respective ΔΔG values.

We then asked whether somatic mutations in TBX3 are present in humans with MASLD (Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/JCI191855DS1). Genomic sequencing of liver tissues from MASLD and alcohol-related liver disease patients identified somatic mutations in TBX3 (10), several of which fell within the T-box DNA-binding domain (Figure 1C), indicating that TBX3 mutations may also promote clonal expansion in human disease. To assess potential loss-of-function mechanisms, we mapped the mutations onto the AlphaFold-predicted structure of full-length TBX3. Mutations outside the DNA-binding domain had no predictable impact on TBX3 structure or function (Figure 1D). In contrast, mutations that occurred within the DNA-binding domain localized to key structural elements and were suggestive of loss-of-function mechanisms. To explore this further, we mapped the DNA-binding domain mutations onto the x-ray crystal structure of the TBX3 T-box bound to DNA (17). We then used DDMut to predict changes in Gibbs free energy (ΔΔG) between the WT and mutant proteins (18). The L207P, L263P, and I155S mutations were predicted to have moderate to severe destabilizing effects (ΔΔG ≤ –2.0 kcal/mol) (Figure 1E). Interestingly, the A280S mutation was predicted to only modestly destabilize TBX3 (ΔΔG = –0.94 kcal/mol), but is located near the DNA/protein interface (Figure 1E), suggesting that it may alter TBX3-DNA interactions. These findings suggested that several somatic TBX3 mutations likely result in loss of functionality and could mirror the effects that were observed in mouse MASLD models.

Tbx3 loss protects against MASLD development. To determine whether Tbx3 levels physiologically change during MASLD progression, we performed quantitative PCR (qPCR) on livers from mice on a WD containing high fat, cholesterol, and sugar for up to 36 weeks. Compared with mice fed normal chow (NC) for 12 weeks, hepatic Tbx3 expression was elevated in mice fed a WD for 8, 12, and 20 weeks, but decreased in mice fed a WD for 36 weeks (Figure 2A). To examine the role of Tbx3 in MASLD development, we generated liver-specific Tbx3-knockout (Tbx3-KO) mice. Mice harboring loxP sites flanking exon 1 of Tbx3 (14) were injected with AAV8-TBG-Cre, which leads to hepatocyte-specific Cre expression and loss of Tbx3 mRNA expression in the liver (Figure 2B). To induce MASLD, we fed Tbx3-KO mice a WD for 3 months (Figure 2C). Tbx3-KO mice had decreased liver weights despite unchanged body weights, resulting in lower liver/body weight ratios (Figure 2D), suggesting reduced lipid accumulation. Compared with control mice, macrosteatosis and microsteatosis were attenuated in Tbx3-KO mice (Figure 2E) along with mRNAs involved in fibrosis (Supplemental Figure 1, A and B). We also observed reduced plasma alanine transaminase (ALT) in KO mice, indicative of reduced liver damage (Figure 2F). To ask whether the protection from MASLD could be sustained for longer periods, WT and KO mice were fed a WD for 6 months. Again, Tbx3-KO mice had lower liver/body weight ratios, liver triglycerides, and liver cholesterol (Figure 2, G–I). We then assessed the NAFLD activity score (NAS) (19) after 6 months of WD and found decreased NAS in KO versus WT mice (Figure 2, J and K). Despite the protection from MASLD, these mice had similar levels of liver fibrosis and plasma ALT (Supplemental Figure 1, C–E). Because apoptosis is known to play a role in MASLD progression, and TBX3 has been reported to regulate apoptosis, we checked Tbx3-KO livers for cleaved PARP, a common marker of apoptosis. However, we found no appreciable levels of cleaved PARP in either Tbx3-WT or -KO mice after 3 and 6 months of WD feeding (Supplemental Figure 1F). To determine whether this result is sex specific, we induced MASLD in Tbx3-KO female mice. While female mice develop less MASLD than male mice, Tbx3-KO females still showed reduced steatosis (Supplemental Figure 1, G and H).

Liver-specific Tbx3 deletion protects against MASLD. (A) Tbx3 mRNA levels in WT mouse livers after NC or WD feeding for indicated durations. (B) Tbx3 mRNA levels in Tbx3fl/fl livers 1 week after AAV-TBG-Cre injection. (C) Experimental schema for diet-induced MASLD. (D) Liver weight (left), body weight (middle), and liver/body weight ratio (right) of Tbx3-KO mice on WD for 3 months. (E) Representative H&E images from mice from D. (F) Plasma ALT levels from mice from D. (G) Liver/body weight ratio of Tbx3-KO mice on WD for 6 months. (H) Liver triglycerides from mice from G. (I) Liver cholesterol from mice from G. (J) Representative H&E images from mice from G. (K) NAS scores from mice from G. Significance of relative Tbx3 mRNA in A was calculated using a 1-way ANOVA with Dunnett’s post hoc test. *P < 0.05; **P < 0.01; ****P < 0.0001. Scale bars: 500 μm, left panels; 100 μm, right panels.

We then asked whether Tbx3 deletion would impact MASLD-induced cancer. We induced MASLD-driven tumorigenesis by feeding Tbx3-KO or -WT mice a WD for 48 weeks, allowing tumors to develop within the context of MASLD. After 48 weeks, Tbx3-KO mice again had decreased liver/body weight ratios (Supplemental Figure 2A). They also had decreased surface tumor numbers and sizes (Supplemental Figure 2, B and C). Together, these results suggest that liver-wide Tbx3 loss is protective against WD-induced MASLD.

Tbx3 overexpression exacerbates WD-induced MASLD development. To determine whether increased Tbx3 would be sufficient to promote MASLD, we generated an AAV8 overexpressing a V5-tagged Tbx3 under the control of TBG, a hepatocyte-specific promoter (AAV8-TBG-V5-TBX3). One week after retro-orbital injection into WT mice, immunohistochemistry showed V5-positive staining and reverse transcription–qPCR confirmed increased Tbx3 mRNA expression in the liver (Figure 3, A and B). After AAV delivery, we provided NC or WD for 3 months. On NC, Tbx3 overexpression was insufficient to cause MASLD (Figure 3, C–F). On WD, Tbx3 overexpression led to increases in liver weight, body weight, liver/body weight ratios, and liver triglycerides after 3 months (Figure 3, G and H). Histologically, overexpression led to increased steatosis and lipid droplet accumulation (Figure 3I). Tbx3-overexpressing mice also had increased plasma ALT and aspartate transaminase (AST), indicating increased liver damage (Figure 3, J and K). We observed increased Tbx3 mRNA expression in livers after 3 months (Supplemental Figure 3, A and B), demonstrating that AAV-mediated overexpression was sustained for the entire feeding period. These results showed that Tbx3 overexpression alone cannot drive MASLD, but can accelerate WD-induced MASLD.

Overexpression of Tbx3 exacerbates MASLD. (A) Representative V5 staining from WT mice injected with AAV8-TBG-V5-GFP or AAV8-TBG-V5-TBX3. (B) Tbx3 mRNA levels from livers of mice overexpressing GFP or TBX3. (C)Liver/body weight ratio of Tbx3-overexpressing mice on NC for 3 months. (D) Plasma ALT from mice in C. (E) Plasma AST from mice in C. (F) Representative H&E images from mice in C. (G) Liver weight (left), body weight (middle), and liver/body weight ratio (right) of Tbx3-overexpressing mice on WD for 3 months. (H) Liver triglyceride measurements from mice in G. (I) Representative H&E images from mice in G. (J) Plasma ALT from mice in G. One outlier from each group was excluded. (K) Plasma AST from mice in G. One outlier from each group was excluded. (L) Experimental schema to test the fitness of Tbx3-expressing hepatocytes during WD feeding. (M) Representative V5 staining from WT mice injected with AAV8-TBG-V5-GFP or AAV8-TBG-V5-TBX3 after 4 weeks of NC or WD. (N) Quantification of relative abundance of V5-expressing cells from H. Significance of the V5+ quantification in N was calculated using a 2-way ANOVA with Tukey’s post hoc test. *P < 0.05; **P < 0.01; ***P < 0.001. Scale bars: 500 μm, left panels; 100 μm, right panels.

Because Tbx3-KO hepatocytes are positively selected during MASLD (11), we asked whether forced expression of Tbx3 would confer a selective disadvantage during MASLD. To test this, we overexpressed Tbx3 in WT mice, followed by NC or WD feeding for 4 weeks (Figure 3L). After 4 weeks on an NC diet, there were similar levels of V5-positive cells in GFP- and Tbx3-overexpressing livers (Figure 3, M and N). However, after 4 weeks on WD, there were significantly fewer V5-positive cells in the Tbx3-overexpressing versus GFP-overexpressing control livers (Figure 3, M and N). One possibility is that Tbx3 overexpression led to clonal demise in the context of WD-induced MASLD. Another possibility is that the selective pressure exerted by MASLD caused a subset of Tbx3-overexpressing hepatocytes to silence the AAV transgene in the context of WD feeding. Regardless of the mechanism, these results suggest that forced Tbx3 expression in hepatocytes causes a selective disadvantage specifically in the context of MASLD, which is consistent with results showing that Tbx3 deletion confers a selective advantage in the context of MASLD.

Tbx3 loss does not improve fatty liver through altered insulin resistance. Because insulin resistance is correlated with MASLD progression (20), we asked whether Tbx3-KO mice were protected from MASLD through altered insulin sensitivity. We measured several metabolic parameters in Tbx3-KO mice, including glucose tolerance, fasting insulin levels, plasma non-esterified fatty acids, and plasma triglycerides. Surprisingly, Tbx3-KO mice exhibited increased glucose intolerance, fasting hyperinsulinemia, and increased plasma triglycerides and non-esterified fatty acids (Figure 4, A–D), all of which indicated increased insulin resistance. While selective insulin resistance seen in diabetic patients is a major driver of fatty liver disease, complete loss of hepatic insulin signaling is known to ameliorate MASLD (21), so we asked whether Tbx3 might induce complete insulin resistance. To test this, we maintained Tbx3-KO mice on an NC diet for 6 months. Even on NC, Tbx3-KO mice trended toward lower liver weight and liver/body weight ratios, and had lower plasma ALT (Supplemental Figure 4). Under these conditions, Tbx3-KO mice had similar glucose tolerance, fasted insulin, and non-esterified fatty acids compared with WT mice (Figure 4, E–G). These mice still had elevated plasma triglycerides and plasma cholesterol (Figure 4, H and I). Overall, these results show that Tbx3 loss alone is insufficient to drive insulin resistance, making complete insulin resistance unlikely to be responsible for MASLD protection.

Tbx3 loss exacerbates diet-induced metabolic syndrome. (A) Intraperitoneal glucose tolerance test from Tbx3-KO or -WT mice on WD for 6 months. Tbx3 KO, n = 10; Tbx3 WT, n = 10. (B) Fasting plasma insulin in mice from A. (C) Plasma free fatty acids (FFA) in mice from A. (D) Plasma triglycerides in mice from A. (E) Intraperitoneal glucose tolerance test in Tbx3-KO or -WT mice fed an NC diet for 6 months. Tbx3 KO, n = 8; Tbx3 WT, n = 8. (F) Fasted plasma insulin in mice from E. (G) Plasma FFAs in mice from E. (H) Plasma triglycerides in mice from E. (I) Plasma cholesterol in mice from E. Significance of the glucose tolerance test in A and E was calculated using a 2-way ANOVA with Šidák’s post hoc test. *P < 0.05; **P < 0.01; ***P < 0.001.

Tbx3 deletion transcriptionally upregulates genes involved in VLDL secretion. We next investigated differences in major lipid metabolic pathways in the liver. The liver accumulates lipids primarily through de novo lipogenesis and free fatty acid uptake, while it disposes of lipids through fatty acid oxidation and very low-density lipoprotein (VLDL) particle secretion (22). After 6 months on a WD, Tbx3-KO mice had increased expression of de novo lipogenesis genes, including Srebp1c, Acc1, Acly, and Scd1 (Figure 5A). Similarly, KO mice had elevated expression of genes involved in regulation of free fatty acid uptake, such as Slc27a2, Slc27a5, and Fabp5 (Figure 5B). Upregulation of de novo lipogenesis and free fatty acid uptake are associated with insulin resistance, consistent with the metabolic syndrome in these mice, but are unlikely to mediate protection against MASLD. Next, we assessed regulators of fatty acid oxidation and VLDL secretion. Genes regulating fatty acid oxidation such as Cpt1b, Cpt2, and Acadl were upregulated (Figure 5C). In addition, critical genes for VLDL particle formation such as Mttp and Sar1b were upregulated (Figure 5D). Because phosphatidylcholine (PC) is the major phospholipid found in lipoproteins and known to be required for lipoprotein assembly (23), we also assessed components of the PC biosynthesis pathway. Tbx3-KO mice had elevated expression of PC biosynthesis genes including Chka, Pcyt1a, and Pemt (Figure 5D). These results suggested that Tbx3 loss may protect from MASLD by increasing lipid disposal, either through increased fatty acid oxidation, VLDL secretion, or both.

Tbx3 deletion protects against MASLD by transcriptionally upregulating VLDL-TG particle secretion. (A) qPCR results for genes involved in de novo lipogenesis from Tbx3-KO or -WT mice fed a WD for 6 months. (B) qPCR for free fatty acid uptake genes in mice from A. (C) qPCR for β-oxidation genes in mice from A. (D) qPCR for PC biosynthesis and VLDL secretion genes in mice from A. (E) qPCR for PC biosynthesis and VLDL secretion genes in Tbx3-WT or -KO mice fed a WD for 4 weeks. (F) qPCR for β-oxidation genes in mice from E. (G) Ratio of M+14 myristoylcarnitine to M+16 palmitoylcarnitine in the liver of Tbx3-KO or -WT mice fed a WD for 2 weeks. (H) Lipoprotein fractionation (left) and total plasma triglyceride concentration (right) from Tbx3-KO mice fed a WD for 4 weeks. For lipoprotein fractionation, plasma was pooled from 4 mice per group. (I) Experimental setup for in vivo VLDL triglyceride secretion assay. (J) Plasma triglyceride levels over time from Tbx3-KO or -WT mice fed a WD for 2 weeks. Tbx3 KO, n = 8; Tbx3 WT, n = 7. (K) Quantification of the triglyceride secretion rate from mice from K. (L) Plasma concentration of total apolipoprotein B (ApoB) 3 hours after poloxamer injection in mice from K. (M) Experimental setup to induce MASLD with a CDA-HFD. (N) Liver weight (left), body weight (middle), and liver/body weight ratio (right) of Tbx3-KO or -WT mice fed a CDA-HFD for 14 weeks. (O) Representative H&E images in mice from N. 61. Significance of the difference in plasma triglycerides at 180 minutes after P407 injection in J was calculated using a 2-way ANOVA with Šidák’s post hoc test. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Scale bars: 500 μm, left panels; 100 μm, right panels.

Both fatty acid oxidation and VLDL secretion are impacted by the presence and severity of MASLD, so it is difficult to attribute transcriptional differences directly to loss of Tbx3 versus differences caused by reduced MASLD. To disentangle these possibilities, we examined changes in these pathways early in MASLD progression, or after only 4 weeks of WD (Supplemental Figure 5, A and B). At this time point, we again found transcriptional upregulation of the PC biosynthesis and VLDL secretion genes, but no differences in fatty acid oxidation genes (Figure 5, E and F). These results suggested that upregulation of VLDL particle secretion occurs before fatty acid oxidation, making it more likely to be a direct effect of Tbx3 loss.

Although we did not see a transcriptional upregulation in fatty acid oxidation early in disease, we wanted to ensure that Tbx3 deletion was not increasing flux through β-oxidation to reduce MASLD. After 2 weeks of WD, we tested β-oxidation in Tbx3-KO mice using [13C]palmitate, a recently developed and validated in vivo tracing approach (24). We infused [13C]potassium palmitate into Tbx3-WT or -KO mice and performed metabolomics to calculate the ratio of M+14 myristoylcarnitine to M+16 palmitoylcarnitine to measure the rate of β-oxidation. We found that Tbx3-WT and -KO mice had similar rates of β-oxidation after 2 weeks on a WD (Figure 5G), suggesting that the Tbx3 loss did not strongly perturb fatty acid oxidation.

While Tbx3-KO mice had increased expression of Slc27a2, Slc27a5, and Fabp5, we observed a decrease in the fatty acid transporter CD36 in Tbx3-KO mice after 6 months of WD feeding (Figure 5B). This prompted us to more directly investigate fatty acid uptake in Tbx3-KO mice. We performed lipidomics on Tbx3-KO mice infused with [13C]potassium palmitate after 2 weeks of WD feeding and measured fractional enrichment of M+16 palmitate in the liver and plasma of these mice. We observed no difference in M+16 palmitate enrichment in the plasma (Supplemental Figure 5C), confirming equivalent levels of tracer in the KO and WT mice. Similarly, we found no significant difference in M+16 palmitate enrichment in the liver (Supplemental Figure 5D), suggesting that Tbx3 KO does not directly influence fatty acid uptake.

Loss of Tbx3 accelerates VLDL particle secretion to protect against MASLD. We further investigated the role of TBX3 in VLDL secretion. To understand the effect of Tbx3 deletion on plasma lipids, we performed lipoprotein fractionation in Tbx3-KO mice. After 4 weeks on a WD, Tbx3-KO mice showed an increase in plasma triglycerides, specifically within the VLDL fractions (Figure 5H). To directly test the effect of Tbx3 loss on VLDL particle secretion, we performed an in vivo VLDL-TG secretion assay using the detergent poloxamer-407 after 2 weeks of WD feeding (Figure 5I). KO mice had an increased rate of VLDL-TG secretion (Figure 5, J and K). Interestingly, we also found increased plasma apolipoprotein B (ApoB), suggesting that KO mice had increased numbers of VLDL-TG particles, rather than increased triglycerides per particle (Figure 5L). To test whether the increased VLDL secretion was the driving force behind the protection from MASLD, we used a choline-deficient, l-amino acid–defined high-fat diet (CDA-HFD) to induce MASLD (Figure 5M). This commonly used diet is deficient in both choline and methionine, which restricts PC biosynthesis, leading to lipoprotein retention, steatosis, and other features of MASLD (25). In contrast to mice fed a WD, Tbx3-WT and -KO mice had similar liver weight, body weight, and liver/body weight ratios (Figure 5N). Additionally, H&E showed equivalent levels of lipid accumulation in Tbx3-WT versus -KO livers (Figure 5O). Tbx3-KO mice also had similar levels of plasma ALT and AST, as well as similar levels of liver fibrosis (Supplemental Figure 6, A–D). Liver triglyceride measurements showed a small, but significant, decrease in Tbx3-KO mice (Supplemental Figure 6E), suggesting that the CDA-HFD is incapable of fully rescuing the anti-MASLD phenotype of Tbx3 deletion. These results demonstrated that CDA-HFD impairs the central mechanism of Tbx3 loss, namely accelerated VLDL particle secretion, thereby ablating the protective phenotypes associated with Tbx3 deletion.

Because Tbx3-KO clones are positively selected as a result of increased VLDL-TG secretion, we asked whether altering VLDL secretion is a generalizable mechanism for MASLD-dependent clonal expansion. To see whether VLDL genes are associated with clonal fitness, we reanalyzed previous MOSAICS data (11). In WD-fed livers, Mttp-KO clones were significantly negatively selected, and Tm6sf2-KO clones trended toward negative selection (Supplemental Figure 6F). Notably, Mttp is required for the VLDL particle secretion, while Tm6sf2-KO clones can still secrete VLDL particles that are triglyceride poor. These results suggest that secreting lipids, rather than storing them, is a general mechanism of hepatocyte protection that is exemplified by Tbx3-KO clones.

TBX3 suppresses VLDL secretion through regulating HDLBP. To determine the transcriptional targets of TBX3 in the context of MASLD, we generated Tbx3-KO and V5-tagged Tbx3 overexpression hepatocyte cell lines (Figure 6A). To identify genome-wide TBX3 binding sites, we performed CUT&RUN on Tbx3-overexpressing cells in tandem with Ty1-TBX3 ChIP-Seq on Ty1-TBX3–overexpressing mouse livers (Figure 6B). When we intersected these in vitro and in vivo data, we identified 1,518 genes containing a TBX3 binding site within 10 kb of their transcriptional start sites. Among these loci were Mttp and Chka, canonical regulators of VLDL secretion and PC biosynthesis, respectively. This suggested that TBX3 directly regulates some VLDL secretory components (Figure 6C). We then subjected the genes associated with all overlapping hits to Gene Ontology pathway enrichment analysis (Figure 6D). This identified several pathways related to developmental processes and apoptosis, known functions of TBX3 (12).

TBX3 regulates VLDL secretion through regulating cholesterol homeostasis. (A) Western blots showing Tbx3 KO and overexpression in H2.35 cells. (B) Heatmaps from in vitro CUT&RUN (left) and in vivo ChIP-Seq for TBX3 binding loci. (C) Genomic tracks of VLDL secretion and cholesterol biosynthesis genes showing TBX3 binding in vitro and in vivo. (D) Gene Ontology pathway enrichment analysis from overlapping in vitro and in vivo TBX3 binding sites. (E) Relative secretion of Gaussia luciferase from Tbx3-KO and overexpression H2.35 cell lines. (F) Venn diagram showing the overlap of genes that are transcriptionally upregulated during MASLD in Tbx3-KO livers and have a TBX3 binding site in vitro and in vivo. (G) Hdlbp genomic tracks showing TBX3 binding in vitro and in vivo. (H) Western blot showing HDLBP expression in Tbx3-overexpressing H2.35 cells. (I) qPCR of Hdlbp mRNA levels in livers from Tbx3-KO mice fed a WD for 4 weeks. (J) Western blot showing HDLBP protein levels in livers from Tbx3-KO mice (top) or Tbx3-overexpressing mice (bottom) fed a WD for 4 weeks. (K) Western blot showing Tbx3 and Hdlbp double knockout (DKO) in vivo and triglyceride secretion assay from Tbx3/Hdlbp DKO mice fed a WD for 2 weeks (sgLacZ/sgNT, n = 7; sgLacZ/sgHdlbp, n = 7; sgTbx3/sgNT, n = 7; sgTbx3/sgHdlbp, n = 8). (L) qPCR of cholesterol biosynthesis genes from Tbx3-KO or -WT mice fed a WD for 4 weeks. Significance of the difference in plasma triglycerides among all groups at 180 minutes after P407 injection in K was calculated using a 2-way ANOVA with Tukey’s post hoc test. *P < 0.05; **P < 0.01; ***P < 0.001.