Lipogenesis and innate immunity in hepatocellular carcinoma cells reprogrammed by an isoenzyme switch of hexokinases

During the cancerous transformation of normal hepatocytes into hepatocellular carcinoma (HCC), the enzyme catalyzing the first rate-limiting step of glycolysis, namely the glucokinase (GCK), is replaced by the higher affinity isoenzyme, hexokinase 2 (HK2). The transcriptomic analysis of HCC tumors shows that highest expression level of HK2 in tumor lesions is inversely correlated to GCK expression, and is associated to poor prognosis for patient survival. To further explore functional consequences of the GCK-to-HK2 isoenzyme switch occurring during carcinogenesis, HK2 was knocked-out in the HCC cell line Huh7 and replaced by GCK, to generate the Huh7-GCK+/HK2− cell line. HK2 knockdown and GCK expression rewired central carbon metabolism, stimulated mitochondrial respiration and restored essential metabolic functions of normal hepatocytes such as lipogenesis, VLDL secretion, glycogen storage. It also reactivated innate immune responses and sensitivity to natural killer cells, showing that consequences of the HK switch extend beyond metabolic reprogramming.

basis of etiology and outcomes revealed metabolic deregulation as a hallmark of HCC progression 71 (4). Indeed, metabolic remodeling is critically required for tumor growth, since bioenergetic 72 requirements and anabolic demands drastically increase (5-7). For instance, HCC cells have lost 73 their ability to secrete very low-density lipoproteins (VLDL), a highly specialized function of 74 hepatocyte and can only secrete low-density lipoproteins (LDL)-like lipoproteins, indicating a 75 defective lipogenesis and/or lipoprotein assembly (8). 76 Metabolic reprogramming in cancer cells involves the modulation of several enzymes by 77 oncogenic drivers (6). Targeting these enzymes is now considered as a therapeutic strategy for 78 several types of cancers (6). Among these enzymes, hexokinase 2 (HK2) stands out because of its 79 elevated or induced expression in numerous cancers, including HCC (9). Hexokinases control the 80 first rate-limiting step of glucose catabolism by phosphorylating glucose to glucose-6-phosphate 81 (G6P), fueling glycolysis as well as glycogen, pentose phosphate and triglyceride synthesis. The 82 human genome contains four genes encoding distinct hexokinase isoenzymes, named HK1 to HK4 83 (HK4 is also known as glucokinase or GCK), with distinct enzymatic kinetics and tissue 84 distributions. A fifth putative hexokinase enzyme was recently discovered but has not been fully 85 characterized yet (10). A switch from GCK to HK2 isoenzymes is occurring during the transition 86 from primary to tumor hepatocytes so that HCC cell lines express HK2 but no longer GCK. HK2 87 expression level has been correlated with disease progression and dedifferentiation of HCC cells 88 (11). When HK2 is artificially knocked-down in HCC cell lines, glycolysis is repressed, and 89 tumorigenesis is inhibited while cell death increases (9). In addition, hexokinase function extends 90 beyond metabolism towards autophagy, cell migration, and immunity, suggesting that the GCK-91 to-HK2 isoenzyme switch has broader consequences than initially suspected (12-15). Here, we 92 analyzed transcriptomic data of HCC biopsies and correlated hexokinase isoenzyme expression 93 level with patient survival. This led us to generate a new human HCC model expressing GCK 94 instead of HK2. A comparative analysis of GCK + vs HK2 + HCC cell lines provided a unique 95 opportunity to look into HK isoenzyme-dependent metabolic features, lipoprotein production and 96 resistance to immune signals of liver cancer cells. 97

Results 98
Relative expression level of GCK and HK2 in HCC patients 99 Although an isoenzyme switch from GCK to HK2 has been observed during the carcinogenesis 100 process (16), whether hexokinase isoenzymes expression is predictive of patient survival is 101 performed a Spearman's rank correlation test (Fig. 1c). Patient survival was positively correlated 117 to GCK expression but inversely correlated to HK2 expression in line with the Kaplan-Meier 118 analysis. In addition, GCK and HK2 expression tends to be inversely correlated in tumor samples 119 (Fig. 1c). Therefore, there is a trend for mutual exclusion of GCK and HK2 expression in HCC 120 tumors, and this profile is associated to clinical outcome. Huh7-GCK + /HK2cell lines, respectively, was validated, while HK1 and HK3 were not expressed 126 (Fig. 2a). The hexokinase activity in the presence of increasing concentration of glucose was 127 determined in protein lysates from the two respective cell lines. Hexokinase activity in Huh7 lysate 128 reached its maximum at low glucose concentration, presenting a saturation curve according to 129 Michaelis-Menten kinetics (Fig. 2b). In contrast, the hexokinase activity in Huh7-GCK + /HK2 -130 lysates followed a pseudo-allosteric response to glucose (20, 21). Thus, the expected HK2 and GCK 131 activities were observed in the Huh7 and Huh7-GCK + /HK2cells respectively. The cell proliferation 132 capacity remained identical between the two cell lines ( Supplementary Fig. 1). We then compared 133 the genome edited Huh7-GCK + /HK2and the parental Huh7 cell lines (i.e. Huh7-GCK -/HK2 + ) at a 134 transcriptomic, metabolic and immunological level.  Table 2). Overall, 4.2% of the gene transcripts were reduced and 6.9% 139 were induced in Huh7-GCK + /HK2compared to Huh7 ( Fig. 2c; |fold-change (FC)|>2 and p 140 value<0.05). We first determined the metabolic consequences of the HK isoenzyme switch by 141 mapping the differentially expressed genes onto the a well-established bipartite metabolic 142 network Recon2, connecting gene products and metabolites ( Supplementary Fig. 2-4) (22, 23). 143 After trimming highest-degree metabolites as currency metabolites, clusters of genes that are 144 both differentially expressed and connected by common metabolites emerged. Interestingly, we 145 found that across a wide range of analysis parameters, including varying rates of currency 146 metabolites and gene expression fold-change, the differentially expressed metabolic genes are 147 substantially better connected than expected by chance (Fig. 2d). This highlights the specificity of 148 the transcriptomic changes with respect to metabolic pathways. The spanned network presented 149 in Figure 2e corresponds to a stringent fold-change threshold for transcriptomic data 150 (log2(|FC|)>3) while removing 2 percent of highest-degree currency metabolites. This network 151 shows connected components within glycolysis, but also across distant modules including the 152 gamma-aminobutyric acid (GABA) shunt (ALDH5A1), urea cycle (CPS1, OTC), glycogen metabolism 153 (GYS1, GYS2, AGL) and lipid synthesis (GPAM, AGPAT4, DGKG, CDS1, A4GALT) or degradation 154 (ACADL, HSD17B4, AMACR). This analysis highlights the global impact of the HK isoenzyme switch 155 that spreads beyond glycolysis across distant connected metabolic modules. 156 Enrichment of molecular and cellular functions in differentially expressed genes was also analyzed 157 using Ingenuity Pathway Analysis (IPA). This revealed that cellular movement and lipid 158 metabolism were the most affected functions (Fig. 3a and Supplementary Table 3). A closer look 159 at these annotations pointed to differences in the migratory capacities (Fig. 3b) as well as lipid 160 concentration and synthesis (Fig. 3c). The migratory capacities of Huh7 and Huh7-GCK + /HK2were 161 compared using transwell-migration cell assays (Fig. 3d- The intracellular lipid content of the two cell lines was further analyzed. In Huh7-GCK + /HK2 -, an 174 enrichment in phosphatidylcholine, cholesterol, triglycerides (TG) and free fatty acids was 175 observed compared to Huh7 (Fig. 4a). One major function of hepatocytes is to secrete triglyceride-176 rich VLDL and this function is altered in HCC cells that secrete smaller lipoproteins with the density 177 of LDL (24, 25). The secretion of lipids and lipoproteins by both cell lines was analyzed after a 24h-178 culture in the absence of FCS to exclude any participation of exogenous lipids in the production 179 of lipoproteins. Huh7-GCK + /HK2secreted more free fatty acids than Huh7 while secretion of 180 cholesterol and triglycerides (TG) remained unchanged (Fig. 4b). However, under the same 181 conditions, the secretion of apolipoprotein B (ApoB) by Huh7-GCK + /HK2was reduced compared 182 to Huh7. Since ApoB is a non-exchangeable protein with only one copy in VLDL and LDL particles, 183 an elevated TG/ApoB ratio indicates that ApoB + -lipoproteins secreted by Huh7-GCK + /HK2cells 184 are enriched in TG compared to those secreted by Huh7 (Fig. 4c). This was confirmed by the ApoB 185 distribution in density gradient fractions. As expected, lipoproteins secreted by Huh7 sediment at 186 the density of LDL, while those secreted by Huh7-GCK + /HK2 - (Fig. 4d) match the density of VLDL 187 found in human plasma or secreted by primary human hepatocytes in culture (26, 27). This 188 indicates that GCK expression is essential for the VLDL assembly/secretion pathway and could 189 explain the loss of this crucial metabolic pathway in hepatoma cells expressing HK2 instead of GCK 190 (28). Huh7-GCK + /HK2cells compared to Huh7 ( Fig. 5l and m). Even though SDH is also part of the 231 complex II of the mitochondrial respiratory chain, we observed that the overall oxygen 232 consumption was increased in Huh7-GCK + /HK2 - (Fig. 5n) with increased basal and maximal 233 respiration, ATP production and spare respiration capacity ( Supplementary Fig. 6). Functional 234 analysis of the respiratory chain showed that oxygen consumption in Huh7 and Huh7-GCK + /HK2 -235 cells was mainly dependent on complex I which is fueling complex III ( Fig. 5n and o). Thereby, the 236 HK isoenzyme switch rewired the TCA cycle promoting carboxylation of pyruvate into OAA in the 237 presence of a reduced SDH activity and increased respiration through complex I. 238 To investigate whether this differential sensitivity to RLR ligands is linked to GCK expression or 266 HK2 knockout, we used Huh7 cells transduced for GCK expression so that they express both HK2 267 and GCK (Huh7-GCK + /HK2 + ). In these cells, the response to RIG-I ligation did not differ from that 268 of Huh7 cells suggesting that GCK expression alone is not sufficient to restore immune sensitivity 269 (Fig. 6d). When HK2 expression was repressed in these cells with a shRNA (Huh7-GCK + /HK2 -Sh) 270 (see Supplementary Fig. 8 showing 95% extinction of HK2 protein), ISRE response to RIG-I signaling 271 was restored to a level similar to that observed in Huh7-GCK + /HK2cells (Fig. 6d). This is pointing 272 towards HK2 as a negative regulator of RLR signaling in HCC cells and suggests that the GCK-to-273 HK2 isoenzyme switch during malignant transformation of hepatocytes is accompanied by a 274 reduced sensitivity to innate immune signals. The higher sensitivity to RLR ligands of  GCK + /HK2cells also resulted in increased secretion of inflammatory interleukins (IL-6 and IL-8), 276 antiviral cytokines (IFN-λ1, IFN-λ2/3, IFN-β), and IP-10 (Fig. 6e), indicating that both NF-kB-and 277 IRF3-dependent signaling pathways were induced. IL-1β, TNFα, IL-12p70, GM-CSF, IL-10 and IFNγ 278 were not detected in the supernatants of none of the cell lines, whether they were stimulated or 279 not. 280 As natural killer (NK) cell-mediated lysis of tumor cells is crucial for the anti-cancer immune 281 defense, we compared the susceptibility of the two cell lines to NK cells cytotoxicity. Figure 6f  282 shows that Huh7 cells are resistant to NK cell-mediated lysis in contrast to Huh7-GCK + /HK2 -. Thus, 283 replacing HK2 by GCK restored Huh7 sensitivity to NK cell-mediated lysis. Similar results were 284 obtained when NK cells were pre-activated with IL-2 ( Supplementary Fig. 9). Altogether, these 285 results demonstrate that HCC cells expressing HK2 instead of GCK exhibit an impaired response 286 to immune signals and also a strong resistance to NK cells. Significantly, these two observations 287 are in line with clinical data showing that elevated GCK expression is associated with prolonged 288 survival, while elevated HK2 expression coinciding with GCK reduction correlates with shorter 289 overall survival (Fig. 1). 290

Discussion 291
Metabolic network rewiring is a hallmark of cancer although for most tumors, mechanisms at the 292 origin of this metabolic reprogramming have not been elucidated. While GCK, but not HK2, is 293 expressed in normal hepatocytes, the expression of HK2 occurs during cirrhosis and increases as 294 the disease progresses to carcinoma. Several signaling pathways such as hypoxia inducible factors 295 (HIF), peroxisome proliferator-activated receptors (PPAR) and phosphatidylinositol-4,5-296 bisphosphate 3-kinase (PI3K) might contribute to HK2 induction in fatty liver disease and its 297 evolution towards cirrhosis and carcinogenesis (35-37). Consequently HK2 induction has been 298 proposed as a risk marker of HCC development (16). Analyzing TCGA data from human HCC 299 tumors, we observed that not only high levels of HK2 but also low levels of GCK are of poor 300 prognosis. In contrast, neither HK1 nor HK3 expression levels were correlated with survival of HCC 301 patients. GCK expression is very low or not detected in biopsies from a majority of patients (65.8% 302 of patients show RSEM values <10), whereas HK2 is widely expressed (16) (only 5.8% of patients 303 show RSEM values <10). This probably explains that HK2 expression is a better prognostic marker 304 than GCK for HCC. However, when GCK and HK2 expression were combined into a single ratio, 305 this prognostic marker outperformed HK2 or GCK expression alone. This suggests that both HK2 306 induction and GCK loss play a role in HCC progression. As HK2 and GCK expression tend to be 307 does not induce a shift from LDL to VLDL density, indicating that intracellular fatty acid 318 accumulation of exogenous origin cannot rescue VLDL production (28). Here we show that 319 replacing HK2 by GCK in Huh7 cells restored de novo fatty acid synthesis, allowing VLDL 320 assembly/secretion in the absence of exogenous lipids. To our knowledge Huh7-GCK + /HK2is the 321 first human cell model with a functional VLDL secretion pathway. Such a tool will strongly benefit 322 the field of cardiovascular diseases and hepatic steatosis. 323 De novo fatty acid synthesis from carbohydrates requires an adequate supply in metabolic 324 substrates, especially citrate that is produced by the TCA cycle from incoming pyruvate. The 325 glycolytic entry point into the TCA cycle is controlled by PDH and PC that convert pyruvate into 326 acetyl-CoA or OAA, respectively. Our data revealed that in addition to the increased production 327 of pyruvate from glucose, PC activity is increased whereas PDH is inhibited. This suggests that 328 pyruvate metabolism is rebalanced in favor of OAA in Huh7-GCK + /HK2cells, as described in 329 healthy liver. Such a mechanism of anaplerosis is known to replenish TCA cycle intermediates and 330 compensate citrate export out of the mitochondria for lipogenesis fueling. Increased PC activity 331 is observed in both normal and pathological situations, mainly as a result of an increased 332 transcription of the PC gene. In our model, mRNA and protein levels were not affected, indicating 333 that PC activity can be regulated by alternative mechanisms depending on HK isoenzyme 334 expression. This may relate to lower levels of oxalate, a known inhibitor of PC activity, in Huh7-335 GCK + /HK2cells ( Fig. 5c and Fig. 7 discussed below). 336 A rebalanced pyruvate usage in favor of OAA is also described for instance in SDH-deficient 337 neuroendocrine tumor cells, where succinate accumulates and PC activity is increased to maintain 338 OAA production, replenish the oxidative TCA cycle and support aspartate synthesis (29). 339 Interestingly, in comparison to Huh7 cells, succinate and aspartate levels are elevated in Huh7-340 GCK + /HK2where SDH activity is reduced, suggesting a direct link between PC and SDH activity in 341 hepatocytes. Several mechanisms inhibiting SDH have been described (38). Modification of the 342 expression of SDH subunits is unlikely as no variation was observed at the transcriptomic level. 343 Itaconate is a weak inhibitor of SDH produced by IRG1 from aconitate but this metabolite was not 344 detected and IRG1 mRNA was absent from the transcriptome of both cell lines. Whether fumarate 345 or other metabolites are responsible for the reduced SDH activity in GCK-expressing cells remains 346 to be investigated. Finally, SDH-deficient cells and LPS-stimulated macrophages have been shown 347 to elicit a hypoxic-like phenotype through accumulation of large amounts of succinate and 348 stabilization of HIF-1α (39, 40). Despite an elevated succinate steady-state level in  GCK + /HK2compared to Huh7 cells, we observed no difference in HIF-1α stabilization neither at 350 basal level nor upon induction (Supplementary Fig. 10). This suggested that the reduction of SDH 351 activity in Huh7-GCK + /HK2cells was not strong enough to induce such a pseudo-hypoxic 352 phenotype. 353 Our gene-centric metabolic analysis of transcriptomic data revealed a wide spreading of 354 metabolic modifications resulting from HK isoenzyme switch. Illustrating these modifications, 355 Figure 7 is an attempt to integrate the observed changes in central carbon metabolism and closely 356 connected metabolic pathways. In particular, decreased level of alanine and increased aspartate 357 concentration in Huh7-GCK + /HK2cells could be an indirect effect of PC activation that uses 358 pyruvate for the synthesis of OAA. As a consequence, hepatic transaminases may balance 359 intracellular pools of OAA, aspartate, alanine and pyruvate. Glutamate and GABA levels were also 360 modified, thus supporting anaplerosis of the TCA cycle through glutamine consumption and the 361 GABA shunt pathway, respectively. We also observed lower levels of oxalate, an end-product of 362 glyoxylate degradation. In Huh7-GCK + /HK2cells, increased levels of the enzyme alanine-363 glyoxylate and serine-pyruvate aminotransferase (AGXT) could account for this phenotype as it 364 converts alanine and glyoxylate into pyruvate and glycine, which is also increased. Interestingly, 365 high level of AGXT is a good prognostic marker for HCC (41). Consistently, it was found that oxalate 366 inhibits liver PC, resulting in reduced gluconeogenesis and lipogenesis (42, 43). Thus, a higher PC 367 activity could be explained by lower levels of oxalate in Huh7-GCK + /HK2cells. We also observed 368 that isoleucine and valine levels increased while BCAT1 (branched chain amino acid transaminase 369 1) predominant transcripts decreased. This suggests a reduced catabolism of branched chain 370 amino acids in Huh7-GCK + /HK2cells. Again, low levels of BCAT1 is a good prognostic marker for 371 HCC and oral supplementation with branched chain amino acids has been shown to reduce the 372 risk of liver cancer in cirrhotic patients (44, 45). If some metabolic modifications seem to advocate 373 for the restoration of a normal hepatocyte phenotype following the replacement of HK2 by GCK, 374 it cannot be a general statement. Indeed, the urea cycle was also impacted in Huh7-GCK + /HK2 -375 cells with lower levels of CPS1 and OTC, which are also observed in aggressive HCC tumors (46). 376 Altogether, our results demonstrate the broad impact of replacing HK2 by GCK in HCC cells, and 377 the key role played by the HK isoenzyme switch in HCC tumor metabolism. 378 We discovered that HK isoenzyme expression not only controls hepatic metabolic functions but 379 also interferes with intrinsic innate immunity of hepatocytes and antitumor immune surveillance. Here we show that Huh7 cells expressing GCK instead of HK2 exhibit a higher sensitivity to RIG-I 387 and MDA5 ligands, and produce higher levels of type I/III IFNs and inflammatory cytokines. This 388 immune phenotype occurs in a context of reduced SDH activity and increased intracellular content 389 in succinate (Fig. 5l-m). A pro-inflammatory function of immune cells such as macrophages was 390 previously linked to TCA rewiring, with reduced SDH activity resulting in succinate accumulation 391 (39, 50, 51). Succinate can also be secreted from LPS-activated macrophages and activate its 392 cognate receptor, succinate receptor 1 (SUCNR1, previously known as GPR91) in an autocrine and 393 paracrine manner to further enhance production of IL-1β (52). Interestingly, glucose metabolism 394 promotes RIG-I and MDA5 signaling through the O-GlcNAcylation of the mitochondrial adaptor 395 MAVS (47). Thus, an intriguing hypothesis is that GCK expression could facilitate MAVS signaling 396 by increasing UDP-GlcNAc through upregulation of the hexosamine biosynthetic pathway. HK2 397 binding at the surface of mitochondria may also compete with pyruvate carboxylase, metabolites 398 or mitochondria factors known to control MAVS signaling (47-49). Here we show that HK2 399 knockdown promotes RIG-I-induced ISRE-dependent transcription (Fig. 6d). This is consistent with Cells were seeded in 96-well or 24-well plates. After 24 h, cells were co-transfected with indicated 507 doses of the RIG-I ligand 3p-hpRNA or the MDA5/TLR3 ligand poly(I:C) HMW together with the 508 pISRE-luc (1.25 µg/ml) and pRL-SV40 (0.125 µg/ml) reporter plasmids using the JetPEI-Hepatocyte 509 reagent (Polyplus Transfection). Manufacturer's instructions were followed. After 48 h, 510 supernatants were collected for cytokine quantification. Firefly and Renilla luciferase expressions 511 within cells were determined using the Dual-Glo luciferase Assay system (Promega) and an Infinite 512 M200 microplate reader (TECAN). 513

Cytokine assays 514
Clarified culture supernatants were collected and stored at -20°C. IL-8 was quantified using the 515 Cytometric Bead Array for human IL-8 (BD Biosciences). Other cytokines were assayed using the 516 LEGENDPlex multiplex assay (Human Anti-Virus Response Panel, BioLegend). Fluorescence was 517 analyzed using a FACS Canto II (BD Biosciences). 518 NK cell cytotoxicity test 519 NK cells were isolated as described in Supplementary Methods from human buffy coats of healthy 520 donors obtained from the Etablissement Français du Sang. Informed consent was obtained from 521 donors and experimental procedures were approved by the local institutional review committee. 522 Huh7 or Huh7-GCK + /HK2were seeded at 1x10 5 cells per well in a 24-well plate in RPMI-1640 523 (Gibco) with 10% FCS and 40 µg/ml gentamycin. After 24 h, 3x10 5 or 3x10 6 NK cells were added 524 to the culture wells. The cytotoxicity assay was performed for 4 h at 37°C, under 5% CO2. Target 525 hepatoma cells were harvested after trypsination, labelled with propidium iodide (PI) and 526 analyzed by FACS. Cell death was monitored after morphological gating on hepatocytes. 527

Statistics and reproducibility 528
All the statistical analyses were performed with GraphPad Prism or Analyse-it softwares. Details 529 of statistical analyses can be found in figure legends. 530 Acknowledgments 531 532 We