The PERK-mediated UPR activates FUT1 promoter activity via ATF4 induction under glucose restriction.

Further analysis of our RNA-seq profiling data identified the top 3 most significantly enhanced genes (FUT1, heat shock protein family A member 6 [HSPA6], and fosB proto-oncogene, AP-1 transcription factor subunit [FOSB]) when cells were cultured under glucose restriction. Two of these harbor ATF4 binding sites on their promoter regions (Figure 2A). With ATF4 being an important downstream transcription factor of PERK-mediated UPR signaling, we hypothesized that PERK-UPR-ATF4 induction under glucose restriction would enhance the liver TIC phenotype through HSPA6, FUT1, or FOSB. Analysis of FUT1 and ATF4 mRNA levels in The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) data set revealed that ATF4 expression was positively correlated with FUT1 (Figure 2B) but not FOSB expression (Supplemental Figure 4). Notably, HSPA6, which was also overexpressed when HCC cells were cultured under low glucose, also showed a positive correlation with FUT1 expression in HCC (Supplemental Figure 4), although it did not harbor an ATF4 binding site and thus was not further analyzed. In addition to the TCGA LIHC data set, we further confirmed our observations by qPCR, where we found FUT1 to be significantly enriched by approximately 8-fold when the HCC cell lines Huh7 and CLC13 and HCC patient–derived spheroids were cultured under glucose restriction (Figure 2C). Likewise, in vivo HCC tumors formed by injecting HCC cells treated with low glucose also showed elevated FUT1 expression compared with that of HCC cells treated with high glucose (Supplemental Figure 2). Notably, the expression of FUT2, a close family member of FUT1, was not altered when cells were cultured under the same conditions, suggesting the importance of FUT1 alone in this process (Supplemental Figure 5).

Open in a separate window

Figure 2

The PERK-mediated unfolded protein response activates FUT1 promoter activity via ATF4 induction under glucose restriction.

(A) A volcano plot revealed the top 3 most significantly enhanced genes when HCC cells were cultured in low glucose. (B) Analysis of FUT1 and ATF4 expression in The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) database (unpaired Student’s t test and Pearson’s correlation test). (C) FUT1 expression in low-glucose culturing conditions was validated by qPCR in Huh7 and CLC13 cells and HCC patient–derived spheroids (unpaired Student’s t test). The data shown are representative of 3 independent experiments. (D) Confirmation of ATF4 binding to both predicted sites on FUT1 by ChIP-qPCR using anti-ATF4 and control IgG antibodies (unpaired Student’s t test and 1-way ANOVA). The data shown are representative of at least 3 independent experiments. (E) Luciferase reporter assays found that only ATF4 binding site 2 (Δ3′) was critical in modulating FUT1 transcriptional activity and FUT1 transcriptional activity was upregulated under glucose-deprived conditions in Huh7 and CLC13 HCC cells. ATF4 expression was induced by low-glucose treatment (unpaired Student’s t test and 1-way ANOVA). The data shown are representative of 3 independent experiments. HG, high glucose; LG, low glucose; FL, full-length. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

To determine whether ATF4 directly regulates FUT1 expression, we assessed the binding of ATF4 to the FUT1 promoter, which is predicted to contain 2 consensus binding sites (at –74 to –102 and –533 to –591 relative to the transcription start site) (Supplemental Figure 6A). Chromatin immunoprecipitation (ChIP) of ATF4 followed by qPCR with 2 different sets of primers spanning the 2 putative ATF4 binding sites in the FUT1 promoter showed a 4- to 6-fold enrichment in the FUT1 signal over ChIP with nonspecific IgG in both Huh7 and CLC13 HCC cells (Figure 2D). To confirm the ability of ATF4 to control FUT1 expression, we tested the effects of introducing full-length ATF4 with one or both putative ATF4 binding sites deleted on luciferase activity (with luciferase expression driven by the FUT1 promoter). Although ATF4 bound to both predicted binding sites, we observed only a significant reduction in luciferase activity when the predicted ATF4 binding site close to the 3′ end (i.e., site 2) was deleted, suggesting that only 1 of the 2 predicted ATF4 binding sites is critical in controlling FUT1 transcription (Supplemental Figure 6B and Figure 2E). We also tested the level of FUT1 promoter activity under glucose-deprived conditions, and consistently, an increase in luciferase activity was observed in both Huh7 and CLC13 cells treated with glucose-restricted medium (Figure 2E). ATF4 suppression in HCC cells cultured under low glucose consistently led to a reduction in FUT1 expression in both Huh7 and CLC13 cells (Supplemental Figure 7). Notably, GRP78, p-PERK, p-eIF2α, ATF4, and FUT1 showed a stepwise increase in expression when going from high (hyperglycemic; 4.5 g/L) to medium (physiological; 1 g/L) to low (hypoglycemic; 0.45 g/L) glucose concentrations (Supplemental Figure 8A), while the addition of the PERKi reversed the change in FUT1 expression (Figure 1D). Culturing the immortalized normal liver cell line MIHA in high or low glucose did not alter ATF4 or FUT1 expression, suggesting that the observed phenomenon was HCC specific (Supplemental Figure 8B). Overall, our data identified PERK-mediated UPR-driven ATF4 as a key regulator of FUT1 transcriptional induction in HCC under glucose-restriction stress. Notably, we also cultured Huh7 and CLC13 HCC cells under normoxic and hypoxic conditions, but ATF4 and FUT1 expression was not altered, suggesting that PERK-mediated UPR activation of FUT1 promoter activity via ATF4 induction is not relevant to the hypoxic tumor microenvironment (Supplemental Figure 9).

FUT1 overexpression is tightly associated with aggressive clinical features. FUT1 encodes the FUT1 enzyme, which is responsible for the addition of fucose to glycans via the α-(1,2) linkage, resulting in the formation of Lewis and H antigens. To confirm the clinical relevance of FUT1 expression, we interrogated publicly available HCC data sets in which we could compare FUT1 expression in specimens with different tumor grades and survival oucomes (GSE14520, TCGA LIHC; ref. 10). FUT1 mRNA expression was significantly elevated in HCC tumor tissues compared with nontumor liver or normal liver tissues in both data sets (Figure 3A). High FUT1 expression was also correlated with the presence of multinodular tumors, advanced tumor stage (nontumor liver vs. stage III), and worse overall survival (Figure 3, B–D). Immunohistochemical analysis of 61 paired primary HCC and corresponding nontumor liver tissues also showed that the FUT1 protein was overexpressed in most HCC samples examined (Figure 3E and Supplemental Table 2). Analysis of the GSE109211 data set also found that HCC patients with low FUT1 and ALDH1L1 expression were more responsive to sorafenib treatment, suggesting that FUT1 together with the cancer stem cell (CSC) marker ALDH1L1 may be a predictive marker for the response to sorafenib in the clinic (Supplemental Figure 10).

Open in a separate window

Figure 3

FUT1 overexpression is tightly associated with aggressive clinical features.

(A) Box-and-whisker plot analysis of FUT1 mRNA levels in nontumor (NT) liver or normal liver and HCC tissues using information gathered from the GSE14520 data set from the NCBI Gene Expression Omnibus (GEO) (left) or The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) database (right) revealed FUT1 to be frequently overexpressed in HCC (unpaired Student’s t test). (B–D) FUT1 overexpression in HCC correlated significantly with multinodular tumors (B), advanced TNM stage (C), and worse overall survival (D) (unpaired Student’s t test, 1-way ANOVA, and Kaplan-Meier survival curve using the log-rank test, respectively). (E) FUT1 immunostaining on a tissue microarray comprising 61 paired NT liver and HCC tissue samples found FUT1 to be frequently overexpressed in HCC. Representative images of immunostaining in 2 patients are shown. Scale bar: 50 μm. The bar graph shows the percentage of cases displaying no, low, medium, and high staining intensity of FUT1 in paired NT and HCC samples (Fisher’s exact test). The dot plot illustrates the quantification of FUT1 by ImageJ (NIH), with both the intensity (mean OD) and number of cells according to the number of nuclei taken into consideration (paired Student’s t test). Mean OD per cell indicates mean OD relative to the number of nuclei. **P < 0.01; ***P < 0.001; ****P < 0.0001.

Knockdown of endogenous Fut1 expression in the liver of an immunocompetent HCC mouse model attenuates hepatocarcinogenesis.

To investigate the causative relationship between FUT1 overexpression and the functional role of FUT1 in HCC, we performed endogenous knockdown of Fut1 expression in an immunocompetent mouse model. Using the hydrodynamic tail vein injection (HTVI) delivery approach, we induced HCC tumors in C57BL/6 mice using a combination of the activated forms of the myristylated AKT (myr-AKT) and N-RasV12 proto-oncogenes (NRAS+AKT) and sleeping beauty (SB) transposase (Figure 4A). A stepwise increase in Fut1 mRNA expression was noted from harvested livers of the mice injected with the empty vector (EV) control alone from early (1 to 2 weeks after injection) to advanced HCC (3 to 5 weeks after injection) to the endpoint, at which time the mice died naturally (6 to 8 weeks after injection) (Figure 4B). Knockdown of endogenous Fut1 expression delivered via tail vein injection of lentiviral particles significantly attenuated the liver to body weight ratio as well as the number and size of tumor nodules compared with those of the nontargeting shRNA control (shNTC) (Figure 4, C and D, and Supplemental Figure 11). Notably, the liver-free body weight was not found to be significantly different between the 2 groups, excluding data bias due to tumor burden–induced cachexia (Supplemental Figure 12). Immunohistochemical analysis of resected tumors showed a marked decrease in the expression of FUT1, the liver CSC markers CD133 and ALDH1L1, and the proliferation marker PCNA in the Fut1-knockdown group (Supplemental Figure 13). The apoptotic marker cleaved caspase 3 did not exhibit any expression in either the control or shFut1 groups and TUNEL staining was negative, suggesting that FUT1 does not affect spontaneous HCC cell death (Supplemental Figure 13). Subsequent ex vivo limiting-dilution assays using cells isolated from the liver of this HCC mouse model demonstrated that the frequency of TICs capable of forming spheres also decreased upon Fut1 knockdown (Figure 4E). These findings support a key role of FUT1 in promoting HCC growth and self-renewal.

Open in a separate window

Figure 4

Knockdown of endogenous Fut1 expression in the liver of an immunocompetent HCC mouse model attenuates hepatocarcinogenesis.

(A) Schematic representation of the hydrodynamic tail vein injection (HTVI) model in C57BL/6 mice. (B) qPCR analysis of Fut1 expression in mice that received HTVI of either empty vector control (EV CTRL) or NRAS, AKT, and sleeping beauty (SB) transposase for HCC induction with samples collected at various time points. n = 22 in total (EV, n = 3; 1 week, n = 3; 2 weeks, n =3; 3 weeks, n = 3; 4 weeks, n = 3; 5 weeks, n = 3; endpoint, n = 7) (1-way ANOVA). (C) Strategy for testing the functional significance of Fut1 in hepatocarcinogenesis. NRAS, AKT, and SB transposase were delivered by HTVI for HCC induction. Lentiviral particles with nontargeting control shRNA (shNTC) or shFut1 were administered twice at 4 and 5.5 weeks. Mice were sacrificed at 10 weeks after plasmid injection. (D) Representative images of dissected livers at the end of the experiment. Scale bars: 5 cm and 2 cm (enlarged images). Successful Fut1 knockdown confirmed by qPCR. n = 11–12 per group. (E) An ex vivo limiting-dilution assay of HCC tumor cells harvested from HTVI mouse models found that the frequency of tumor-initiating cells decreased in the mice administered shFut1 lentiviral particles. The data shown are representative of 3 independent experiments (pairwise tests for differences in stem cell frequencies). *P < 0.05.

FUT1 overexpression is important in mediating growth, self-renewal, and drug resistance in HCC cells.

To functionally demonstrate the importance of FUT1 in TIC maintenance, we used a lentivirus-based approach to stably overexpress FUT1 in Huh7 and CLC13 HCC cells. Efficient FUT1 overexpression was confirmed by Western blot analysis (Figure 5A). Note that manipulation of FUT1 expression did not affect FUT2 expression, suggesting a specific effect of FUT1 on HCC cells (Supplemental Figure 14). FUT1 overexpression resulted in a 2.5- to 7-fold increase in the frequency of TICs, as demonstrated by an in vitro limiting-dilution assay (Figure 5B). Overexpression of FUT1 also resulted in an enhanced ability of the cells to resist the molecularly targeted therapeutic sorafenib, as demonstrated by a decrease in the number of dead cells measured by annexin V/propidium iodide (PI) flow cytometry (Figure 5C). As a complementary approach, we also used lentivirus-based targeting to express shRNAs to reduce FUT1 in Huh7 cells (shFUT1 clones 544 and 565) and observed the opposite functional phenomena (Figure 5, B and C). Of note, FUT1 suppression completely reversed the enhanced self-renewal ability induced by culturing cells in low glucose (Figure 5, B and C). To further determine whether FUT1 overexpression resulted in enhanced tumor propagation in vivo, we subcutaneously implanted HCC cells with or without FUT1 overexpression into immunocompromised NOD/SCID mice. Overexpression of FUT1 resulted in a profound increase in the ability of cells to initiate tumor growth. FUT1 overexpression in HCC cells also resulted in increased tumor incidence, expedited tumor latency, and a higher frequency of TICs. HCC cells with high FUT1 also exhibited significantly worse tumor-free survival than HCC cells without FUT1 overexpression (Figure 5D and Supplemental Table 3).

Open in a separate window

Figure 5

FUT1 overexpression is important in mediating growth, self-renewal, and drug resistance in HCC cells.

(A) Validation of FUT1 overexpression or knockdown in Huh7 and CLC13 HCC cells by Western blotting. (B) An in vitro limiting-dilution assay revealed that FUT1 overexpression and knockdown promoted and attenuated the frequency of tumor-initiating cells, respectively (pairwise tests for differences in stem cell frequencies). (C) Annexin V/PI analysis found that FUT1 overexpression augmented apoptosis in Huh7 cells treated with 4 μM sorafenib (72 hours) and CLC13 cells treated with 4 μM sorafenib (48 hours). DMSO (2%) was used as a control (1-way ANOVA). The reverse phenotype was observed in a Huh7 FUT1-knockdown model (4 μM sorafenib, 72 hours). (D) FUT1 overexpression promoted HCC tumor growth in an in vivo subcutaneous implantation model. Representative image of tumors harvested from mice injected with 5000 cells. Scale bar: 1 cm. Kaplan-Meier curve showing a significant decrease in the percentage of tumor-free survival upon FUT1 overexpression (n = 6). In vivo limiting-dilution assays found that FUT1 overexpression enhanced tumor incidence, expedited tumor latency, and increased the frequency of tumor-initiating cells (pairwise tests for differences in stem cell frequencies and Kaplan-Meier survival curve using the log-rank test). The data shown in A–C are representative of 3 independent experiments. NTC, nontargeting control; OE, overexpression. *P < 0.05; **P < 0.01; ***P < 0.001.

Inhibition of α-(1,2) fucosylation by 2DGal increases the efficacy of sorafenib and eradicates TICs.

In light of the functional and clinical significance of FUT1 in mediating the TIC phenotype, we then speculated whether targeting FUT1 would be beneficial in the treatment of HCC. As a proof of concept, we tested the effects of inhibiting α-(1,2) fucosylation with 2DGal on the ability of sorafenib to suppress HCC proliferation and self-renewal (11). HCC cells stably overexpressing FUT1 were subjected to 2DGal and sorafenib treatment. While FUT1 overexpression enhanced the ability of the cells to resist sorafenib, combination treatment with 2DGal reversed this protective effect and sensitized cells to sorafenib (Figure 6, A and B). Treatment of HCC cells cultured under low glucose with 2DGal also reversed the TIC frequency, as demonstrated by an in vitro limiting-dilution spheroid assay (Figure 6C). We also extended our study to a more physiological setting, utilizing organotypic ex vivo culture of HCC tumor tissues where the pathophysiology of the original tumor is better preserved than that of cell lines. Note that all 3 HCC patient–derived organoids (HCC-HK P1, HCC-HK P2, and HCC10) have been thoroughly characterized, either in-house or by our collaborator, at both the molecular and phenotypic levels, with comparisons made against the original tissue samples (12). HCC-HK P2 and HCC10 both tested positive for FUT1 expression, while HCC-HK P1 did not express FUT1 (Figure 6D). The treatment efficacy of sorafenib and 2DGal was evaluated in ex vivo cultures of HCC patient–derived organoids treated with either sorafenib or 2DGal alone or the combination of both. In the FUT1-expressing HCC-HK P2 and HCC10 cells, combination treatment resulted in the most significant reduction in tumor growth, where 2DGal treatment was found to sensitize HCC cells to sorafenib (Figure 6E). Conversely, in the FUT1-deficient HCC-HK P1 cells, combination treatment did not further enhance the sorafenib response (Figure 6E). Additional analyses were performed to discern the in vivo benefit of their combinatorial effect in HCC in the NRAS+AKT HTVI immunocompetent HCC mouse model, where we found Fut1 to be overexpressed (Figure 4B). Sorafenib and/or 2DGal were administered following HCC formation at 5–7 weeks. At the endpoint, HCC tumor cells were harvested for flow cytometric cell sorting to capture live cells for the ex vivo limiting-dilution assay (Figure 6F). Although sorafenib treatment alone led to a decrease in the frequency of TICs, combination treatment led to an even more significant decrease in the self-renewal ability of the cells compared with the control (Figure 6G), suggesting that 2DGal sensitizes HCC cells to sorafenib and preferentially eradicates the TIC subset of the tumor. FUT1 is a key rate-limiting enzyme in the synthesis of Lewis Y, a membrane-associated carbohydrate antigen (13, 14). Hence, we used the Lewis Y antigen as a readout of the level of FUT1 modification on its target proteins. Immunohistochemical analysis of Lewis Y expression in the resected livers of the 2DGal and combination treatment groups showed a decrease in expression, suggesting that 2DGal effectively impaired the synthesis of Lewis Y antigen on cell-surface proteins (Supplemental Figure 15). Collectively, these data demonstrated the potential of utilizing 2DGal to enhance the efficiency and efficacy of sorafenib and to preferentially eradicate the TIC subset of HCC for long-term tumor eradication.

Open in a separate window

Figure 6

Inhibition of α-(1,2) fucosylation by 2DGal increases the efficacy of sorafenib and eradicates tumor-initiating cells.

(A) Schematic representation of the effects of 2DGal and sorafenib on Huh7 HCC cell proliferation (10 mM 2DGal and 4 μM sorafenib) and HCC patient-derived organoids (HCC-HK P1 and HCC-HK P2, 10 mM 2DGal and 2 μM sorafenib; HCC10, 10 mM 2DGal and 4 μM sorafenib) after 72 hours. (B) CellTiter Glo analysis found that Huh7 cells with FUT1 overexpression (OE) responded to a combination of 2DGal and sorafenib more significantly than either drug alone when compared with the empty vector (EV) control (1-way ANOVA). (C) In vitro limiting-dilution assays of Huh7 cells cultured in low glucose treated with 2DGal (10 mM) (pairwise tests for differences in stem cell frequencies). (D) Western blot analysis of FUT1 expression in HCC-HK P1, HCC-HK P2, and HCC10. (E) CellTiter Glo analysis found that FUT1-expressing HCC organoids (HCC-HK P2 and HCC10) responded to a combination of 2DGal and sorafenib more significantly than either drug alone when compared with HCC-HK-P1 (1-way ANOVA). (F) Strategy for testing the effects of 2DGal and sorafenib in NRAS+AKT+SB HTVI–driven HCC immunocompetent mouse models. (G) Ex vivo limiting-dilution assay of HCC tumor cells harvested from the HTVI mouse models (pairwise tests for differences in stem cell frequencies). The data shown in B–E and G are representative of 3 independent experiments. 2DGal, 2-deoxy-D-galactose; LDA, limiting dilution assay; SB, sleeping beauty; HTVI, hydrodynamic tail vein injection; Combo, combination. *P < 0.05; **P < 0.01; ****P < 0.0001. NS, not significant.

HTVI NRAS+AKT mouse model of HCC.

Six- to 8-week-old male wild-type C57BL/6 mice (Jackson Laboratory) were used, and the procedure was performed as previously described (41, 42). In brief, 20 μg of plasmid encoding human AKT1 (myr-AKT1) and human neuroblastoma Ras viral oncogene homolog (N-RasV12) along with sleeping beauty (SB) transposase at a ratio of 25:1 was diluted in 2 mL of saline (0.9% NaCl), filtered through a 0.22 μm filter, and injected into the lateral tail vein of C57BL/6 mice in 5 to 7 seconds. The constructs used in this study showed long-term expression of genes via hydrodynamic injection (42). Four weeks after HTVI of proto-oncogenes and SB transposase, the mice were separated into 2 groups and administered 5 × 10⁷ transducing units of lentiviruses encoding either shNTC or shFut1 in 100 μL of PBS via tail vein injection. At 5.5 weeks, mice were given another dose of the shNTC or shFut1 lentiviruses. Mouse shNTC (CGTTCTCCGAACGTGTCACGT) and mouse shFut1 target (GTTCTCCGAACGTGTCACGT) sequences were cloned into the LV3 vector.

Tumor-initiating and self-renewal animal studies.

Tumor-initiating and self-renewal abilities were investigated by limiting-dilution and serial transplantation assays. Four- to 6-week-old male NOD/SCID mice (Jackson Laboratory) were injected subcutaneously with 2500, 5000, 10,000, or 50,000 cells. For the high- and low-glucose study, HCC cells were cultured under low- or high-glucose conditions for 6 days prior to primary implantation. For secondary implantation, only 1 tumor was dissociated from each of the low- and high-glucose treatment groups for passage into an additional 5 mice. Tumor incidence and tumor latency were recorded. Tumor-initiating frequency was calculated using extreme-limiting-dilution analysis. Tumor sizes were measured every 3 days by calipers, and tumor volumes were calculated as volume (cm³) = L × W² × 0.5, with L and W representing the largest and smallest diameters, respectively. Tumor initiation and incidence were determined to be positive when the tumor volume reached 14 mm³. Tumors formed were harvested for histological analysis. Only tumors with volumes less than 200 mm³ were harvested and dissociated for subsequent passage to secondary mouse recipients or for in vitro limiting-dilution analysis to minimize the number of necrotic cells within the tumor bulk. Animals that were injected with tumor cells but showed no sign of tumor burden were generally euthanized 6 months after tumor cell inoculation, and animals were opened up at the injection sites to confirm that there was no tumor development.

RNA-seq.

See Supplemental Methods. Gene expression data were deposited in the Gene Expression Omnibus, and the accession number is GSE152927.

Statistics.

Statistical analyses were performed using GraphPad Prism 5.0 or SPSS 21.0. Unpaired and paired Student’s t tests were used to compare the mean values of 2 groups. One-way ANOVA with Tukey’s honestly significant difference (HSD) post hoc test was used to compare the mean values of more than 2 groups. Clinicopathological significance in clinical samples was evaluated by Fisher’s exact test and independent Student’s t test for categorical data and continuous data, respectively. The differences in Kaplan-Meier survival curves were calculated using the log-rank test. The differences in survival analysis in the mouse model were determined by the log-rank test. Data are shown as the mean ± standard deviation. Statistical significance was defined as P < 0.05. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

This project is supported in part by grants from the Research Grants Council of Hong Kong – Theme Based Research Scheme (T12-704/16-R and T12-710/16-R), Collaborative Research Fund (C7026-18G), and the Health and Medical Research Fund from Food and Health Bureau of the Hong Kong Government (07183206). We also acknowledge funding support from “Laboratory for Synthetic Chemistry and Chemical Biology” under the Health@InnoHK Program launched by the Innovation and Technology Commission, The Government of Hong Kong Special Administrative Region of the People’s Republic of China. We thank the Centre for PanorOmic Sciences (The University of Hong Kong) for providing and maintaining the equipment and technical support needed for transcriptome sequencing, mass spectrometry, flow cytometry, animal imaging, and confocal microscopy studies. We thank the Centre for Comparative Medicine Research (The University of Hong Kong) for supporting our animal work studies. We also sincerely thank Meritxell Huch of The Gurdon Institute at the University of Cambridge for providing the patient-derived HCC organoid line HCC10; Lijian Hui of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, for providing CLC13 HCC cells; and Phillip Chiu and Keith Lee of The University of Hong Kong for sharing antibodies.