Viral infection is one of the leading medical challenges of the 21st century. Viruses are parasites that lack the basic metabolic machinery needed to replicate or assemble and therefore must hijack host processes to complete their life cycle. Early research on metabolic changes in infected cells used 13C tracing of glucose and glutamine to show that human cytomegalovirus (HCMV) infection induces glycolysis and fatty acid synthesis in fibroblasts1. Direct pharmacological inhibition of this pathway suppresses HCMV replication2. More recently, lipidomics has been used to identify fatty acids and lipid mediators critical for influenza replication in A549 cells3,4. Again, reversal of virus-suppressed ω3-polyunsaturated fatty acid (PUFA) metabolism was shown to block viral replication4. System-wide proteomic and transcriptomic analysis of HCV infection in replicating Huh7.5 cells has suggested that the virus similarly hijacks host metabolism to support its replication, assembly and egress5,6. Together, these findings suggest that viruses evolved as metabolic engineers, modulating host processes to optimize virus production.

Our focus was HCV infection, a pandemic affecting 3% of the world's population7. This nonintegrating virus replicates in liver cells, causing fatty liver disease and type 2 diabetes8,9. The underlying cause of these metabolic changes is difficult to study owing to the lack of biopsy samples from patients in the early stages of infection and difficulties infecting primary hepatocytes in culture10. Nevertheless, studies using derivatives of the Huh7 cell line have shown that HCV strongly depends on host lipid metabolism for its replication, assembly and egress11. Proteomic analysis of HCV infection in Huh7 cells has shown that, like HCMV and influenza, HCV induces glycolytic enzymes but paradoxically upregulates the enzymes of oxidative phosphorylation as well6. Enigmatically, several studies have demonstrated upregulation of both lipid peroxidation and lipid accumulation in infected Huh7 cells5,6.

Interestingly, host metabolic processes are regulated at the transcriptional level by a network of ligand-activated transcription factors called nuclear receptors12. Unraveling the connections between transcriptional regulators and the virus-induced metabolic response would permit systematic modulation of metabolic pathways, enabling scientists to probe the complexity of host–virus interactions. Regretfully, transcriptional control of metabolism is often obscured by other, massive disease-induced transcriptional responses, such as inflammation and wound healing. Consequently, attempts to link metabolic fluxes to transcriptional regulation thus far have been carried out primarily in single-cell organisms13.

Here we overcame the technical barriers to in vitro study of HCV infection by using oxygenated cocultures of primary human hepatocytes that support normal gene expression levels and metabolic function14. We found that these cocultures supported robust HCV infection, which caused metabolic changes that stabilized after 8 d of culture (Fig. 1). This stable experimental system allowed us to use targeted metabolomics to derive a complete metabolic network induced by HCV infection. Using this metabolic network, we identified differentially activated metabolic fluxes and corresponding changes in gene expression. We identified transcriptional regulators, validated them experimentally using a library of GFP reporters, and systematically perturbed their activity using small-molecule inhibitors. Our analysis showed that HCV-infected primary hepatocytes become dependent on an HNF4α-induced shift from oxidative phosphorylation to glycolysis. In addition, we clarify the paradoxical disconnect between lipid oxidation and accumulation, and concomitantly show that drug metabolism was significantly induced by HCV infection in this study. Finally, we confirmed our findings using patient serum and biopsy samples. Our work elucidates the metabolic fingerprint of HCV infection, identifying distinct assemblies of transcriptional–metabolic regulatory circuits in primary human hepatocytes.

Figure 1: Metabolic fingerprint of HCV infection in oxygenated cocultures of primary human hepatocytes.
figure 1

(a) Schematic of the experimental design. Primary human hepatocytes were cocultured with lung endothelial cells under high-oxygen, serum-free conditions. Cultures were infected with cell culture variants of HCV (JFH-1 or JC1), and their metabolic function was tracked for 10 d in culture. Once metabolism had restabilized, we carried out metabolic flux analysis to identify differentially activated metabolic pathways. Transcriptional regulatory analysis was used to identify the upstream regulators by their differentially activated target genes in each metabolic pathway. (b) Phase images of naive and JFH-1-infected cultures on day 10 post-infection. Numerous droplets are seen in infected cells. (c) Fluorescent image of JC1-RFP-infected culture counterstained for neutral lipids on day 10 post-infection. (d) Top, production of HCV RNA and infectious virus titer as a function of time post-infection, normalized to a culture of Huh7.5.1 infected with JFH-1 (n = 3). Bottom, intracellular levels of HCV core protein, HCV RNA and triglycerides on day 10 post-infection (n = 3). (e) Changes in glucose uptake and in production of urea, albumin, lactate, ketone bodies and triglycerides in naive and HCV-infected primary hepatocytes (n = 3). *P < 0.05; **P < 0.01, Student's t-test n refers to the number of experimental replicates. Scale bars, 10 μm (b) or 20 μm (c). Error bars in d and e indicate s.d.


Robust HCV infection of primary human hepatocytes

We cocultured freshly isolated primary human hepatocytes with endothelial cells expressing the HCV capture receptor L-SIGN15. Cells were seeded in 95% oxygen under serum-free conditions14 and infected with the JFH-1 cell-culture variant of HCV or a mock-infection control on the first day of culture (Fig. 1a). Hepatocytes cocultured with L-SIGN-expressing endothelial cells showed 2.6-fold higher levels of intracellular HCV RNA after 10 d of infection compared with those cultured alone (P < 0.05; Supplementary Results, Supplementary Fig. 1). Primary hepatocyte morphology was markedly altered by HCV infection. Naive cells displayed a cuboidal morphology, smooth cytoplasm and bright cell borders indicative of differentiated, healthy cells. In contrast, JFH-1-infected cells were noticeably smaller and contained numerous lipid droplets in their cytoplasm (Fig. 1b). To visualize infection and lipid accumulation, we infected hepatocytes with a JC1 variant of HCV that contained an in-frame insertion of RFP in the NS5A viral protein16 (hereinafter called JC1-RFP) and counterstained for intracellular triglycerides (Fig. 1c). Fluorescence microscopy showed that 50% ± 15% of the hepatocytes were positive for HCV after 10 d of infection. HCV production stabilized after 8 d of infection, reaching 30% of the RNA level of model cells (Huh7.5.1 infected with JFH-1) and infectivity titers of 400 focus forming units (f.f.u.) per ml (Fig. 1d). Intracellular analysis showed HCV core protein levels of 6.2 ± 2 fmol per mg of total protein and a 4.4-fold increase in intracellular triglycerides (P < 0.05); these results confirm steatosis, a clinical hallmark of infection (Fig. 1d). Interestingly, although JFH-1-infected Huh7.5.1 cells showed a marked increase in apoptosis upon infection, no significant cell death was observed in primary hepatocytes. Cell viability remained constant above 90% ± 5% with minimal ALT release (Supplementary Fig. 1).

Long-term analysis of metabolic activity showed clear differences between HCV-infected and uninfected cocultures (Fig. 1e, Supplementary Fig. 1). Glucose uptake rates in JFH-1-infected cocultures were 29% ± 5% higher (P < 0.01), whereas urea and albumin production were consistently 39% ± 4% and 70% ± 7% lower (P < 0.01), respectively. Lactate and ketone body production were both elevated, whereas the production of triglyceride was dramatically reduced by 69% ± 20% (P < 0.01). Similar to previous reports10, we observed metabolic changes occurring 4 d post-infection, when HCV infectivity began to peak. Metabolic changes stabilized 8 d post-infection, correlating with HCV production reaching a steady state (Fig. 1d,e).

Metabolic fingerprint of HCV infection

The stabilization of primary hepatocyte metabolism, and the quiescent state of the cells, simplifies the derivation of intracellular fluxes by applying mass balance around basic metabolites at pseudo-steady state17 (Supplementary Tables 1 and 2). In total, 31 different metabolites were measured by targeted metabolomics over 48 h on days 9–11 of culture (Online Methods), and the cells were lysed on day 11 for protein and mRNA analysis. Figure 2a shows the metabolic fingerprint of HCV infection, with HCV up- and downregulated metabolic fluxes. Several pathways show clear, statistically significant (P < 0.05), coordinated changes after HCV infection. These include the upregulation of glycolysis toward lactate production (flux 1–8) as well as upregulation of ketone body production (flux 47–50). In contrast to previous proteomic studies using Huh7 cells6, cholesterol and bile acid synthesis were strongly downregulated (flux 76–79), as were glutamine and nitrogen metabolism (flux 15–20) and oxidative phosphorylation (flux 10–14).

Figure 2: Respiratory and glycolytic flux analysis of HCV-infected and sofosbuvir-treated primary human hepatocytes.
figure 2

(a) Metabolic flux balance map of JFH-1-infected primary hepatocytes compared to naive cells. “Upregulated flux” and “downregulated flux” refer to changes in after infection. Statistically significant changes determined as those with P < 0.05 by Student's t-test (n = 3). Metabolic fluxes measured by targeted metabolomics or HPLC are marked with an asterisk. TCA, tricarboxylic acid. (b) Flux measurements (Supplementary Fig. 2) of mitochondrial basal respiration, oxidative phosphorylation, glycolysis and fatty acid oxidation, normalized to naive cells. JFH-1 infection caused a 34% drop in mitochondrial function and a 36% drop in oxidative phosphorylation (P < 0.0001, n = 3). Sofosbuvir reversed this effect, but function remained significantly lower in sofosbuvir-treated cells than in naive cells (P < 0.001, n = 3). JFH-1 infection increased fatty acid oxidation by 85% (P < 0.02, n = 3) and glycolysis by 169% (P < 4 × 10−5, n = 3); both effects were drastically diminished by treatment with sofosbuvir (P < 0.05, n = 3). (c) Metabolic phenotype graph of naive cells, JFH-1-infected cells and infected cells treated with sofosbuvir, showing the metabolic shift toward glycolysis in HCV-infected cells and its reversal after sofosbuvir treatment (n = 3). **P < 0.001; n refers to experimental replicates. Error bars in b and c indicate s.d.

To validate our results, we quantified changes in oxidative phosphorylation, fatty acid oxidation and glycolysis in JFH-1-infected primary hepatocytes and in cells treated with sofosbuvir, an antiviral with direct action against HCV18 (Online Methods). JFH-1 infection caused a 36% ± 4% drop in oxidative phosphorylation, indicating a significant shift to glycolysis (P < 0.0001; Fig. 2b,c, Supplementary Figs. 2 and 10). Sofosbuvir treatment reversed this effect, but function remained lower than in naive cells (P < 0.001), possibly because of residual HCV or lingering damage (Fig. 2b,c). In contrast, HCV infection increased fatty acid oxidation by 85% ± 29% (P < 0.02), and this effect was also reversed by treatment with sofosbuvir (P = 0.058; Fig. 2b,c, Supplementary Fig. 2). Finally, HCV infection increased glycolysis by 169% ± 10% (P < 4 × 10−5), and the effect was completely reversed by treatment with sofosbuvir (P = 0.368; Fig. 2b,c, Supplementary Fig. 2). This well-coordinated transition from energetic to glycolytic phenotype (known as the Warburg effect) and fatty acid oxidation, coupled with inhibition of the urea cycle, is markedly similar to the metabolic transformation in liver cancer19.

Regulation of HCV-induced metabolic transformation

Gene expression analysis of naive and HCV-infected primary hepatocytes identified 893 differentially expressed genes (false discovery rate (FDR) q-threshold < 0.05). Unbiased enrichment analysis identified multiple affected processes, including wound healing (P < 6.3 × 10−12), oxidative stress (P < 3.2 × 10−8) and drug and lipid metabolism (Supplementary Fig. 3). Gene-set enrichment analysis (GSEA) focused on metabolic terms showed significant HCV-induced upregulation of drug metabolism (P < 0.0001), fatty acid oxidation (P < 0.0065) and glycolysis (P < 0.05), as well as downregulation of oxidative phosphorylation (Supplementary Fig. 4). Changes in gluconeogenesis and fatty acid biosynthesis were not significant (P > 0.135).

These HCV-induced metabolic changes suggest underlying transcriptional regulation. To elucidate this regulation, we grouped functional annotations into six metabolic categories that were sufficiently large for our statistics20,21 (Supplementary Table 3). The differentially expressed genes were enriched in lipid metabolism (P < 3.2 × 10−8), cholesterol metabolism (P < 2.4 × 10−12), glucose metabolism (P < 1.2 × 10−14) and drug metabolism (P < 1.1 × 10−13). We found 214 differentially expressed genes (24%) involved in metabolic processes (Supplementary Fig. 3). To identify regulators, we searched for transcription factors with target genes that were differentially expressed during HCV infection (Supplementary Tables 4 and 5). Targets were taken from the TRANSFAC database and computationally validated using genome-wide binding data for HNF4α and CEBPα in human hepatocytes22, as well as mouse overexpression data23. Our unbiased search for regulators identified 14 transcription factors controlling the observed changes in metabolic genes, including 12 nuclear receptors known to have a role in regulating the metabolic pathways identified above. We constructed GFP activity reporters for the 14 transcription factors but were able to experimentally validate only seven functional reporters (Supplementary Fig. 5).

HNF4α-induced glycolysis is an HCV proviral response

HCV-infected primary hepatocytes significantly upregulate glycolysis toward the production of lactate (P < 0.01) while decreasing mitochondrial oxidative phosphorylation, suggesting a Warburg-like effect in nonproliferating primary hepatocytes (Figs. 2b,c and 3a). GSEA showed significant HCV-induced upregulation of glycolysis but no significant change in gluconeogenesis (Fig. 3a, Supplementary Fig. 4). An unbiased search for transcriptional regulators of glycolysis showed that HCV infection activated HNF1α (MODY3) and HNF4α (MODY1), factors implicated in the development of maturity-onset diabetes of the young (MODY), a hereditary form of diabetes (Fig. 3b,c). The change in expression of regulators and key target genes was independently confirmed by qRT-PCR (Fig. 3d).

Figure 3: HCV-infected cells become dependent on HNF4α-induced glycolysis.
figure 3

(a) Schematic of hepatic glycolysis showing metabolic fluxes and involved genes. Red, upregulated fluxes and genes; green, downregulated; black, no change. (b) Transcriptional regulatory analysis of glycolysis showed enrichment of HNF4α and HNF1α targets. (c) Interaction network of the identified regulators HNF4α and HNF1α, their differentially expressed target genes, and glucose metabolism. (d) qRT-PCR showed a threefold increase in expression of HNF4α and its target gene PKLR (P < 0.001, n = 3). Although HNF1α expression did not increase significantly, the expression of its target genes SLC2A2 and ALDOB did significantly increase (P < 0.05, n = 3). (e) HNF4α GFP reporter activity was fourfold greater (P < 0.001, n = 3) in JC1-RFP-infected Huh7.5.1 cells compared to Pol controls. Inhibition of HNF4α with Medica16 blocked its activation25 while simultaneously decreasing JC1-RFP expression. Scale bars, 10 μm. (f) JC1-RFP-infected Huh7.5.1 cells showed increased glucose uptake and lactate production (P < 0.05, n = 5). Stimulation with HNF4α inhibitors Medica16 and BI6015 reversed the increase in glycolysis (P < 0.001, n = 5; Supplementary Fig. 6). (g) Inhibition of HNF4α with Medica16 (P = 0.016, n = 3) or BI6015 (P < 0.001, n = 3) led to decreased expression of both HCV RNA and NS5A-RFP. (h) Inhibition of glycolysis with 2DG (P = 0.004, n = 3) or BrPA (P = 0.003, n = 3) led to a twofold decrease in HCV replication. (i) Cellular apoptosis increased by fourfold in JC1-RFP-infected cells exposed to Medica16 (P < 0.001, n = 3), whereas Pol control cells showed no response. (j) Fluorescent micrographs of apoptotic nuclei (green) in Medica16-treated JC1-RFP-infected cells and Pol controls. Scale bars, 50 μm. *P < 0.05, **P < 0.001; n refers to experimental replicates. Images in e and j are representative of three biological replicates and five technical replicates (15 repeats total). Error bars indicate s.d. throughout.

To validate the casual role of HNF4α in a rigorous model of HCV infection, we electroporated JC1-RFP or polymerase-negative (Pol) HCV RNA into Huh7.5.1 cells (Online Methods). Cells were then infected with an HNF4α activity reporter in which binding of an activated nuclear receptor drives EGFP expression (Fig. 3e). HNF4α activity was greater by (4 ± 1)-fold in HCV-positive cells compared to Pol controls (Supplementary Fig. 5). HNF4α activation was reversed after treatment with the HNF4α antagonists Medica16 and BI6015 (refs. 24, 25; Fig. 3e). Huh7.5.1 cells replicating JC1-RFP produced more lactate, the end product of glycolysis, than Pol controls did (P < 0.05). However, treatment with Medica16, BI6015 or short interfering RNA (siRNA) reversed this increase in glycolysis (P < 0.001; Fig. 3f, Supplementary Figs. 6 and 7). Importantly, HCV replication was blocked by treatment with either Medica16 (P = 0.016) or BI6015 (P < 0.001), leading to a three- to fourfold decrease in viral RNA both in cells infected with JC1 (genotype 2a) and in full-length genotype 1b replicon cells26, as well as in primary human hepatocytes (Fig. 3e,g, Supplementary Figs. 8 and 9).

To confirm that HCV relies on glycolysis, we treated Huh7.5.1 cells replicating JC1-RFP with the glycolytic inhibitors 2DG and BrPA, as well as DCA, a PDK1 inhibitor (Fig. 3h, Supplementary Fig. 13). Treatment with 2DG and BrPA blocked HCV replication (P < 0.001), whereas downstream treatment with DCA showed marginal inhibition (P < 0.05), demonstrating viral reliance on glycolytic by-products (Fig. 3h). Moreover, HCV infection caused liver cells to become sensitive to Medica16, showing a fourfold increase in apoptosis over Pol controls (Fig. 3i,j). These results show a strong proviral effect of HCV-induced glycolysis.

HCV-induced lipid oxidation is an antiviral response

HCV infection has a puzzling effect on lipid metabolism, as cells show both lipid accumulation and oxidation at the same time5,6. The involvement of the PPARα transcription factor in this process has been under debate27,28. We found that HCV-infected primary hepatocytes significantly upregulated fatty acid oxidation toward the production of ketone bodies (P < 0.01) while downregulating cholesterol biosynthesis (P < 0.01; Figs. 2b and 4a, Supplementary Fig. 2). Surprisingly, GSEA showed upregulation of both (Supplementary Fig. 4). However, we identified two bifurcation points. First, induction of HMGCL over HMGCR directs acetyl-CoA toward ketone bodies rather than cholesterol (Fig. 4a), and induction of ADRP directs part of the triglyceride flux to lipid droplets rather than lipoproteins (Fig. 4a). An unbiased search for transcriptional regulators of lipid metabolism revealed a hierarchal interconnected network of transcription factors with a role in lipid metabolism (Fig. 4b). The nuclear receptors PPARα, SREBP, FXR and LXRα are known regulators of fatty acid oxidation (PPARα and SREBP), cholesterol metabolism and bile acid metabolism (FXR and LXRα). Figure 4c shows the links between PPARα, FXR and the metabolic pathway. Regulators and target genes for each factor were independently confirmed by qRT-PCR (Fig. 4d).

Figure 4: HCV induction of PPARα- and FXR-dependent lipid oxidation is a host antiviral metabolic response.
figure 4

(a) Schematic of fatty acid metabolism, triglyceride synthesis and bile acid production showing metabolic fluxes and involved genes. Red, upregulated fluxes and genes; green; black, no change. Panels at right show quantitative changes leading to intracellular lipid accumulation (top) and hypolipidemia (bottom). (b) Transcriptional regulatory analysis of lipid metabolism showed significant enrichment of PPARα and FXR, known regulators of fatty acid oxidation and cholesterol biosynthesis. (c) Interaction network of PPARα and FXR, their target genes, and lipid metabolism. (d) qRT-PCR showed no change in the expression of PPARα and FXR, but their coactivators and target genes were significantly induced. (e) PPARα and FXR GFP reporter activity increased by threefold in JC1-RFP-infected Huh7.5.1 cells. Inhibition of PPARα with GW9662 and of FXR with BML-GR235 blocked their activation but increased viral replication. Scale bars, 10 μm. (f) JC1-RFP-infected Huh7.5.1 cells showed a twofold decrease in bile acid production and a twofold increase in ketone body production. BML-GR235 reversed the decrease in bile acid production, and GW9662 and GW6471 reversed the increase in ketone body production. (g) Inhibition of PPARα or FXR led to an increase in HCV RNA and NS5A-RFP expression. (h) Inhibition of fatty acid oxidation with etomoxir or ranolazine led to an increase in HCV replication. (i) Fluorescent micrographs of intracellular lipids (orange) in GW9662-treated and untreated cells. Scale bars, 100 μm. (j) Intracellular lipid accumulation after PPARα inhibition. *P < 0.05, **P < 0.001; n refers to experimental replicates. n = 5 (d,fh) or 3 (j). Error bars indicate s.d. throughout. Images in e and i are representative of three biological replicates and five technical replicates.

We again sought to validate our findings in Huh7.5.1 cells robustly infected with JC1-RFP or Pol controls. Cells were infected with PPARα, SREBP, FXR and LXRα activity reporters (Fig. 4b and Supplementary Fig. 5). Although the activity of SREBP and LXRα did not change, we observed a threefold induction of PPARα activity (P < 0.001) and a 3.8-fold induction of FXR activity (P < 0.001) in HCV-infected cells versus Pol controls (Fig. 4e). Activation of PPARα was reversed by treatment with its inhibitors GW9662 and GW6471, and FXR activation was blocked by its inhibitor BML-GR235 (Z-guggulsterone; Fig. 4e,g). Huh7.5.1 cells replicating JC1-RFP showed a two-fold increase in ketone body production and a two-fold decrease in bile acid production, respective end products of fatty acid oxidation and cholesterol biosynthesis (Fig. 4f). Importantly, treatment with GW9662 or GW6471 reversed the HCV-induced increase in ketone body production (P < 0.001), and treatment with BML-GR235 reversed the decrease in bile acid production (P < 0.001; Fig. 4f). siRNA against each regulator showed similar results (Supplementary Fig. 6). Inhibition by these compounds provides internal controls showing independent PPARα regulation of fatty acid oxidation and FXR regulation of cholesterol biosynthesis (Supplementary Fig. 7). Interestingly, all three inhibitors increased replication of HCV genotype 2a by 40–50%, leading to a 1.4-fold increase in viral RNA (Fig. 4g). A similar 1.8- to 2.4-fold increase in HCV was seen for genotype 1b replicon cells and primary human hepatocytes (Supplementary Figs. 8 and 9). To confirm that HCV relies on fatty acid oxidation, we treated Huh7.5.1 cells replicating JC1-RFP with etomoxir and ranolazine, inhibitors of fatty acid oxidation (Fig. 4h, Supplementary Fig. 13). Treatment with etomoxir and ranolazine increased HCV replication (P < 0.001) by 2- and 1.5-fold, respectively.

To probe the second bifurcation point (Fig. 4a), we used intracellular fluorescent staining to quantify lipid accumulation (Online Methods). We found that PPARα inhibition led to increased lipid accumulation (Fig. 4i,j). We note that the HCV life cycle is dependent on lipid accumulation in droplets29 as well as on cholesterol biosynthesis for FBL2 anchoring30. Therefore, in contrast to HNF4α induction of glycolysis, the observed activation of PPARα and FXR during HCV infection is detrimental for the virus life cycle and might be part of a host metabolic antiviral response to infection.

HCV activation of PXR induces hepatic drug metabolism

Changes in hepatic drug metabolism can affect drug clearance rates but are difficult to assess in Huh7.5.1 cells owing to marginal expression of CYP450 enzymes. CYP3A4 is responsible for the clearance of 40% of drugs on the market, including newly introduced antivirals. Our experimental system of primary human hepatocytes does not share this limitation and showed eightfold and sixfold upregulation of CYP3A4 and CYP2C9, respectively, in response to viral infection (P < 0.001; Fig. 5c). Other important drug-clearance enzymes such as CYP1A2, 2E1 and 2D6 showed a similar two- to sixfold increase in expression (P < 0.05; Supplementary Fig. 11). To confirm these results, we carried out CYP450 activity assays in JFH-1-infected primary human hepatocytes. CYP3A4 and CYP2E1 showed twofold and threefold increases in activity, respectively (P < 0.05), confirming our results (Fig. 5e).

Figure 5: HCV activates PXR and induces CYP450 expression.
figure 5

(a) Transcriptional regulatory analysis of drug metabolism showed significant enrichment in a number of transcription factors, including PXR and CAR, known regulators of CYP450 expression. Inferred regulators of drug metabolism are highlighted in blue on the full regulatory network (Supplementary Table 4). (b) Interaction network of the identified regulators HNF4α, PXR and CAR, and their differentially expressed targets. (c) qRT-PCR showed no change in the expression of PXR but a ninefold induction of CAR (P < 0.001, n = 3). PXR and CAR target genes CYP3A4 and CYP2C9 were upregulated eightfold and sixfold, respectively (P < 0.001, n = 3). (d) PXR GFP reporter activity was fivefold greater (P < 0.05, n = 3) in JC1-RFP-expressing Huh7.5.1 cells compared to Pol controls. Inhibition of PXR with silibinin blocked its activation without affecting viral replication (Supplementary Fig. 8). Scale bars, 10 μm. Images are representative of three biological replicates and five technical replicates. (e) Functional assays showed a twofold induction of CYP3A4 activity (P < 0.005, n = 3) and a threefold induction of CYP2E1 activity (P = 0.013, n = 3) in HCV-infected primary hepatocytes. PXR-controlled CYP3A4 was completely blocked by silibinin. *P < 0.05; **P < 0.001; n refers to experimental replicates. Error bars in c and e indicate s.d.

An unbiased search for transcriptional regulators of drug metabolism revealed a small network of transcription factors with roles in drug metabolism (Fig. 5a). Figure 5b shows the links among HNF4α, PXR, CAR and the CYP450 enzymes. Regulators and target genes were independently confirmed by qRT-PCR (Fig. 5c). We sought to validate these findings in Huh7.5.1 cells robustly infected with JC1-RFP or Pol controls. Cells expressing the PXR activity reporter showed a 5.3-fold induction of PXR activity (P < 0.05) in HCV-infected cells compared to Pol controls (Fig. 5d, Supplementary Fig. 5). Activation of PXR was reversed by its inhibitor silibinin31 without observable effects on viral replication (Fig. 5d). Exposure to silibinin blocked the HCV-induced induction of CYP3A4 activity, reversing the effect of PXR (Fig. 5e).

Clinical validation of HCV-induced metabolic changes

Our in vitro primary human hepatocyte cocultures reproduced known clinical features of HCV infection (i.e., steatosis and hypolipidemia) while predicting others (i.e., decreased bile acids and increased ketones) (Fig. 1). To test these predictions, we analyzed gene expression in liver biopsy samples obtained from HCV patients in early stages of infection and compared them to healthy controls (Online Methods). Although in vivo gene expression levels seldom correlate with in vitro levels, expression in primary human hepatocytes was strongly correlated to expression in biopsies of healthy liver, with R2 = 0.85 (Fig. 6a). We identified 1,259 genes that were differentially expressed in liver biopsies during HCV infection (FDR q < 0.01). However, in spite of the well-known metabolic problems associated with HCV infection, unbiased enrichment analysis did not identify any metabolic terms due to overlapping processes such as an immune response (P < 5.2 × 10−3), underscoring the need for in vitro analysis.

Figure 6: Clinical biopsy and serum samples validate experimental model results.
figure 6

(a) Log-scale scatter plot of whole-genome gene expression correlating in vitro expression levels with expression in biopsy samples from healthy liver. R2 was 0.85, compared with 0.72 for HepG2 cells. (b) Log-scale scatter plot of virus-induced gene expression in vitro (culture) versus in vivo (biopsy). Changes in the expression of genes related to metabolism showed a positive (P < 0.05) correlation between in vivo and in vitro samples. 88% of the metabolic genes upregulated in vitro (red squares) were also upregulated in vivo, but downregulated genes (blue diamonds) appeared to have randomly distributed in vivo expression. (c) Metabolic analysis of serum samples from HCV-infected subjects versus controls. The results confirm previous reports that HCV induces hyperglycemia and hypolipidemia, but they also show a decrease in bile acids and a corresponding increase in ketone bodies and lactate as predicted by the in vitro model. (d) Unbiased transcriptional regulatory analysis of patient biopsy samples showed activation of key nuclear receptors, mirroring those identified in vitro.

However, genes differentially expressed in vivo were enriched with those found in vitro (P < 0.005), as well as with genes involved in glucose, lipid and drug metabolism (Supplementary Figs. 3 and 4). Limiting our analysis to metabolic genes, we found that 88% of the genes induced by HCV infection in vitro (Fig. 6b) were similarly induced in vivo. For example, expression of genes from the cytochrome P450 family increased with HCV infection both in vivo and in vitro (Supplementary Fig. 11). However, genes downregulated by HCV infection in vitro were randomly distributed in terms of in vivo expression (Fig. 6b), possibly because of the presence of multiple cell types (i.e., immune cells) in liver biopsies. An unbiased search for transcriptional regulators in vivo revealed a similar network of transcription factors enriched for differentially expressed metabolic genes, including HNF4α, PPARα and FXR (Fig. 6d).

To further test our metabolic findings, we obtained plasma samples from HCV patients in early stages of infection and compared them to samples from healthy control subjects (Fig. 6c). There was marginally more alanine transaminase (ALT) in plasma from HCV patients (97 U/L versus 21 U/L), confirming low levels of inflammation due to steatosis (P < 0.001). Serum cholesterol levels were 17% lower (P < 0.01) during infection, as has been previously shown32. In addition, there was a small (13%) increase in serum glucose levels (P < 0.05), suggesting disturbed glycemic control. Importantly, we found that circulating bile acid levels decreased by 53% (P < 0.05), whereas levels of β-hydroxybutyrate (βHB) ketone bodies increased by 36% (P < 0.05). Lactate levels showed a minor increase of 13%. These results strongly support our in vitro observations, confirming the rapid metabolic changes taking hold in early stages of HCV infection.


Metabolism is a complex phenomenon regulated on multiple levels. In recent years, a family of ligand-activated transcription factors called nuclear receptors have emerged as key regulators of metabolic processes. Nuclear receptors are activated by metabolites such as fatty acids and glucose, and they control negative feedback loops12,33, offering an opportunity to reprogram metabolism at the transcriptional level. This approach is appealing in the context of metabolic disease and viral infection, where pharmaceutical targeting of transcriptional regulators could rewire chronic metabolic states.

In this work, we studied the effects of HCV infection on primary human hepatocytes. Our cocultures of primary human hepatocytes with L-SIGN+ endothelial cells showed robust infection, affecting 50% of the cells (Fig. 1), compared with previously reported infection rates of 6–10% (ref. 10). Similar infection rates, ranging from 20% to 50%, have been found in vivo34,35. In contrast to earlier work on Huh7 cells, infection of primary hepatocytes reproduced lipid accumulation and hypolipidemia, clinical hallmarks of infection, matching the transcriptional and metabolic signature of patient samples (Fig. 6). This allowed us to track metabolic changes with confidence and correlate these changes to their underlying transcriptional regulators (Figs. 3,4,5).

The effect of HCV infection on CYP450 expression is an area of major interest for pharmaceutical development, and it could not be studied in Huh7 cells, as tumors show minimal expression of CYP450 enzymes. We show that HCV upregulates CYP3A4 both in vitro and in vivo (Fig. 5). This increase would affect the clearance rates of emerging antiviral drugs that are substrates of CYP3A4, such as telaprevir, boceprevir and simeprevir.

We also show that HCV infection of primary hepatocytes induces glycolysis over oxidative phosphorylation in a manner similar to HCMV infection of fibroblasts2 and the Warburg effect in cancer36,37. Although upregulation of glycolysis has been previously observed in HCV-infected Huh7.5 cells via proteomic analysis6, so have oxidative phosphorylation and glutamine metabolism6. This suggests that the global upregulation of metabolic pathways previously reported was a result of HCV-induced upregulation of Huh7.5 proliferation, rather than the host response to infection38. In contrast, our model shows a striking induction of glycolysis over oxidative phosphorylation in both primary human hepatocytes and growth-arrested Huh7.5 cells (Supplementary Fig. 10), strengthening earlier evidence of HCMV-induced glycolysis in serum-starved fibroblasts2.

An unbiased search for metabolic regulators showed that HNF4α, a nuclear receptor associated with MODY39, has a critical role in host glycolytic response. Although there are conflicting reports regarding HNFs in HCV infection40,41, our work clearly shows that HNF4α inhibition reverses glycolysis in infected cells. Suppression of glycolysis by Medica16, BI6015 or the glycolytic inhibitors 2DG and BrPA blocked HCV replication as a result of substrate limitation (Fig. 3). Importantly, the treatment also increased apoptosis in HCV-infected cells, suggesting that the process is driven by host reliance on nonoxidative energy (Fig. 2) rather than by the needs of the virus.

Unraveling lipid metabolism proved more complex. Early analysis of HCV-infected cells suggested cross-purpose induction of both fatty acid oxidation and lipogenesis6. Lipogenesis is similarly induced in HCMV infection and certain types of cancer42, but not in HCV infection of primary hepatocytes. In fact, lipid peroxidation has recently been shown to attenuate HCV replication and egress43,44. Our results show PPARα-dependent induction of fatty acid oxidation and FXR-dependent suppression of cholesterol biosynthesis (Fig. 4). Similar activation of FXR has a critical role in liver carcinogenesis45. These results are surprising, as HCV relies on cholesterol biosynthesis to anchor its NS5A cofactor through FBL2 (ref. 46) and on lipid droplets for replication and lipoprotein-dependent egress29,47. Importantly, inhibition of either nuclear receptor, or of fatty acid oxidation directly, caused considerable upregulation of HCV replication, suggesting that in contrast to HNF4α, the effects of PPARα and FXR are part of a host antiviral response.

In summary, our work provides a metabolic fingerprint of HCV infection in primary human hepatocytes, mapping it to transcriptional regulators that can be pharmaceutically modulated. We show that viral infection dramatically affects host metabolism, but not always for the benefit of the virus. The results show that virus needs can conflict with the host's requirement for energy and that virus replication might be tightly controlled by positive and negative interactions with the host48.


Hepatocyte oxygenated cocultures.

Primary human hepatocytes were obtained from BD Biosciences or were kindly provided by Dr. Stephen C. Strom (University of Pittsburgh). Human cells were purified in 33% Percoll solution centrifuged at 500g for 5 min before seeding. Cell viability post-purification was greater than 90%, and purity was greater than 95%. After purification hepatocytes were mixed with lung endothelial cells (Lonza) at a 4:3 ratio in ice-cold culture medium and seeded at 175,000 cells/cm2 in a humidified incubator set at 37 °C and conditioned by a gas mixture of 95% oxygen and 5% CO2 (Airgas). After overnight seeding under 95% oxygen, the cultures were returned to atmospheric levels of oxygen and 5% CO2. HCV infection was carried out 24 h after seeding (day 1). Cells were cultured in serum-free basal medium (Lonza) supplemented with ascorbic acid, transferrin, EGF and antibiotics. Medium was further supplemented with 100 μg/mL of rat-tail collagen during seeding to enhance attachment14. Hepatocyte culture medium was collected every 48 h and stored at −80 °C for metabolic analysis. After 11 d of culture (10 d post-infection) endothelial cells were removed by trypsinization and the hepatocyte population was scraped off the surface for protein and mRNA analysis. Hepatocyte purity post-trypsinization was assessed by immunofluorescence microscopy and was found to be greater than 95%. Cultures were confirmed to be free of mycoplasma contamination using a PCR test kit (Biological Industries).

Drugs and reagents.

2-DG (sc-202010), BrPA (sc-296029), and DCA (sc-260854) were purchased from Santa Cruz Biotechnology. GW9662 (M6191), GW6471 (G5045), etomoxir (E1905), ranolazine (R6152), silibinin (S0417), and Medica16 (M5693) were purchased from Sigma-Aldrich. BI6015 (4641) was purchased from Tocris, and BML-GR235 (39025) was purchased from Enzo. A Human Albumin Elisa Quantitation Set (E80-129) was purchased from Bethyl Laboratories. A Human α-Fetoprotein ELISA kit (ab108838) was purchased from Abcam.

JFH-1, JC1-RFP, and polymerase-negative (Pol) RNA production and electroporation.

The Huh7.5.1 human hepatoma cell line was provided by F. Chisari (Scripps Research Institute). Huh7.5.1 cells were cultured in DMEM supplemented with 10% FBS and 200 units/mL penicillin–streptomycin in a 5% CO2-humidified incubator at 37 °C. Plasmids containing the JFH-1 and JC1-RFP full-length genome were kindly provided by T. Wakita (National Institute of Infectious Diseases) and R. Bartenschlager (University of Heidelberg), respectively. The JFH replicon containing the GND inactivating mutation in the viral polymerase (Pol) served as negative control. In vitro transcribed JFH-1, JC1-RFP and Pol RNA were delivered to cells by electroporation using a BTX model 830 electroporator (820 V, five 99-μs pulses given at 220-ms intervals). Cells were left to recover for 15 min at 22 °C and then mixed with 10 mL of pre-warmed growth medium and seeded for further analysis. Fluorescence microscopy showed robust JC1-RFP infection affecting over 90% of the Huh7.5.1 cells. To generate JFH-1 and JC1-RFP viruses for infection, we collected Huh7.5.1 conditioned medium for 72 h after electroporation. Conditioned medium was concentrated and used to infect primary human hepatocytes at multiplicity of infection of 0.5–1 for 24 h. Culture medium was subsequently changed daily. The mock control consisted of similarly processed medium conditioned by naive Huh7.5.1 cells.

Quantitative reverse-transcription polymerase chain reaction (qRT-PCR).

Virus samples were purified using a QIAamp viral RNA mini kit (Qiagen). The reverse-transcription reaction step was performed on a Mastercycler ep Gradient S (Eppendorf) instrument using a High Capacity cDNA reverse transcription kit (Applied Biosystems). Real-time PCR was performed on a MyiQ real-time PCR detection system using the iScript One-Step RT-PCR kit with SYBR Green (Bio-Rad) according to the respective manufacturers' instructions.

Small-molecule inhibition of nuclear receptor and metabolic targets.

JC1-RFP or Pol electroporated Huh7.5.1 cells were grown to 90% confluence and treated with small-molecule inhibitors or DMSO control for 24 h. Primary human hepatocytes were similarly treated 10 d post-infection. Cell viability was >90% for all concentrations tested (Supplementary Fig. 12).

Central glycolysis inhibitors 2-deoxy-D-glucose (2DG) and 3-bromopyruvate acid (BrPA) were used at 1 mM and 10 μM, respectively. Dichloroacetate (DCA), which blocks PDK1, redirecting pyruvate to oxidative phosphorylation, was used at 100 μM (ref. 49). Fatty-acid-oxidation inhibitors etomoxir50 and ranolazine51 were used at 100 μM and 50 μM, respectively.

The specificity of nuclear-receptor antagonists was validated using qRT-PCR (Supplementary Fig. 5). HNF4α activation was reversed after treatment with 250 μM Medica16 or 5 μM BI6015 (refs. 24, 25). PPARα activation was reversed by treatment with 10 μM GW9662 or 10 μM GW6471. FXR was blocked by treatment with 100 μM BML-GR235 (ref. 52), and PXR was blocked by treatment with 200 μM silibinin31. HCV levels were determined by qRT-PCR and fluorescence quantification of JC1-RFP normalized to the number of Hoechst-positive nuclei.

HCV infectivity.

The production of infectious JFH-1 virus particles by primary hepatocytes was measured as previously described44. Briefly, naive Huh7.5.1 cells were grown to 80% confluence and exposed to cell culture supernatants serially diluted tenfold in the culture medium. After overnight incubation at 37 °C, 5% CO2, the medium was replaced, and cells were cultured for an additional 3 d. Levels of HCV infection were determined by immunofluorescence staining for HCV core protein47. The viral titer is expressed in focus forming units (f.f.u.) per milliliter of supernatant.

Biochemical assays.

Albumin concentration was determined with a commercial ELISA kit (Bethyl Laboratories). Urea concentration was measured with a blood urea nitrogen kit (Stanbio Labs). Glucose and lactate levels were determined using commercial kits testing for glucose oxidase (Stanbio Labs) and lactate oxidase (Trinity Biotech, Bray), respectively. Triglycerides were quantified using a commercial kit (Sigma Chemical) based on glycerol quantification. Free fatty acids were determined on the basis of their CoA derivatives using a commercial kit (BioVision). The ketone bodies acetoacetate and β-hydroxybutyrate were measured by the disappearance of NADH in the conversion of acetoacetate to β-hydroxybutyrate in the presence of β-hydroxybutyrate dehydrogenase and by the production of NADH in the reverse reaction. Cholesterol was measured using a commercial kit (Stanbio Labs) based on cholesterol oxidase. Total bile acids were determined through NADH formation using a commercial kit (BioQuant). Individual amino acid concentrations in medium samples were quantified by HPLC as previously described53. ALT and AST levels were quantified using a commercially available kit (Thermo Fisher Scientific). HCV core proteins were measured using the ORTHO HCV antigen ELISA kit (Ortho Clinical Diagnostics). The oxygen uptake rate was measured in parallel cultures using the Ocean Optics FOXY fiber optic oxygen sensor immersed in a well-stirred chamber as previously described54.

Metabolic flux balance analysis (MFA).

MFA is a stoichiometric model of hepatocyte metabolism in which intracellular fluxes are calculated via the application of mass balances around basic metabolites. The analytical predictions of the model used in this work were previously confirmed using radiolabeled substrates55. The stabilization of primary hepatocyte metabolism 8 d after infection, and the lack of cell proliferation, allowed us to use the model under steady-state conditions, which reduced the balance equations to a set of linear algebraic equations given by dz/dt = S × v = 0, where z represents metabolites, S is an n × m matrix of coefficients, and v is a vector of m fluxes17,56. This approach accounts for the complex interdependence between pathways due to a common pool of cofactors and thus provides a comprehensive picture of the hepatic metabolic state17,56. Here we expanded the model to include lipid metabolism, adding fluxes for cholesterol synthesis, bile production, triglyceride synthesis and accumulation. The mathematical model uses 35 metabolites and 80 chemical reactions implemented and solved using MATLAB. Because of linear dependence the matrix is completely determined.

Metabolic flux quantification of HCV-infected and sofosbuvir-treated cells.

Primary human hepatocytes were infected with JFH-1 containing medium for 7 d and treated with 10 μM sofosbuvir (Cayman Chemical) or DMSO vehicle control for 72 h. 3.5 × 103 cells were plated on collagen-coated XFp Flux Analyzer Miniplates (Seahorse Bioscience). A mitochondrial stress test assay was conducted per the manufacturer's instructions (Seahorse Bioscience). Briefly, cells were incubated in unbuffered DMEM supplemented with 2 mM glutamine, 1 mM sodium pyruvate, and 10 mM glucose (pH 7.4) for 1 h at 37 °C in a non-CO2 incubator. We measured the basal oxygen consumption rate (OCR) for 30 min and then injected 1 μM oligomycin, a mitochondrial complex V inhibitor that blocks oxidative phosphorylation. The decrease in OCR due to oligomycin treatment was defined as the oxidative phosphorylation rate. 1 μM carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP), an uncoupling agent, was added at 60 min to measure maximal mitochondrial activity, and complete inhibition was induced at 90 min using a mixture of 0.5 μM antimycin A and rotenone, mitochondrial complex III and mitochondrial complex I inhibitors.

Free fatty acid oxidation was measured as previously described57. Briefly, after overnight incubation in substrate-limited medium containing 0.5 mM glucose to prime cells for exogenous fatty acid utilization, cells were incubated in unbuffered DMEM supplemented as described above. Basal OCR was measured for 30 min, and then 200 μM palmitate (C16:0) was injected. Fatty acid oxidation was blocked at 60 min by injection of 100 μM etomoxir, a carnitine palmitoyltransferase I (CPT1) inhibitor. OCR attributable to exogenous fatty acid oxidation was defined as the difference between maximal palmitate-treated and minimal etomoxir-treated conditions.

A glycolysis stress test assay was conducted per the manufacturer's instructions (Seahorse Bioscience). Briefly, cells were incubated in unbuffered DMEM supplemented with 2 mM glutamine and 1 mM sodium pyruvate (pH 7.4) for 1 h at 37 °C in a non-CO2 incubator. The extracellular acidification rate (ECAR) was measured for 30 min in the absence of glucose; this served as the baseline measurement. We measured the rate of glycolysis by injecting 10 mM glucose at 30 min, followed by 1 μM oligomycin, blocking oxidative phosphorylation at 60 min. Finally, 100 mM 2DG, a nonmetabolized glucose analog, was injected at 90 min to competitively inhibit glycolysis. The increase in ECAR due to the addition of glucose was defined as the glycolysis rate, and the difference between oligomycin and 2DG treatments defined the maximal glycolytic capacity of the cells. Data are presented normalized to 104 cells.

Metabolic phenotype comparison was conducted per the manufacturer's instructions (Seahorse Bioscience), as previously described (ref. 58). Briefly, absolute basal OCR and ECAR levels, normalized to protein concentration, were graphically plotted for naive, JFH-1-infected and sofosbuvir-treated primary human hepatocytes. The graphical representation indicates a relative metabolic phenotype between the three groups, the relative reliance on glycolysis or oxidative phosphorylation.

Generation of transcription factor activity reporter library.

We generated transcription factor activity reporters by cloning DNA response elements (i.e., PPRE and LXRE) upstream of a minimal CMV promoter. Binding of an activated transcription factor to its DNA response element induces the transcription of destabilized copGFP with a half-life of 2 h, permitting the measurement of both transcriptional activation and inhibition. Reporters were generated for 14 computationally identified transcription factors including HNF4α, PPARα, FXR, SREBP, LXRα, PPARγ and PXR. To validate their activity, we introduced reporters to Huh7.5.1 or HepG2 cells and stimulated them with a classical agonist (Supplementary Fig. 5).

SREBP1c, LXRα, and PPARγ pGreenFire1 activity reporters were purchased from System Biosciences and validated in-house. Reporters contained ATCACGTG, GGGTTACTGGCGGTCATTGTA and TGTAGGTCACGGTGACCTAC, respectively, upstream of a minimal CMV promoter. The FXR reporter was constructed by PCR amplification of human genomic DNA from −145 to +86 of the BSEP promoter and subsequent cloning upstream of copGFP. The promoter of BSEP contains an FXR response element (GGGACATTGATCCT). The PPARα reporter was constructed by PCR amplification of human genomic DNA from −562 to +1,890 of the human CPT1A gene and cloning upstream of copGFP. The PXR reporter construct was cloned from the human CYP3A4 promoter and was a kind gift from Dr. C. Liddle (University of Sydney, Australia). The promoter from the original luciferase construct was subcloned into the copGFP construct. We constructed the HNF4 reporter by cloning three HNF4 response elements (GGGTCAAAGGTCA) upstream of the minimal CMV promoter in the same pGreenFire1 reporter construct.

Reporter activity screen in infected Huh7.5.1 cells.

Huh7.5.1 cells electroporated with JC1-RFP or Pol RNA were re-plated in a black 96-well plate at a density of 5 × 104 cells per well 24 h after electroporation. GreenFire lentivirus reporters were generated according to the manufacturer's directions. Briefly, 293T cells were cotransfected with a pGreenFire1 activity reporter plasmid, a plasmid expressing GAG-Pol, and a plasmid expressing VSV-G in a ratio of 3:2:1. 24 μg of total DNA was diluted in 750 μL of OptiMEM and 25 μL of polyethylenimine, mixed, allowed to settle for 5 min and added to 293T packaging cells cultured without antibiotics. Lentivirus was collected from the supernatant on the third day of culture, filtered through a 45-μm syringe filter, mixed with 4 ng/mL polybrene, and added at 100 μL/well to Huh7.5.1 cells expressing JC1-RFP or Pol RNA. Lentivirus infection was repeated the following day and reached >80% efficiency based on the EGFP control.

24 h after the second infection (72 h after electroporation) the cells were imaged using Carl Zeiss LSM700 confocal microscope. Photos were analyzed using ImageJ software. Activity is presented as the number of GFP-positive cells above a preset threshold normalized to the Pol control. We quantified HCV replication in the same samples by defining a low threshold, identifying HCV-positive cells, and measuring RFP fluorescence in HCV-positive cells divided by the area of the cells. A minimum of four regions were quantified for each experimental condition.

Intracellular lipid accumulation.

Intracellular lipid accumulation was quantified using Nile Red staining (Sigma Chemical) according to the manufacturer's instructions. Briefly, cultures were fixed with 4% formaldehyde in PBS for 20 min at room temperature, washed, treated with 1 μM Nile Red and counterstained with Hoechst 33258 (Sigma Chemical). For intracellular lipid accumulation in JC1-RFP-infected cultures, we used LipidTOX Green (Thermo Fisher Scientific) neutral lipid stain according to the manufacturer's instructions. Images were taken using a Carl Zeiss LSM700 confocal microscope and analyzed using ImageJ software. Lipid accumulation is presented as lipid intensity normalized to Hoechst. A minimum of four regions were quantified for each experimental condition.

Quantification of apoptosis.

Quantitation of apoptotic cell nuclei was carried out with the DeadEnd Fluorometric TUNEL assay (Promega) according to the manufacturer's instructions. Briefly, cultures were fixed with 4% formaldehyde in PBS for 20 min at 4 °C, washed, and permeabilized with 0.2% Triton X100 for 5 min. Cultures were then equilibrated to room temperature for 10 min in buffer and treated with TdT reaction mixture for 60 min at 37 °C. The reaction was stopped using SSC for 15 min, at which point cells were washed and counterstained with Hoechst. Images were taken using aCarl Zeiss LSM700 confocal microscope and analyzed using ImageJ software. Apoptosis is presented as TUNEL intensity normalized to Hoechst. A minimum of four regions were quantified for each experimental condition.

Clinical HCV sample collection.

Plasma and liver biopsy samples were obtained from individuals infected with HCV attending the Department of Gastroenterology and Hepatology of the University Hospital Duisburg-Essen, Germany, after obtaining written informed consent. The study was approved by the Institutional Review Board of the University of Essen, Germany (IRB 08-3662). Inclusion criteria included a diagnosis of early HCV infection, defined as HCV RNA abundance greater than 100,000 IU/mL, but fibrosis stage 1 determined using FibroScan or pathological assessment. Patients with HIV or HBV coinfection, decompensated liver disease, or organ transplantation were excluded, as were those on medications that alter lipid metabolism.

Patient plasma samples.

Plasma samples were collected from 15 people positive for HCV infection (genotypes 1a, 1b and 3a), after an overnight fast. Patient age ranged from 33 to 72 years; their ALT levels were below 190 U/l, and their fasting glucose levels were below 130 mg/dL. FibroScan results showed only minimal fibrosis.

Patient biopsy samples.

Liver biopsy samples were collected from five patients, at early stages of HCV infection (genotypes 2b, 1a, 2a and 3a), after an overnight fast. Patient age ranged from 25 to 39 years, and their GPT levels were below 112 U/l. FibroScan and histological assessment showed minimal fibrosis and low-grade inflammation in the liver samples. Normal human liver biopsy data from 19 patients were taken from GEO accession GSE14323 (ref. 58). The age for these subjects ranged from 9 to 66 years, and their cause of death was primarily head trauma.

Affymetrix gene arrays.

In vitro RNA samples were tested on the Affymetrix GeneChip Human Exon 1.0 ST array at the Center for Genomic Technologies at the Hebrew University of Jerusalem. In vivo RNA samples were isolated from snap-frozen liver biopsy samples and tested on Affymetrix GeneChip Human Genome U133A 2.0 array at the Genomic Facility of the University Hospital Essen, Germany. Both processes were carried out in an Affymetrix accredited facility according to the company instructions. GSE14323 samples deposited58 were similarly collected and hybridized to the same Human Genome U133A 2.0 array. Microarray data were deposited in GEO (GSE84587).

Gene expression analysis.

After a robust multi-array average (RMA) procedure and quantile normalization, the probes were filtered using BioDiscovery's Nexus Expression software. Probes with a variance less than 0.2 and intensity less than 4.0 were removed. Differential expression of HCV-infected versus naive cells in vitro was evaluated via two-sample Student's t-test, using a threshold of P < 0.05 after Bonferoni correction for multiple hypotheses. Under these conditions, 893 genes were differentially regulated in vitro. Patient-to-patient differences required a more stringent test for the in vivo samples in order to reduce variability. We therefore carried out a two-sample Student's t-test using a threshold of P < 0.01 after Bonferoni correction for multiple hypotheses and 1.5-fold change. Under these conditions, a similar-sized group of 1,259 genes were found to be differentially regulated in vivo.

Assembly of metabolic categories.

We defined metabolic categories by uniting various functional annotations chosen by hand, taken from GO20 or KEGG21 (Supplementary Table 3). We assembled five sets for metabolic categories and a set of oxidative stress response genes. General liver metabolism was defined as the combination of all metabolic categories.

Functional annotations of gene expression.

To functionally annotate differentially expressed genes, we tested for their enrichment in functional gene sets. The general test for functional enrichment of the differentially expressed genes against various functional categories was done using the DAVID tool59. We tested the enrichment of differentially expressed genes with metabolic categories (defined above), using all genes expressed in vitro as the background list. Enrichment P values were calculated using Fisher's exact test and with Bonferoni correction for multiple hypotheses. The analysis of the in vivo and in vitro data was done the same way, including only genes measured on both arrays.

Target genes of transcription factors.

Transcription factor target genes were defined by TRANSFAC on the basis of expression analysis and binding data. We added CAR as a target gene of HNF4α on the basis of literature reports60. We found target genes in the liver for the factors HNF4α and CEBPα for validation using DNA binding data measured by ChIP-seq in primary human hepatocytes22. To determine target genes for these two factors, we used the binding-site locations previously determined22 and defined a gene as a target of the factor if we found a binding site 3,000 bp upstream or of the transcription start site (TSS) of the gene. TSS locations were downloaded from the human genome version hg19 in the UCSC Genome Browser. We found target genes in the liver for the factors SREBF1, SREBF2 and their negative regulator SCAP using their defined target genes based on overexpression in mouse primary hepatocytes23. We mapped the mouse target genes to their human orthologs using orthologous mapping from the MGI database (Informatics).

Transcriptional activity analysis.

We conducted an unbiased search for transcription factors differentially regulating metabolic pathways after HCV infection (Supplementary Table 4). A transcription factor was considered to be differentially regulating a pathway if its target genes were enriched with differentially expressed genes within this pathway. Enrichment P values were calculated using Fisher's exact test and corrected for multiple hypotheses using a false discovery rate (FDR) threshold of 1%. The analysis was done separately for each constraint metabolic pathway defined above (Supplementary Table 4). We created the full regulatory network of each metabolic pathway in a similar way, but using all genes in each metabolic category.

Similarly, we searched for transcription factors regulating differentially expressed metabolic genes in response to HCV infection by testing the enrichment (FDR threshold of 5%) of transcription factor target genes with differentially expressed metabolic genes in vitro (total of 235 genes). To compare the in vivo and in vitro data, we tested each in vitro enriched factor for enrichment (FDR threshold of 5%) of its target genes with the set of metabolic genes differentially expressed in vivo (total of 189 genes; Supplementary Table 5).

HCV pseudoparticles (HCVpp).

HCVpp were created by binding of HCV surface receptors E1–E2 to 100-nm-diameter fluorescent beads. Green-fluorescent beads were purchased from Polysciences, Inc. Properly folded E1–E2 heterodimers were recombinantly expressed and linked to beads as previously described47.

Full-length genotype 1 replicon cells.

Huh7 clonal cells harboring a full-length genotype 1b replicon were cultured in DMEM supplemented with 10% FBS and 250 μg/ml G418 in a 5% CO2–humidified incubator at 37 °C. Cells were grown to 90% confluence and exposed to 250 μM Medica16, 5 μM GW-9662, 10 μM GW-6471, 100 μM BML-GR235, or 0.5% DMSO vehicle control for 24 h. RNA was isolated and purified using a Macherey-Nagel NucleoSpin RNA II kit according to the manufacturer's instructions. RNA concentration and purity were determined using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). Gene expression analysis was carried out using a Bio-Rad iScript One-Step RT-PCR kit on the Bio-Rad CFX96 Real-Time system. HCV (forward, 5′-AGAGCCATAGTGGTCTGCGGAA-3′; reverse, 5′-AAATCTCCAGGCATTGAGCGGGTT-3′) gene transcription was evaluated using the ΔΔCt method normalized to RPL32.

LIVE/DEAD quantification of cell viability.

Huh7.5.1 cells and primary human hepatocytes cocultured with endothelial cells were stained using a fluorescent LIVE/DEAD viability assay (Invitrogen Life Sciences) in which the cytoplasm of live cells accumulates green-fluorescent calcein because of esterase activity, while the nuclei of dead cells are labeled red by ethidium homodimer owing to a loss of nuclear membrane integrity. Cells were quantified using ImageJ.

AFP secretion ELISA analysis.

Control (Pol) and JC1-infected Huh7.5.1 cells were treated with Medica16 and BI6015 for 24 h. Supernatant and cell lysate samples were collected. AFP concentration was analyzed using the AFP human ELISA kit (Abcam) according to the manufacturer's directions. All data were normalized to total protein content using the Bradford assay.

Small interfering RNA analysis In Huh7.5.1 cells.

Cells were plated in 12-well plates 24 h before transfection at 30–50% confluency. Transfection experiments were performed using Lipofectamine 2000 (Invitrogen). 50 μL solution of siRNAs and LF2K were prepared in Opti-Mem and allowed to mix for 30 min before their addition to cells at final concentrations of 5–100 nM siRNA and 2.3 μg/mL LF2K. Cells were incubated in the transfection solutions at 37 °C and 5% CO2. After 24 h, cells were washed twice with DPBS (Gibco), and EGFP fluorescence was quantified using a plate reader at 480 nm excitation/525 nm emission. Fluorescence intensity was normalized to control wells that were treated with transfection reagent but no siRNA. Cytotoxicity was assessed by LIVE/DEAD staining.

Statistical analysis.

Experiments were repeated two or three times with triplicate samples for each experimental condition. Data from representative experiments are presented, and similar trends were seen in multiple trials. Unless otherwise noted, a parametric two-tailed Student's t-test was used to calculate significant differences between groups.

Accession codes.

Microarray data were deposited in GEO (GSE84587).

Additional information

Any supplementary information, chemical compound information and source data are available in the online version of the paper. Reprints and permissions information is available online at http://www.nature/com/reprints/index.html. Correspondence and requests for materials should be addressed to Y.N.