Close this search box.

NOTCH localizes to mitochondria through the TBC1D15-FIS1 interaction and is stabilized via blockade of E3 ligase and CDK8 recruitment to reprogram tumor-initiating cells – Experimental & Molecular Medicine

The TBC1D15-NICD interaction increased the mitochondrial diameter and recruited mitochondria to the perinuclear region

Previous analyses of TICs by NANOG ChIP (chromatin immunoprecipitation)-seq demonstrated that NANOG represses OXPHOS genes and induces fatty acid oxidation genes to maintain TICs35. To specifically identify NOTCH1 target genes, ChIP-seq was performed on CD133(+) and CD133(−) TICs with an anti-NOTCH1 antibody. Gene Ontology (GO) analysis of the data from this study indicated that CD133(+) TICs presented altered expression of genes encoding proteins involved in mitochondrial metabolism-related pathways, such as proteins related to the mitochondrial proton-transporting ATP synthase complex, mitochondrial structure and morphology (Supplementary Tables 24 and Fig. 1a). These results demonstrated that TBC1D15 and NOTCH1 may play important roles in the regulation of mitochondrial metabolic pathways in CD133(+) TICs.

Fig. 1: The TBC1D15-NICD interaction results in the recruitment of mitochondria to the perinuclear region with an increase in the mitochondrial diameter and regulates mitochondrial gene expression.
figure 1

a ChIP-seq analysis of TBC1D15 with an anti-NOTCH1 antibody in CD133(+) and CD133(−) TICs. (Left, black bar graphs) Analysis of downregulated target genes identified by ChIP-seq with the anti-NOTCH1 antibody in CD133(+) TICs vs. CD133(−) non-TICs using Gene Ontology (GO) cellular component terms via the Enrichr web server ( The bar graphs show statistical significance, as indicated by the p values (white bar graphs). All the genes that were significantly downregulated according to ChIP-seq with the anti-NOTCH1 antibody in TBC1D15 KO vs. WT CD133(+) TICs and in CD133(+) vs. CD133(−) TICs were used for the analysis. b Western blot analysis of the cytoplasmic, nuclear, and mitochondrial fractions from NICD mutant-transfected CD133(+) TICs. c Co-IP–Western blot analysis of the cytoplasmic, nuclear, and mitochondrial fractions from NICD mutant-transfected CD133(+) TICs. d Confocal images of TOM20 (red, outer mitochondrial membrane of mitochondria marker), FIS1 (green), TBC1D15 (green) and DAPI (nucleus) localization in TICs. Scale bar = 7.2 µm. e Quantification of the number of mitochondria in (d) confocal images. The number of mitochondria was calculated as the mean ± SD (n = 450). p values were calculated by a two-tailed unpaired t test. f Graphs of mitochondrial diameters and mitochondrial-nuclear distances. The mitochondrial diameters are presented as the means ± SDs (n = 400). p values were calculated by a two-tailed unpaired t test. g Graph of the distance of mitochondria from the nucleus. The distances of mitochondria from the nucleus was calculated as the mean ± SD (n = 400). p values were calculated by a two-tailed unpaired t test. h, i qRT‒PCR analysis of mitochondrial metabolism-related genes (SLC25A16, FIS1, ATP5B and MRPS10) in the indicated cells. The data are presented as the means ± SDs (n = 3). p values were calculated by a two-tailed unpaired t test.

We next investigated the interactions of TBC1D15 with the NICD PEST domain and mitochondria in CD133(+) TICs. TBC1D15 and NICD were present in the mitochondrial fraction of CD133(+) TICs transfected with NICD mutants (Fig. 1b). In addition, the colocalization of TBC1D15 and NICD was decreased in the mitochondrial fraction of TICs expressing the NICD deletion mutant, as indicated from immunoprecipitation of this cell fraction using the anti-TBC1D15 antibody (Fig. 1c). Furthermore, the endogenous NICD was not displaced from the mitochondrial fraction by expression of the deletion mutant. This demonstrated that the full-length NICD was required for the observed mitochondrial localization.

Next, we investigated the mitochondrial recruitment of endogenous TBC1D15 to the NICD PEST domain. Surprisingly, confocal imaging of the mitochondrial markers TOM20 and TBC1D15 revealed that, compared to non-TICs, the mitochondrial sizes increased in CD133(+) TICs; in addition, the subcellular distribution of mitochondria was noticeably altered, as was the localization of mitochondria to the perinuclear region (Fig. 1d). The same groups of cells were stained with antibodies against TBC1D15 and TOM20 and with DAPI to examine whether TBC1D15 colocalizes with mitochondria and/or nuclei. In CD133(+) TICs, both the numbers and sizes of mitochondria increased, and the mitochondria exhibited pronounced perinuclear localization (Fig. 1f, g). However, in CD133(+) TICs expressing the NICD-PEST deletion mutant, the number of mitochondria was reduced compared to that in NICD-FL expressing cells (Fig. 1e, g). CD133(−) TICs contained fewer mitochondria, which were more evenly distributed throughout the cytoplasm and were smaller in size than mitochondria in CD133(+) TICs. These data showed that CD133(+) TICs were permissive for the association of TBC1D15 with NICD-PEST, leading to their mitochondrial association and a change in the morphology and intracellular localization of mitochondria.

The expression levels of mitochondrial metabolism-related genes (SLC25A16, ATP5B, and MRPS10) decreased in CD133(+) TICs expressing the NICD-PEST deletion mutant compared to those expressing NICD-FL (Fig. 1h, i). These results suggested that TBC1D15 promotes the self-renewal of liver TICs by increasing the mitochondrial volume and altering the expression of mitochondria-related genes. These changes are associated with the recruitment of mitochondria to the perinuclear region. Further studies are warranted to clearly define the causal relationships.

Target genes identified by NOTCH1 ChIP-seq were positively or negatively regulated by TBC1D15 through activated Notch signaling

After TBC1D15 was knocked down in Huh 7 cells, many genes exhibited significant changes in expression (Supplementary Fig. 1a). Among these genes, PD-L1 showed the most significant upregulation, and NFKBIA (IκB), RXRA (RXRα), CHUK (IKKα), and AKT2 (AKT) also showed significant upregulation. HNF4A (HNF4α) showed significant downregulation, which was confirmed by immunoblot analysis (Supplementary Fig. 1b). After knocking down TBC1D15, the protein expression of PD-L1 exhibited the same pattern as that observed for the Notch1 intracellular domain (Supplementary Fig. 1c). Treatment with the Notch signaling activator oxaliplatin (0.75 µM and 1.5 μM) reactivated Notch by facilitating its cleavage to generate the active form of NOTCH (i.e., the NOTCH intracellular domain, NICD). After treatment with the Notch signaling activator oxaliplatin, the mRNA level of TBC1D15 was unchanged (Supplementary Fig. 1a). In contrast, after reactivating Notch by treatment with the NOTCH activator oxaliplatin to generate NICD, the initial mRNA expression level of PD-L1 was restored (Supplementary Fig. 1d). TBC1D15 knockdown reduced the PD-L1 mRNA level. These results suggest that TBC1D15-mediated activation of NOTCH transactivates the PD-L1 gene. These results imply that TBC1D15 regulates PD-L1 induction by activating the Notch signaling pathway. The expression of NFKBIA (IκB), CHUK (IKKα), and AKT2 (AKT) was also restored by reactivating the Notch signaling pathway, suggesting that they are also downstream of NOTCH signaling (Supplementary Fig. 1d). There was almost no difference in the expression of RXRA (RXRα) between the treated group and the control group (Supplementary Fig. 1d). HNF4A (HNF4α) was downregulated by TBC1D15 knockdown but was still upregulated after Notch activation. These results indicated that oxaliplatin-mediated Notch cleavage generates the active form of NOTCH (NICD) to induce NFKBIA (IκB), CHUK (IKKα), and AKT2 (AKT) but repress the hepatocyte differentiation factor HNF4A (Supplementary Fig. 1d).

TBC1D15 preferentially interacted with NUMB isoform 5 and the NOTCH1 PEST domain in CD133(+) TICs

Our laboratory previously showed that TBC1D15 interacts with the NOTCH1 PEST domain9. To determine whether TBC1D15 stabilizes NOTCH1, CD133(+) TICs were lentivirally transduced with shRNA against TBC1D15 (Supplementary Fig. 4c, Left). Co-IP followed by Western blot analysis demonstrated that expression of the NICD PEST deletion mutant reduced the interaction between TBC1D15 and NOTCH1 in TICs (Supplementary Fig. 4c, Middle). By restoring the expression of TBC1D15 in shTBC1D15-CD133(+) TICs, the interaction between TBC1D15 and the NICD PEST domain was restored (Supplementary Fig. 4c, Right).

To determine whether TBC1D15 regulates the expression and phosphorylation of NUMB splice variants, TBC1D15 was knocked down in CD133(+) TICs and overexpressed in CD133(−) TICs (Fig. 2a). TBC1D15 knockdown reduced NUMB phosphorylation in both mock-transfected CD133(+) TICs. Similarly, both the expression and phosphorylation of NUMB were reduced in mock-transfected CD133(−) TICs and unaffected by TBC1D15 overexpression (Fig. 2a). These findings indicated that TBC1D15 promoted NUMB5/6 expression and phosphorylation in CD133(+)TICs but not in CD133(−) non-TICs.

Fig. 2: TBC1D15 interacts with NUMB isoform 5, which is localized to mitochondria, and with full-length NUMB, which is localized in the cytoplasm.
figure 2

a Co-IP-western blot analysis of NUMB phosphorylation and expression in TBC1D15-KD CD133(+) TICs and TBC1D15-overexpressing CD133(−) TICs. b Upper panel, Schematic of NUMB isoform domain constructs. The red bar(s) are the sites of serine phosphorylation. Bottom panel, Cell-free in vitro binding assay of the identified interaction domains of full-length NUMB, NUMB isoforms 5 and 6 and TBC1D15. Ponceau staining was used as a loading control. NUMB isoforms 5 and 6 lack three serine phosphorylation sites that are critical for its membrane localization. c Co-IP–Western blot analysis of the identified interaction of NUMB isoforms and TBC1D15 in the indicated cells with full-length NUMB or the isoforms 5 and 6. d Co-IP–Western blot analysis of interactions of TBC1D15 (Flag-tag) with full-length NUMB, isoform 5 (GFP-tagged), and the PEST domain of NICD (MYC-tagged) in CD133(+) TICs. e Confocal images of TBC1D15 (endogenous, red) and GFP-tagged NUMB (green) and DAPI (nucleus, blue) localization in TICs. Scale bar: 7.2 µm. The graphs show the mean ± S.D. of three independent experiments. f Confocal images of TOM20 (red, outer mitochondrial membrane marker), GFP-tagged NUMB (green) and DAPI (nucleus) localization in TICs. Scale bar, 6.14 µm. Bottom: Quantification of TBC1D15/NUMB and NUMB/TOM20 localization. The graphs show the mean ± S.D. of three independent experiments.

Alternative splicing isoforms of NUMB are expressed in various cancers36,37. The corresponding alternatively spliced mRNAs are translated into NUMB isoforms with molecular weights of approximately 74 kDa (NUMB isoforms 1 and 3), 72 kDa (NUMB isoforms 2 and 4), and 50 kDa (NUMB isoforms 5 and 6)24,38. The expression of NUMB isoforms as a function of TBC1D15 expression was examined. An array of mRNAs encoding NUMB isoforms was examined by semiquantitative RT‒PCR in CD133(+) and CD133(−) TICs with or without TBC1D15 overexpression (Supplementary Fig. 5b). In CD133(+) TICs, the expression levels of all NUMB mRNA isoforms were examined by semiquantitative RT‒PCR. The mRNA expression levels corresponding to NUMB isoforms 1, 3, 5 and 6 in CD133(−) TICs were lower than those in CD133(+) TICs. Overexpression of TBC1D15 resulted in higher expression levels of NUMB isoforms 1, 3, 5 and 6 than of NUMB isoforms 2 and 4 in CD133(−) non-TICs.

The differences in NUMB isoform expression led us to investigate the interactions of TBC1D15-NICD with the NUMB protein isoforms in CD133(+) TICs. When the NICD PEST domain was deleted in TBC1D15 KD CD133(+) TICs, alternative NUMB isoform expression was observed (Fig. 2a and Supplementary Fig. 5a, b). The NUMB isoform that was expressed in these cells was a 50 kDa isoform that we ascribed to isoform 5 or 6, and its expression was dependent on TBC1D15 (Fig. 2a).

To confirm whether NUMB isoform 5 binds TBC1D15, we performed cell-free in vitro binding assays and co-IP followed by Western blot analysis with CD133(+) and CD133(−) TICs with overexpression of full-length NUMB or NUMB isoforms 5 and 6 (Fig. 2b, c). Recombinant GST-tagged NUMB proteins (NUMB full-length, NUMB-PTB, NUMB-PRR1, NUMB-PRR2, and NUMB-PRR3) were expressed in and purified from E. coli BL21. Flag-tagged TBC1D15 was expressed in and purified from HEK-293A cells. Purified GST-NUMB proteins were immobilized onto glutathione-Sepharose 4B beads, which were coincubated with the Flag-tagged full-length TBC1D15 protein. The resulting complexes were eluted from the beads for PAGE and subjected to immunoblot analysis with an anti-Flag or anti-TBC1D15 antibody. The Flag-tagged TBC1D15 protein was found to strongly interact with NUMB isoform 5 in vitro (red arrowhead) (Fig. 2b). In vitro binding assays were used to compare the binding of full-length NUMB, NUMB isoform 5, and NUMB isoform 6 to TBC1D15 and revealed that NUMB isoform 5 interacted with TBC1D15 in CD133(+) TICs. TBC1D15 binding induced the phosphorylation of the full-length NUMB and NUMB isoform 5. Furthermore, we tested whether a trimeric TBC1D15-NUMB isoform 5-NICD complex was formed in CD133(+) TICs (Fig. 2d). Collectively, the results indicated that the interaction between TBC1D15 and the PEST domain of NICD induced the phosphorylation of NUMB isoform 5 via the TBC1D15 interaction (possibly through aPKCζ, which was previously reported in our paper)39.

NUMB isoform 5 colocalized to the cytoplasm and mitochondria with TBC1D15

To investigate the subcellular localization of TBC1D15, the localization of NUMB isoform 5 and full-length NUMB in CD133(+) TICs was examined via immunofluorescence staining. Mammalian NUMB localizes to the basolateral membrane of polarized epithelial cells40. Furthermore, NUMB depletion accelerates mitochondrial fragmentation by promoting the phosphorylation of DRP1, which in turn leads to mitochondrial dysfunction and apoptosis41. Endogenous TBC1D15 exhibited punctate staining throughout the cytoplasm with accumulation at the cell membrane in CD133(+) (Fig. 2e) and CD133(−) TICs (Supplementary Fig. 6). Under the experimental condition, GFP-tagged full-length NUMB localized to the inner cell membrane. GFP-tagged NUMB isoform 5 (NUMB5) exhibited coimmunostaining with endogenous TBC1D15 in the cytoplasm and cell membrane (Fig. 2e, f, bottom panel). In contrast, this localization pattern was rarely observed for GFP-tagged NUMB isoform 6 (Fig. 2e, bottom). NUMB5 lacks the three sites for aPKC-mediated serine phosphorylation that are required for its membrane localization due to alternative mRNA splicing events42.

As a GTPase-activating protein for RAB7A, TBC1D15 exists both free in the cytoplasm and in association with mitochondria through a direct interaction with FIS143,44. We next investigated whether TBC1D15 interacts with NUMB isoform 5 to recruit TBC1D15 to mitochondria. Indeed, TBC1D15 and NUMB isoform 5 were observed to colocalize with a mitochondrial marker (TOM20) (Fig. 2f). In addition, NICD overexpression resulted in recruitment of the TBC1D15-NUMB isoform 5 complex to mitochondria (Fig. 2f).

TBC1D15 enhanced NOTCH1 protein activation and stabilization in CD133(+) TICs

To determine the importance of TBC1D15-mediated NOTCH stabilization and activation in many cancer types, an extensive meta-analysis of TBC1D15 expression in tumors arising from diverse tissue types was performed. The TCGA database revealed poorer survival rates with higher levels of TBC1D15 mRNA expression in liver, lung, breast, and ovarian cancer patients (Supplementary Fig. 2a, Fig. 3a). TBC1D15 mRNA expression is elevated in cholangiocarcinoma (CHOL), while TBC1D15 mRNA expression is not significantly elevated in HCC (Supplementary Fig. 2a), indicating that posttranscriptional mechanisms are involved in increasing NOTCH1 protein expression in various tumor types, including cholangiocarcinoma, glioblastoma, acute myeloid leukemia, and ovarian cancer (Supplementary Fig. 2a, b). Immunoblot analysis confirmed that NOTCH1 and TBC1D15 expression is elevated in HCC tissues compared to noncancerous tissues (Supplementary Fig. 2b). To confirm whether CD133(+) TICs express a higher level of the TBC1D15 protein, CD133(+)/CD49f(+) TICs were isolated from lung cancer cells (Calu3 and H1299) using magnetic affinity microbeads (Supplementary Fig. 2c). The TBC1D15 protein levels were confirmed to be higher in CD133(+)/CD49f(+) TICs than in CD133(−)/CD49f(−) TICs, indicating that TBC1D15 indeed associated with TIC characteristics, such as NOTCH-mediated stemness induction (Supplementary Fig. 2d).

Fig. 3: Coexpression of TBC1D15 and NOTCH1 promotes oncogenesis, and TBC1D15 interacts with the NOTCH1-PEST/STR domains.
figure 3

a Kaplan-Meier analysis of liver, lung, breast, and ovarian cancer patients with primary tumors expressing high or low levels of TBC1D15. p values were calculated via a stratified Cox proportional hazards model. b Immunoblot analyses confirmed that NOTCH1 and TBC1D15 expression is elevated in HCC tissues compared to noncancerous tissue regions. c Representative immunohistochemical images showing TBC1D15, NOTCH1, and NUMB localization in human normal liver tissues and hepatocellular carcinoma tissues. Images were acquired, and the mean fluorescence intensity was quantified using ImageJ software. The fluorescence intensity percentage relative to the control is shown and expressed as the mean ± SD (n = 3). Scale bars, 50 µm. The insets show the 10X images at 40X magnification. d Schematic illustration of tumor incidence and tumor growth in Alb::CreERT2;Tbc1d15FL/FL;NS5A and Tbc1d15FL/FL;NS5A mice. e The tumor incidence was reduced in mice with hepatocyte-specific TBC1D15 deficiency. Representative livers are shown. The error bars indicate the SEMs. p values calculated by a chi-square test are shown. *p = 0.001. f The tumor mass/total liver mass ratio (%) decreased in mice with hepatocyte-specific TBC1D15 deficiency. The error bars indicate the SEMs. p values calculated by a chi-square test are shown. *p = 0.001.

We next conducted a coexpression analysis of the TBC1D15 protein using the Oncomine database ( The levels of NOTCH1, NOTCH2, NOTCH3, NOTCH4, and TBC1D15 were compared among 115 normal liver tissue samples and 95 HCC tissue samples (Supplementary Fig. 3a). This analysis revealed a significant correlation between NOTCH gene expression and TBC1D15 expression (Supplementary Fig. 3b). To validate these findings, we performed immunoblotting and immunohistochemical (IHC) staining of TBC1D15, NOTCH, and NUMB in HCC tissues and matched noncancerous liver tissues (clinicopathological factors are listed in Supplementary Table 1) (Fig. 3b, c). The results showed that the fluorescence intensities of TBC1D15, NOTCH1, and NUMB were significantly increased by >80% compared to those in normal tissue (p < 0.05) (Fig. 3c, Right). Moreover, the significant increases in NOTCH and NUMB expression appeared to correlate with HCC and TBC1D15 expression (Fig. 3c).

Double or triple knockout of Tbc1d15 and Notch1/2 reduced the tumor incidence in HCV-NS5A transgenic mice fed an alcohol-containing Western diet for one year

The TBC1D15 protein is required for alcohol-containing WD-mediated hepatocarcinogenesis. The existence of cooperative interactions with NOTCH pathways and specific isoforms of NOTCH that are critical for alcohol-containing WD-mediated hepatocarcinogenesis has not been shown. To address this knowledge gap, we generated NS5A Tg mice harboring a floxed Tbc1d15 allele for tamoxifen-inducible, hepatocyte-specific knockout of this gene (Alb::CreERT;Tbc1d15f/f;NS5A mice) (Fig. 3d). The liver tumor incidence in wild-type Tbc1d15f/f;NS5A mice after 12 months of alcohol-containing Western diet feeding was 55%, which was reduced significantly to 17% in TBC1D15-deficient Alb::CreERT;Tbc1d15f/f;NA5A mice and was almost completely eliminated by double or triple knockout of Tbc1d15 and Notch1/2. We considered knockout of Tbc1d15 accompanied by knockout of both Notch1 and Notch2 to be a triple knockout (Fig. 3e). To exclude any impact of the anticancer effects of tamoxifen, we treated both the control and experimental groups with tamoxifen, and the results revealed no significant difference in hepatocarcinogenesis. These results supported the importance of the TBC1D15-NOTCH protein complex as a master regulator of hepatocarcinogenesis induced by alcohol-containing Western diet feeding and HCV transgene expression. The incidence of liver tumors (left) in the four genotype groups of mice is shown in Fig. 3e, along with representative examples of the corresponding tumors (Fig. 3e). Knockout of TBC1D15, Notch1, or Notch2 alone or in combination was observed to reduce the tumor masses in these animals (Fig. 3f).

TBC1D15 promoted EDTA-mediated NOTCH1 cleavage

To determine how TBC1D15 activates NOTCH1 cleavage, we tested whether TBC1D15 processes NOTCH1 through a combination of a disintegrin and metalloproteinase (ADAM17) and γ-secretase. For this analysis, we overexpressed NOTCH1ΔE-FL, NOTCH1ΔE-ΔPEST, and NOTCH1ΔE-V1754L mutants overexpressed in TBC1D15High CD133(+) TICs and TBC1D15Low CD133(−) TICs and compared the effects of TBC1D15 on endogenous NOTCH1 processing by measuring the amount of NICD produced after EDTA treatment and the degree of inhibition by DAPT treatment in vitro. EDTA treatment triggers NOTCH1 cleavage via γ-secretase, whereas DAPT is a γ-secretase inhibitor (GSI)45 (Fig. 4a). The NOTCH1 derivative NOTCH1ΔE was constitutively cleaved by γ-secretase, which released the active NICD, and was unaffected by DAPT (Fig. 4a). DAPT treatment reduced the amount of the γ-secretase-cleaved form of NOTCH1 (S3) in cells expressing NOTCH1ΔE-ΔPEST (Fig. 4b), indicating that PEST domain deletion reduced the stabilizing effect of TBC1D15 and that γ-secretase inhibition markedly increased the level of transfected Myc-tagged NOTCH1. These results indicated that TBC1D15 interacts with the NOTCH1 PEST domain to induce NOTCH1 processing, which was inhibited by expression of the cleavage site mutant NOTCH1 V1754I (Fig. 4c). Treatment with DAPT inhibited EDTA-mediated S3 cleavage, while the endogenous NOTCH1 protein level did not significantly change (Fig. 4). In the absence of EDTA-mediated S2 cleavage (endogenous level of S3 cleaved NOTCH1), DAPT treatment did not significantly restore the protein level of the uncleavable form, indicating that TBC1D15 induced NOTCH1 processing via the S2 cleavage followed by the S3 cleavage reaction. These results implied that TBC1D15 mediated inhibition of the interaction of PIN1 with the STR domain and inhibited the EDTA-mediated cascade of S3 cleavage reactions.

Fig. 4: TBC1D15 binds phospho-S2142 in the NOTCH1 STR domain and constitutively activates NOTCH1 through S3 cleavage.
figure 4

a Mutations in Val1754 (V1754K or V1754L) in humans impede the intracellular cleavage of NOTCH1, leaving the NOTCH1 protein tethered to the membrane (membrane-tethered Notch1) and reducing NOTCH1 downstream signaling activity58,59. To mimic the inactivation of NOTCH1 via mutation, we transfected Huh7 cells with wild-type membrane-tethered NOTCH1 vectors containing full-length NOTCH1 cDNA or vectors containing mutant NOTCH1 cDNA (NOTCH1V1754L). b TBC1D15 binds to the NOTCH1-PIN1 interaction domain (STR) and overrides γ-secretase inhibitor-mediated inhibition of NOTCH1 S3 cleavage induced by EDTA. Western blot analysis of the expression of endogenous NOTCH1 and the N1ΔE–myc tag mutant in CD133-positive and CD133-negative TICs transfected with N1ΔE-FL or N1ΔE-ΔPEST and treated with vehicle, 20 μM DAPT (for 4 h), or 5 mM EDTA (for 20 min). Western blot analysis of endogenous NOTCH1 and the N1ΔE–myc tag mutant in CD133-positive and CD133-negative TICs transfected with N1ΔE-FL, N1ΔE-ΔPEST, or N1ΔE-V1754L (constitutive inhibition of γ-secretase-mediated cleavage) and treated with vehicle or 20 μM DAPT (for 4 h). c Western blot analysis of the expression of endogenous NOTCH1 and the NOTCH1ΔE-myc tag mutant in TBC1D15High CD133(+) TICs and TBC1D15Low CD133(−) non-TICs transfected with NOTCH1ΔE-FL, NOTCH1ΔE-ΔPEST, or NOTCH1ΔE-V1754L (constitutive inhibition of gamma-secretase-mediated cleavage) and treated with vehicle or 20 μM DAPT. DAPT treatment restored the NOTCH1 protein level in NΔE-ΔPEST-expressing cells, indicating that PEST domain deletion reduced the stabilization of TBC1D15 and that γ-secretase inhibition markedly restored the level of transfected Myc-tagged NOTCH1. The NOTCH1 derivative [NOTCH1ΔE] was constitutively cleaved by γ-secretase, which releases the active NICD, and was not affected by DAPT or EDTA. d Hypothetical model of the mechanism by which TBC1D15 inhibits NICD phosphorylation in CD133(+) TICs. e (Left) Schematic diagram of truncation mutants of NICD1. (Right) Co-IP–western blot analysis of the identified interaction domains of NICD and TBC1D15 in CD133(+) TICs. f TBC1D15 binds phospho-S2142 in the NOTCH1-PIN1 interaction domain (serine/threonine region: STR). PIN1 is a peptidyl-prolyl cistrans isomerase (PPIase) that recognizes and isomerizes the phosphorylated serine/threonine-proline (pSer/Thr-Pro) motif. The structural changes mediated by the prolyl isomerase PIN1 posttranslationally regulate the biological functions of NOTCH146,47,60. Co-IP–Western blot analysis of the identified interactions between the STR domain and PIN1 in CD133(+) TICs with or without PIN1 overexpression. g Luciferase assay of CD133(+) TICs cotransfected with HEY-1/LUC and the NΔE-FL-myc expression vector (pcDNA3-NΔE–FL-myc) along with the expression vector for PIN1-HA (pcDNA3-PIN1-HA). (Top) The transcriptional activity of NΔE-FL-myc was tested with HEY-1/LUC in CD133(+) TICs in the presence of either shTBC1D15 or sh-scrambled. (Bottom) The immunoblot is shown. The graph shows the mean ± S.D. of three independent experiments. Cell lysates were analyzed by western blotting, and the analysis results are shown below the graphs.

Mutation of the phosphorylation sites in the STR or PEST domains of NOTCH1 reduced the TBC1D15-NOTCH1 interaction

To mechanistically demonstrate how TBC1D15 activates the NICD pathway, a protein domain mapping study was performed by generating deletion mutants of NOTCH and NICD, including PEST domain mutants (Fig. 4e), and introducing a substitution mutation at T2512 (see Supplementary Fig. 4a) to eliminate the phosphorylation site (Supplementary Fig. 4b, c). NUMB isoforms 5 and 6 lack three serine phosphorylation sites that are critical for membrane localization. Several NICD deletion mutants were constructed and overexpressed in CD133(+) and CD133(−) TICs, after which the expression of these genes was confirmed via IP–western blotting (Supplementary Fig. 4b–d). Deletion or absence of the PEST and STR domains of NICD reduced its interaction with TBC1D15, which was further diminished by mutation of the STR domain phosphorylation sites (3 M and 4 M) (Fig. 4d, Right). Substitution of the Ser/Thr-Pro residues Ser 2122, Thr 2133 and Ser 2137 with Ala (NICD-3M) and Ser 2142 with Ala (NICD-4M) hampered the increase in processing by γ-secretase dependent on the prolyl isomerase PIN1. These results showed that TBC1D15 binds both the PEST domain and the STR domain of NOTCH and NICD.

The NOTCH STR domain has four Ser/Thr-Pro motifs that are phosphorylation sites. The prolyl isomerase PIN1 binds to the NOTCH STR domain and induces a conformational change through prolyl isomerase-mediated, NOTCH1 cleavage by γ-secretase46,47. Proline (Pro)-directed serine/threonine (Ser/Thr) phosphorylation through peptidyl-prolyl cistrans isomerases (PPIases) transforms peptidyl-prolyl structures to regulate the folding, subcellular location, stability, activation, and interaction of multiple proteins48. Therefore, prolyl cis–trans isomerization serves as a molecular timer to amplify cellular processes48. Overexpression of the prolyl isomerase PIN1 in CD133(+) TICs increased the binding of this protein to NICD and NICD-3M, with a corresponding decrease in the binding of TBC1D15 to NICD or the STR domain (Fig. 4e). Co-IP and Western blot analyses demonstrated that PIN1 overexpression in CD133(+) TICs attenuated the interaction between Myc-tagged TBC1D15 and HA-tagged NICD (Fig. 4f). Furthermore, the binding of TBC1D15 to the NICD STR domain was reduced due to a conformational change in NICD induced upon PIN1 binding. These results indicated that endogenously expressed prolyl isomerase PIN1 and TBC1D15 in CD133(+) TICs cooperatively support cleavage by γ-secretase to activate NOTCH, while overexpressed PIN1 outcompeted TBC1D15 for binding to NICD via the NICD STR domain.

To test whether TBC1D15 activates the NOTCH1 pathway, we analyzed the induction of NOTCH1 targeting. For assessment of HEY-1 promoter activity via a luciferase reporter assay, we overexpressed a membrane-tethered, ligand-independent, wild-type NOTCH1 protein and NOTCH1ΔE-FL, a derivative that is constitutively cleaved by γ-secretase, thus releasing the active NICD, in CD133(+) TICs46. The luciferase reporter assay showed that TBC1D15 increased NOTCH1ΔE-FL activity in CD133(+) TICs expressing endogenous TBC1D15 (Fig. 4g). Moreover, TBC1D15 ablation significantly reduced HEY-1 promoter activation by NOTCH1ΔE-FL. These results indicated that TBC1D15 interacts with human NOTCH1 and regulates the downstream pathway.

TBC1D15 binds to T2512 of the NICD PEST domain and blocks the phosphorylation of NICD by CDK8, thereby stabilizing NICD

As TBC1D15 expression promotes NICD production, NOTCH1 activation is regulated by phosphorylation and ubiquitination mediated by CDK849 and the E3-ubiquitin ligase FBW750, respectively. The NOTCH-related molecule MASTERMIND promotes recruitment of the cyclin C-CDK8 complex and hyperphosphorylation of the PEST domain of NOTCH1.

We hypothesized that cyclin-mediated phosphorylation of the NOTCH PEST domain occurs following the formation of the CYCLIN C-CDK8–TBC1D15 complex. We investigated the interaction between CDK8 and TBC1D15 (Fig. 5a). Coimmunoprecipitation and immunoblot analysis of CDK8 and TBC1D15 revealed that these proteins strongly interacted with the PEST domain of NOTCH1. Conversely, overexpression of CDK8 suppressed the interaction between NICD and TBC1D15 (Fig. 5a). Our prior studies demonstrated that TBC1D15 expression promoted NOTCH1 expression39. These observations indicated that TBC1D15 may control the NICD level by blocking the phosphorylation of the NOTCH PEST domain by its interaction partners CDK8 and CDK19. The functional significance of the specific phosphorylation sites within the PEST domain was examined in NICD mutants harboring dominant-negative Ala substitutions (S2490A, S2493A, S2500A, T2512A, and S2514/2517 A) or phosphomimetic T2512E or S2514E substitutions (Fig. 5b). These recombinant proteins were expressed, purified, and used as substrates for in vitro protein kinase assays with cell lysates. We found that the TBC1D15–CDK8 complex could not phosphorylate the NICD. Expression of the dominant-negative or phosphomimetic NOTCH1 mutants caused a lack of function of CDK8 complexes that were outcompeted by TBC1D15. These results suggested that TBC1D15 binds to T2512 of the NICD PEST domain to block phosphorylation of NICD by the CYCLIN C-CDK8 complex, thereby stabilizing NICD (Fig. 5c).

Fig. 5: TBC1D15 sequesters the NOTCH1 PEST domain to inhibit the CDK8-mediated T2512 phosphorylation of the NOTCH1 PEST domain, which is needed for recruitment of the E3 ligase FBW7 and NOTCH1 ubiquitination.
figure 5

a CD133(+) and CD133(-) TICs transfected with myc-NICD-FL or NICD deletion mutants (∆PEST and ∆2171) were analyzed for the expression of the TBC1D15, CDK8 and CDK19 proteins by co-IP–western blotting. b CD133(+) TICs cotransfected with HA-CDK8, myc-NICD-FL, or NICD deletion mutants (∆PEST and ∆2171) were analyzed for the expression of the TBC1D15, CDK8 and CDK19 proteins by co-IP–western blotting. c Top-Representative phosphorylation sites for mutations in the PEST domain of NICD. Bottom-Co-IP–Western blot analysis of the interactions of TBC1D15 with PEST domain mutants in CD133(+) and CD133(-) TICs. d Co-IP–Western blot analysis of the interactions of FBW7 with T2512/S2514 mutants of the PEST domain in CD133(+) TICs. e (Top) Representative images showing the sequence alignment of the NOTCH1, CYCLIN E, c-MYC, and c-JUN FBW7 phosphodegrons. (Bottom) Co-IP–Western blot analysis of the interactions between c-JUN, c-MYC and mutants of the PEST domain in CD133(+) TICs. f Co-IP–Western blot analysis of the interactions between FBW7, c-JUN and mutants of c-JUN in CD133(+) and CD133(-) TICs. g Hypothetical model of the mechanisms by which TBC1D15 interacts with NOTCH to stabilize NOTCH1 in CD133(+) TICs and TBC1D15 inhibits NICD phosphorylation in CD133(+) TICs. h Ubiquitination assay of the anti-MYC immunoprecipitate in CD133(+) TICs treated with the proteasome inhibitor MG132 after transfection with the indicated plasmids. The bar graphs show the mean ± S.D. of three independent experiments.

The T2512 residue of the NICD PEST domain comprises an FBW7 phosphodegron

Phosphorylation converts the NICD PEST domain into a substrate that is recognized by FBW7, which binds to NICD, promoting its polyubiquitination and proteasomal degradation by recruiting the components of the SCF ubiquitin ligase degradation complex51.

We overexpressed Flag-FBW7 (Fig. 5d) to determine whether the TBC1D15-NICD PEST interaction attenuates interactions between FBW7 and the NICD PEST Cdc phosphodegron (CPD), especially those mediated via the T2512 and S2514 residues. We then examined the roles of these phosphorylation sites in the binding interactions between NICD and FBW7. We found that the T2512A and S2514A NICD mutants exhibited decreased binding to FBW7 compared to that of NICD-FL (Fig. 5d). In addition, FBW7 binding was restored when T2512A and S2514A were replaced by phosphomimetic residues (T2512E and S2514E) (Fig. 5d).

The CPD consensus motif is centered at approximately T2512 in human NOTCH1 and is highly conserved across all species of the NOTCH family52,53. These CPD consensus motifs are present in c-JUN, c-MYC and CYCLIN E54,55 (Fig. 5e). As a positive control for FBW7 binding to other CPD sites, we tested whether c-JUN and c-MYC contain a homologous NOTCH1-like CPD that would support TBC1D15 binding. We performed an IP assay using an anti-FBW7 antibody after incubation with a peptide corresponding to c-JUN residues 229–253 (Fig. 5f). This sequence region spans the putative FBW7 binding site of c-JUN and is dependent on phosphorylation of both Thr239 and Ser243. Wild-type c-Jun coprecipitated with FBW7, whereas expression of the double amino acid substitution T239A/S241F mutant reduced the amount of c-JUN binding to FBW7 (Fig. 5f, g). This result suggests that the interaction between FBW7 and c-Jun is dependent on access to the phosphorylation sites of c-JUN.

We next examined whether T2512 of NICD directs endogenous FBW7-mediated NICD ubiquitination. In accordance with our binding studies, the T2512A mutation reduced NICD ubiquitination (Fig. 5h).

Identification of small molecule inhibitors that block interactions between TBC1D15 and NICD CDP phosphodegron sequences

Antagonists of TBC1D15 interactions with NICD are expected to decrease tumor incidence. A small molecule inhibitor library was screened to identify potential antagonists. For this assay, a FITC-tagged NOTCH1-TBC1D15 binding domain peptide (wild type) and a Y > A mutant of this peptide were synthesized for fluorescence polarization assays with recombinant TBC1D15 in the presence of individual library compounds (Fig. 6a, b). The mutant peptide, which does not interact with TBC1D15, served as a negative control. Inhibitor screening revealed that most of the compounds tested exhibited toxicity in both CD133(+) and CD133(−) TICs (R2 = 0.80), except for a subgroup of chemicals that exhibited selective toxicity toward CD133(+) TICs (Fig. 6c). By merging the activity rankings of the tested chemicals determined via these screening methods, the best compounds were identified and included the inhibitor propan-2-yl (Z)-7-[(1 R,2 R,3 R,5 S)-3,5-dihydroxy-2-[(3 R)-3-hydroxy-5-phenylpentyl]cyclopentyl]hept-5-enoate (Inhibitor A; Fig. 6a). Indeed, Inhibitor A effectively blocked the NOTCH-TBC1D15 interaction, inhibited HEY1 and NANOG promoter activity, and selectively killed CD133-positive TICs (Fig. 6d).

Fig. 6: A small molecule inhibitor of the NOTCH1-TBC1D15 interaction inhibited PDX tumor growth.
figure 6

a (Left) Analysis of the 3D structure of the TBC1D15-NOTCH1 interaction identified the docking of the TBC1D15 N-terminal region (green) with aa1805-2113 of NOTCH1 (blue). Inhibitor A efficiently binds the interaction domains for both the TBC1D15-NOTCH and TBC1D15-NUMB interactions. (Right) Structure of Inhibitor A. b A diagram depicting the screening scheme for selective inhibitors of the TBC1D15-NOTCH interaction using the wild-type NOTCH1 PEST domain peptide and its mutant. Both were labeled with FITC and used with recombinant TBC1D15 for a fluorescence polarization assay. The mutant served as a negative control. A diagram depicting the results of the fluorescence polarization assay (FPA) with the wild-type (WT) and mutant NOTCH1 PEST domain peptides and TBC1D15. FPA signal induced by the FITC-labeled wild-type and mutant PEST domain peptides with TBC1D15 as a function of the peptide concentration (Right top); competitive inhibition of the signal with the unlabeled WT peptide and the mutant peptide (Right bottom). c NOTCH1-TBC1D15 inhibitor characterization. Inhibitor A blocked the NOTCH1/3-TBC1D15 interactions (Left) and selectively killed CD133(+) Huh7 cells without killing CD133(−) Huh7 cells or primary hepatocytes. Inhibitor A effectively abrogates the interaction of TBC1D15 with NICD1, as shown by IP-IB. Effects of Inhibitor A on the NICD1-TBC1D15 interaction: (Left) Inhibitor A blocked the interaction between TBC1D15 and NICD1. (Right) Inhibitor A inhibited TIC self-renewal and tumorigenicity. Inhibitor A suppressed TIC self-renewal, as determined by a spheroid formation assay. d Inhibitor A treatment abrogated NOTCH-CFP and NANOG-GFP reporter activity (bottom). **p < 0.01. e Inhibitor A reduced the viability of patient-derived primary TICs but not that of primary human hepatocytes. f Mouse hepatoblast PIL-4 cells become sensitized to Inhibitor A, which exhibited cytotoxic effects against acquired self-renewal activity induced by TBC1D15 expression. g The FP antagonist failed to rescue Inhibitor A-induced cytotoxicity in TICs. (Left) The growth of TIC-derived tumors was dose-dependently reduced by Inhibitor A in NSG mice. *p < 0.05 vs. vehicle. h Dose-dependent inhibitory effect of Inhibitor A on tumor growth in NSG mice transplanted with dsRed-labeled TICs. i Scanning fluorescence imaging of tumors generated from TICs transduced with lentivirus-dsRed on Day 40 post injection, revealing the marked tumor-suppressive effect of the 20 mg/kg dose. Imaging of dsRed-labeled tumors in the mice described above. Red, yellow, and green fluorescence was detected in mice treated with vehicle or the 10 mg/kg but not in those treated with higher doses. j Mass spectrometry analysis of plasma collected 6 h after the last Inhibitor A administration. Inhibitor A was detected in the treated mice but not in the vehicle-treated mice. k Inhibitor A suppressed the growth of patient-derived HCC cells transplanted into NSG mice (PDX model), and this effect was abolished by N1ICD and NUMB-3D transduction. The tumor-killing effects of Inhibitor A are mediated by the activated NICD, the phosphomimetic NUMB and the restoration of p53 protein expression. l Inhibitor A-treated xenograft tumors had an elevated p53 protein level but reduced N1ICD and NANOG protein levels. m Representative images of immunohistochemical staining for liver cancer markers [albumin (ALB), α-fetoprotein (AFP)], NOTCH1, and KI67 in human normal liver tissues vs. hepatocellular carcinoma tissues. Scale bars = 11.53 µm. The insets show the 10X images at 100X magnification.

To extend the inhibitor study to patient HCC-derived TICs, we treated TICs isolated from patient-derived HCC tissues with increasing concentrations of Inhibitor A. Treatment with Inhibitor A at concentrations ranging from 0.001–1 μM reduced tumor cell viability but not the viability of primary human hepatocytes (Fig. 6e), suggesting that the cytotoxic effects may be specific for TICs. To test this hypothesis, we used the mouse hepatoblast cell line PIL-4 without or with TBC1D15 expression (Fig. 6f, left). PIL-4 cells lacking TBC1D15 expression were not affected by Inhibitor A treatment and showed no cytotoxicity (Vector; Fig. 6f, right), while PIL-4 cells with TBC1D15 expression showed a dose dependent cytotoxicity to Inhibitor A (TBC1D15; Fig. 6f, right). This suggested that Inhibitor A is selectively cytotoxic to TBC1D15-expressing self-renewing cells.

Inhibitor A is known as a prodrug for PGF2α (Latanoprost). Therefore, we asked whether its effects are mediated by the PGF2α receptor (FP), which is known to mediate biological effects such as smooth muscle contraction and a reduction in intraocular pressure. Interestingly, the selective FP antagonist (2 μM AL8810) failed to prevent the cytotoxic effects of Inhibitor A on TICs (Fig. 6g), suggesting that the inhibitory effect of Inhibitor A is independent of FP. PGF2α is a hormone shown to reduce the viability of luteal cells via its ability to suppress luteotropic NOTCH signaling56, suggesting the possibility of negative NOTCH regulation by PGF2α. This observation led to the important question of whether Inhibitor A suppresses the NOTCH pathway in TICs via the synthesis of PGF2α.

The properties of Inhibitor A in vivo were examined, including the resulting pharmacokinetics, toxicity, and efficacy data, to determine the optimal dosing scheme. For this purpose, we treated NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ) mice subcutaneously transplanted with TICs with Inhibitor A. This treatment commenced when the tumor volume reached 100 mm3 as determined by a caliper and was continued by injection 5 times/week for 8–10 weeks. The treatment dose-dependently reduced the growth of TIC-derived tumors (Fig. 6h), as monitored by observation of the red fluorescence(dsRed)-labeled TICs transplanted in these NSG mice (Fig. 6h, i). A separate study in NSG mice was conducted with a three dosage regimen (10, 20, and 40 mg/kg, IP) to measure the serum and tissue levels of Inhibitor A and evaluate tissue histology. The serum levels of Inhibitor A were measured via mass spectrometric analysis of serum samples collected from PDX mice 6 h after the final drug injection (40 mg/kg; Fig. 6j).

To generate tumors of the desired size, we monitored the tumor sizes in the untreated HCC PDX model weekly beginning three weeks post-transplantation for a total of 4 weeks (until the end of week 7) (Fig. 6k). For comparison, the injection of an AAV8 vector expressing NICD and NUMB-3D into HCC tissue was done prior to transplantation to show whether this could abrogate the drug response. All experimental animals were euthanized at the end of the seven-week study period, and all the tumor tissues were collected for histological analysis and tumor characterization. These results indicated that the therapeutic effect of Inhibitor A is based on suppression of the NOTCH pathway mediated by inhibition of the TBC1D15-NOTCH interaction and could be antagonized by transduced NICD and NUMB-3D.

To validate whether AAV-mediated transduction changed NICD expression or NANOG expression, immunoblotting was used to measure the levels of NANOG, NICD, p53, and NUMB in the xenograft tumors (Fig. 6l). To determine whether Inhibitor A treatment or NUMB-3D expression changed the NOTCH1 and/or TBC1D15 protein level in hepatocytes, immunohistochemical analysis was performed to monitor the expression of hepatocyte markers (ALB and AFP), KI67, and NOTCH1 (Fig. 6m). These results indicated that Inhibitor A treatment reduced tumor cell proliferation and NOTCH1 expression in hepatocytes.

Deletions within the TBC1D15 N-terminus-NOTCH interaction domain reduced tumorigenesis

We further examined the main sequence of TBC1D15 that is important for its tumorigenic effects. To this end, mutations were introduced in the N-terminal region of this protein. The schematic illustrations show three possible CRISPR guide RNA target positions in TBC1D15 for generating knockout (KO) mutations (Fig. 7a). The illustration shows the intrahepatic injection of CD133(+) TICs or CRISPR RNP-Cas9 TBC1D15 KO CD133(+) TICs into NSG mice. Western blot analysis was used to validate the expression loss of TBC1D15 in CD133(+) TICs after CRISPR RNP-Cas9-mediated TBC1D15 KO. CRISPR-mediated KO of TBC1D15 reduced the number of cell spheres and cell viability (Fig. 7a). The resulting tumor masses in these animals are shown in Fig. 7b. CRISPR RNP-Cas9 TBC1D15 KO CD133(+) TICs formed tumors with lower volumes after intrahepatic injection in NSG mice (Fig. 7c, inset, right bottom). Immunohistochemical analyses revealed increased levels of TBC1D15, NUMB, NOTCH1, and TOM20 in the tumor tissues of NSG mice injected with CD133(+) TICs. Previous figures show targeted mutations, whereas here, domain-targeted knockout of TBC1D15 was performed. By comparison, CRISPR-mediated KO of TBC1D15 in the NICD interacting domain reduced the tumorigenicity of CD133(+) TICs (Fig. 7d). Immunoblot analyses also revealed that the TBC1D15 protein level was reduced in cells transfected with sgRNA targeting TBC1D15 (Fig. 7e).

Fig. 7: CRISPR-mediated knockout of the NUMB-binding domain of TBC1D15 (the CANOE domain) reduced tumorigenesis.
figure 7

a Left panel and inset, Schematic illustrations of three target positions for TBC1D15 gene knockout (KO) and Western blot analysis of the expression of TBC1D15 in CRISPR RNP-Cas9 TBC1D15 KO CD133(+) TICs. Right panel, sphere formation and cell viability assay results in the indicated cells. b Schematic illustrations of intrahepatic injection in NSG mice with CD133(+) TICs or CRISPR RNP-Cas9 TBC1D15 KO CD133(+)TICs and the resulting tumor masses (lower). c Tumor volumes following intrahepatic injection and growth of CD133(+) TICs or CRISPR RNP-Cas9 TBC1D15 KO CD133(+) TICs in NSG mice. (Inset, Right bottom). d Representative immunohistochemical images of TBC1D15, NUMB, NOTCH1, and TOM20 in tumor tissues from NSG mice injected with CD133(+) TICs or CRISPR RNP-Cas9 TBC1D15 KO CD133(+) TICs; scale bar = 11.74 µm. The insets show the 10X images at 63X magnification. e Immunoblot analyses showing that the TBC1D15 protein level was reduced upon CRISPR-sgRNA-mediated targeting of TBC1D15. f Hypothetical model of TBC1D15-mediated FIS1-dependent recruitment of NICD to mitochondria. (Left) In normal cells, CDK8 phosphorylates NOTCH, which is followed by recruitment of the E3 ubiquitin ligase and displacement of NUMB, leading to ubiquitin-dependent proteasomal degradation of NOTCH. (Right) In TICs, TBC1D15 masks the NICD PEST domain to inhibit the CDK8-dependent phosphorylation of NUMB, which stabilizes NICD such that it is recruited to the mitochondrial outer membrane-localized FIS1. Without NICD phosphorylation, NUMB is not recruited to the NICD PEST domain, but the NUMB5 isoform is internalized into mitochondria.