Quantification results were presented as mean ± SD (n = 3; unless otherwise indicated). Statistical analysis was performed by two-tailed unpaired Student's t-test conducted using GraphPad Prism 8.4 (GraphPad Software Inc., Boston, MA). When more than two groups were compared, one-way ANOVA followed by Tukey multiple comparisons test was performed. A value of P < 0.05 were considered statistically significant.

To investigate the biological function of NeuroD1 in HCC, we constructed two lentiviral shRNA expression vectors targeting different sites of NeuroD1, and confirmed their knockdown efficacy in HCC-LM3 and HepG2 cells (Supplementary Fig. 1A-C). To systematically reveal the role of NeuroD1 in HCC cells metabolism, we performed metabolomic analysis by subject ting NeuroD1-knocked down HCC-LM3 cells to liquid chromatography-mass spectrometry (LC-MS). Principal component analysis revealed a significant difference between the metabolites in NeuroD1-knocked down cells and those in control cells (Supplementary Fig. 1D). Metabolomic analysis results revealed 1,705 differential metabolites in NeuroD1-knocked down HCC-LM3 cells, with 968 of them upregulated and 737 of them downregulated (Fig. 1A). Further analysis of these differential metabolites using Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis revealed an obvious enrichment in "lipid metabolism" (Fig. 1B).

To verify the metabolomic analysis results, we next analyzed lipid accumulation and NeuroD1 expression levels in clinical HCC tissues. Oil Red O staining results revealed that compared to normal adjacent tissue, lipid accumulation increased in clinical HCC tissues (Fig. 1C), in which NeuroD1 expression level was also upregulated significantly (Fig. 1D). Furthermore, knocking down NeuroD1 reduced lipid accumulation in HCC-LM3 and HepG2 cells (Fig. 1E), while overexpressing NeuroD1 led to the opposite result (Fig. 1F, Supplementary Fig. 1E). Taken together, these results demonstrate that NeuroD1 is a positive regulator of HCC lipid metabolism.

In order to further determine the detailed pathway by which NeuroD1 regulates lipid metabolism in HCC, we conducted further KEGG enrichment analysis on the differential metabolites associated with lipid metabolism identified in the metabolomic analysis described above. In NeuroD1-knocked down HCC-LM3 cells, 30 lipid metabolism-related metabolites were differentially expressed: 22 of them were upregulated and 7 were downregulated (Fig. 2A). KEGG analysis showed that "cholesterol biosynthesis" and "steroid biosynthesis" pathways were enriched in NeuroD1-knocked down cells (Fig. 2B). We next verified this result by analyzing the effect of altering NeuroD1 expression on HCC cells cholesterol levels. The results showed that knocking down NeuroD1 using lentivirus expressing shRNA significantly decreased total cholesterol (T-CHO), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) levels in HCC cells (Fig. 2C-E), while overexpressing NeuroD1 robustly increased them (Supplementary Fig. 2A-C). Knocking down NeuroD1 using siRNAs further confirmed its positive regulation on cholesterol levels (Supplementary Fig. 2D-G). These results revealed that NeuroD1 could regulate HCC cells cholesterol biosynthesis.

Given that cholesterol is crucial for tumorigenesis, we next assessed the effect of altering NeuroD1 on HCC cell tumorigenic potential. Knocking down NeuroD1 resulted in a significant decrease of HCC cell viability and colony formation potential (Supplementary Fig. 3A, B), most plausibly by reducing their proliferation potential (Supplementary Fig. 3C) and promoting their apoptosis (Supplementary Fig. 3D). Meanwhile, overexpressing NeuroD1 led to the opposite (Supplementary Fig. 3E-G), demonstrating that NeuroD1 plays critical roles in HCC cell proliferation and tumorigenic potential. Furthermore, supplementing NeuroD1-knocked down HCC-LM3 cells with extracellular cholesterol restored their viability, colony formation, and proliferation potentials (Fig. 2F-H). Similar results were also shown by NeuroD1-knocked down HepG2 cells supplemented with extracellular cholesterol (Supplementary Fig. 4A-C). Together, these results clearly showed that NeuroD1 positive regulation on HCC cell cholesterol metabolism is critical for its oncogenic function.

To elucidate the molecular mechanism of cholesterol biosynthesis in HCC regulated by NeuroD1, we analyzed the genomic localization of NeuroD1 around the transcription start site (TSS) of lipid metabolism-related genes using a chromatin immunoprecipitation-sequencing (ChIP-seq) dataset containing DNA fragments that could be bound by NeuroD1 (GSE179072) [23] (Fig. 3A). DIVID analysis of potential genes bound by NeuroD1 revealed significant enrichment in the "cholesterol biosynthesis pathway" (Supplementary Fig. 5A). We further overlapped the potential NeuroD1 binding genes identified by ChIP-seq with those "cholesterol biosynthetic process" obtained by QuickGO screening. Nine genes were identified as potential NeuroD1 direct transcriptional targets (Fig. 3B). Next, we analyzed their mRNA expression levels in NeuroD1-knocked down HCC-LM3 cells. Knocking down NeuroD1 significantly suppressed the expression levels of FDFT1 mRNA and protein in HCC-LM3 cells (Fig. 3C, D), while overexpressing NeuroD1 led to an increase in FDFT1 in HCC cells (Fig. 3E). It is noteworthy that knocking down NeuroD1 only slightly altered the expression levels of PRKAA2, HMGCR, ACAA2, NPC1L1, and IDI2, while it did not induce any significant effect on the expression levels of STK33, TM7SF2, and INSIG2. Moreover, NeuroD1 knockdown did not significantly affect those of SQLE and LSS, which are also key enzymes in cholesterol biosynthesis, as well as LDLR1 and ABCA1, which are key factors for cholesterol uptake and efflux, respectively (Supplementary Fig. 5B). Meanwhile, while NeuroD1 knockdown slightly affected those of SREBP2, another factor in cholesterol biosynthesis, and ACAA1, a key enzyme for cholesterol esterification, the effect was significantly lower than that on FDFT1. These results suggest that NeuroD1 positively regulates FDFT1 expression.

We next investigated FDFT1 expression level in clinical HCC tissues using The Cancer Genome Atlas (TCGA) database, and found that FDFT1 was highly expressed in HCC tissue compared with adjacent tissue (Fig. 3F). This tendency was further confirmed by immunofluorescent staining of clinical HCC tissues (Fig. 3G). Furthermore, FDFT1 expression level positively correlated with poor prognosis (Fig. 3H). Together, these results indicate that NeuroD1 is a positive regulator of FDFT1 expression.

To elucidate the role of FDFT1 in NeuroD1-mediated cholesterol biosynthesis, we first constructed two lentivirus-based shRNA expression vectors targeting different sites of FDFT1, a lentivirus-based FDFT1 overexpression vector (Supplementary Fig. 6A-C), as well as two siRNAs targeting FDFT1 synthesized through in vitro translation (Supplementary Fig. 6D). Knocking down FDFT1, either using shFDFT1 expression vectors or siFDFT1s robustly reduced the T-CHO, LDL-C, and HDL-C levels in HCC-LM3 cells (Supplementary Fig. 6E-J), while overexpressing FDFT1 resulted in the opposite (Supplementary Fig. 6K-M). Furthermore, knocking down FDFT1 suppressed HCC-LM3 cell viability and colony formation potential (Supplementary Fig. 6N, O), while overexpressing FDFT1 enhanced them (Supplementary Fig. 6P, Q). These results confirmed the role of FDFT1 in promoting cholesterol biosynthesis and subsequently, HCC cells tumorigenic potential.

To further elucidate the role of FDFT1 in NeuroD1-mediated HCC cholesterol biosynthesis, we performed rescue experiments by overexpressing FDFT1 in NeuroD1-knocked down HCC-LM3 cells (Supplementary Fig. 7A). Overexpressing FDFT1 robustly restored lipid accumulation and the levels of T-CHO, LDL-C, and HDL-C in these cells (Fig. 4A-D), thus confirming the crucial role of FDFT1 in NeuroD1-mediated HCC cholesterol metabolism. Furthermore, overexpressing FDFT1 clearly re-increased the viability, colony formation potential, and proliferation potential of NeuroD1-knocked down HCC-LM3 cells (Fig. 4E-G), while restoring that of apoptosis (Fig. 4H). Meanwhile, knocking down FDFT1 re-suppressed lipid accumulation and cholesterol levels upregulated by NeuroD1 overexpression (Supplementary Fig. 7B-F), abolishing the effect of NeuroD1 overexpression in increasing the viability, colony formation, and proliferation potentials of NeuroD1-overexpressed HCC-LM3 cells (Supplementary Fig. 7G-I), while restoring their apoptotic rate (Supplementary Fig. 7J).

We next examined the effect of inhibiting FDFT1 using its inhibitor, YM-53601, on NeuroD1-induced cholesterol biosynthesis. Treatment with YM-53601 canceled the effect of NeuroD1 overexpression in increasing the T-CHO, LDL-C, and HDL-C levels in HCC-LM3 cells (Supplementary Fig. 8A-C). Consequently, YM-53601 treatment cancelled the increase of cell viability and restored the apoptotic rate in NeuroD1-overexpressed HCC-LM3 cells (Supplementary Fig. 8D, E). These results demonstrate that FDFT1 is critical for NeuroD1-mediated regulation on HCC cholesterol biosynthesis, and subsequently, on HCC cell tumorigenic potential.

To reveal the molecular mechanism of NeuroD1 regulation on FDFT1 transcription, we predicted three potential NeuroD1 binding sites on FDFT1 promoter using JASPAR (https://www.jaspar.genereg.net). These binding sites are located in the -851 to -843, +226 to +234 and +679 to +689 regions of the FDFT1 promoter (Fig. 5A). Accordingly, we constructed a series of luciferase reporter vectors carrying the -1014 to +1028, +96 to +1 028, +531 to +1028, and +761 to +1028 fragments of the FDFT1 promoter (Fig. 5B). Luciferase reporter assay results showed that knocking down NeuroD1 significantly suppressed the activities of FDFT1-Luc-1, FDFT1-Luc-2, and FDFT1-Luc-3 in HCC-LM3 cells; however, it failed to significantly affect that of FDFT1-Luc-4 (Fig. 5C). Similarly, overexpressing NeuroD1 robustly increased the activities of FDFT1-Luc-1, FDFT1-Luc-2, and FDFT1-Luc-3 without significantly affecting that of FDFT1-Luc-4 (Fig. 5D). These results suggest that NeuroD1 promotes FDFT1 promoter transcriptional activity, and that the +531 to +760 region of the FDFT1 promoter is critical for this regulation.

To further assess whether NeuroD1 could bind to the FDFT1 promoter on the +679 to +689 region as predicted, we then performed chromatin immunoprecipitation (ChIP) assay using a primer set flanking it (Fig. 5E). The results demonstrated that NeuroD1 could directly bind to the +549 to +828 region of the FDFT1 promoter (Fig. 5F, G). Finally, we constructed a mutant FDFT1-Luc-3 reporter vector (FDFT1-Luc-3) by mutating the core sequence in the predicted NeuroD1 binding site on the +679 to +689 region of the FDFT1 promoter (Fig. 5H). Our results show that, unlike its effect on FDFT1-Luc-3, knocking down and overexpressing NeuroD1 did not have any significant effect on FDFT1-Luc-3 (Fig. 5I, J). Taken together, our findings suggest that NeuroD1 can upregulate FDFT1 transcriptional activity, most plausibly through direct binding with the FDFT1 promoter at +679 to +689 region.

Recent studies have shown that epigenetic modification plays an important role in regulating gene transcription activity [24]. For example, histone acetylation, which could increase chromatin opening and accessibility, is usually associated with transcriptional activation, while histone deacetylation is usually associated with transcriptional repression [25]. Interestingly, the genomic tracks analysis of the ChIP-seq (GSE179072) [23] showed that the FDFT1 promoter region enriched by anti-H3K27Ac overlapped with that enriched by anti-NeuroD1, suggesting that NeuroD1 might also be involved in histone acetylation-mediated epigenetic regulation of FDFT1 promoter (Fig. 6A, B). Inhibiting histone acetylation using anacardic acid suppressed FDFT1 expression level in a dose-dependent manner, thereby further confirming the possibility of the FDFT1 promoter epigenetic regulation of FDFT1 promoter by H3K27 acetylation (Fig. 6C). To further elucidate the molecular mechanism of FDFT1 promoter H3K27 acetylation, we synthesized siRNAs targeting currently reported histone acetylase: E1A binding protein p300 (EP300), CREB binding protein (CBP), lysine acetyltransferase 2A (KAT2A), lysine acetyltransferase 5 (KAT5), and lysine acetyltransferase 6A (KAT6A) [26, 27]. The results showed that knocking down KAT5 and KAT6A did not exert any significant effect on the activity FDFT1-Luc-3, while knocking down CBP and EP300 only slightly suppressed it. Meanwhile, knocking down KAT2A robustly suppressed FDFT1-Luc-3 (Fig. 6D, E). These results conformed with the effect of knocking down these histone acetylases on FDFT1 mRNA expression level (Fig. 6F). It is also noteworthy that analysis using TCGA database revealed a positive correlation between FDFT1 and KAT2A expression levels in clinical HCC tissues (Fig. 6G). Together, these results revealed that FDFT1 expression level could be regulated epigenetically by KAT2A-mediated H3K27 acetylation.

Next, to reveal the role of NeuroD1 on KAT2A-mediated FDFT1 promoter H3K27 acetylation, we predicted the possible interaction between NeuroD1 and KAT2A. Molecular docking result showed that multiple hydrogen bonds could be formed between NeuroD1 and KAT2A proteins, with the docking binding energy of -22.2 kcal/mol (Supplementary Fig. 9A). These results, which showed a possible strong binding between the two proteins, were in accordance with the abovementioned genome tracks analysis results showing an overlapping area enriched by NeuroD1 and H3K27 acetylation in FDFT1 promoter. To validate this prediction, we next performed immunoprecipitation using HCC-LM3 cell lysates overexpressing His-KAT2A and FLAG-NeuroD1. The results revealed that KAT2A could be detected in the precipitants pulled down by anti-Flag antibody and vice versa, indicating the physical interactions between KAT2A and NeuroD1 proteins (Fig. 6H, I). Furthermore, ChIP assay results showed that H3K27 acetylation significantly decreased in NeuroD1-knocked down HCC-LM3 cells (Fig. 6J, Supplementary Fig. 9B, C); in contrast, it conspicuously increased in NeuroD1-overexpressed HCC-LM3 cells (Fig. 6K, Supplementary Fig. 9D).

Furthermore, knocking down KAT2A robustly restored lipid accumulation and the levels of T-CHO, LDL-C, and HDL-C in HCC-LM3 cells (Supplementary Fig. 9E-H), thus confirming the crucial role of KAT2A in NeuroD1-mediated HCC cholesterol metabolism. Knocking down KAT2A also clearly re-suppressed NeuroD1 overexpression-induced HCC-LM3 cells viability, colony formation potential, and proliferation potential (Supplementary Fig. 9I-K). Together, these results suggest that NeuroD1 can promote FDFT1 transcription by recruiting KAT2A and promoting H3K27 acetylation, leading to the increase of cholesterol biosynthesis, and subsequently, tumorigenic potential.

To elucidate the pathological function of the NeuroD1/FDFT1 pathway in vivo, we performed xenograft experiments using NeuroD1-knocked down and NeuroD1-knocked down, FDFT1-overexpressed HCC-LM3 cells. Knocking down NeuroD1 reduced tumor growth; however, overexpressing FDFT1 restored it (Fig. 7A). These tendencies were further confirmed by the tumor morphology (Fig. 7B). Western blotting and immunofluorescence staining showed that knocking down NeuroD1 strongly decreased FDFT1 expression in xenograft tumor lesions formed by HCC-LM3 cells, whereas overexpressing FDFT1 restored it (Fig. 7C, D). Moreover, overexpressing FDFT1 restored lipid accumulation and cholesterol levels in the xenograft tumor lesions formed by NeuroD1-knocked down HCC-LM3 cells (Fig. 7E-H). These findings indicate that FDFT1-mediated cholesterol metabolic reprogramming is crucial for NeuroD1 tumorigenic potential.

Collectively, our findings suggest that NeuroD1 not only could promote FDFT1 expression through transcriptional regulation, but also by recruiting histone acetylase KAT2A and enhancing KAT2A/H3K27 acetylation-mediated epigenetic regulation. This in turn enhances cholesterol biosynthesis, and subsequently, HCC tumorigenesis (Fig. 8).