Septin 9 induces lipid droplets growth by a phosphatidylinositol-5-phosphate and microtubule-dependent mechanism hijacked by HCV

The accumulation of lipid droplets (LD) is frequently observed in hepatitis C virus (HCV) infection and represents an important risk factor for the development of liver steatosis and cirrhosis. The mechanisms of LD biogenesis and growth remain open questions. Here, transcriptome analysis reveals a significant upregulation of septin 9 in HCV-induced cirrhosis compared with the normal liver. HCV infection increases septin 9 expression and induces its assembly into filaments. Septin 9 regulates LD growth and perinuclear accumulation in a manner dependent on dynamic microtubules. The effects of septin 9 on LDs are also dependent on binding to PtdIns5P, which, in turn, controls the formation of septin 9 filaments and its interaction with microtubules. This previously undescribed cooperation between PtdIns5P and septin 9 regulates oleate-induced accumulation of LDs. Overall, our data offer a novel route for LD growth through the involvement of a septin 9/PtdIns5P signalling pathway.

L ipid droplets (LDs) are highly dynamic organelles generated from the endoplasmic reticulum (ER) 1 , which can remain attached to the ER or generate contacts with other organelles, including the mitochondria, lysosomes, peroxisomes and Golgi 2 . LDs consist of a hydrophobic core of neutral lipids, which are primarily composed of triacylglycerols (TAGs) and cholesterol esters, surrounded by a monolayer of phospholipids and a large group of proteins that facilitate cellular signalling interactions, control the access of metabolic enzymes, and influence LD movement within the cell 1,3 . These proteins include predominantly the perilipin family (Plins) and proteins of the rab family of GTPases 1,4 . There are five known perilipins (PLINs 1-5) expressed in hepatocytes, which play distinct roles. PLIN2 also called ADRP (adipose differentiation-related protein) is a prominent LD protein, expression level of which is tied to LD accumulation. PLIN3 initially described as a tail-interacting protein of 47 kDa (TIP47) is also ubiquitously expressed 5 . LD growth and degradation are highly regulated processes and excessive accumulation of LDs in the liver causes steatosis, which may contributes to cirrhosis, the final stage for several chronic liver diseases and often evolves to hepatocellular carcinoma.
Chronic HCV infection is a major public health problem and a main risk factor for cirrhosis, and hepatocellular carcinoma, thus remaining the major cause of liver resection and transplantation 6 . There is no vaccine to protect against HCV, although major advances have been recently achieved regarding HCV infection treatment, and the combination of drugs, including protease inhibitors, represents a real breakthrough that cure almost 90% of infected patients. However, these treatments have remained very costly. Additionally, understanding the key mechanisms driving HCV replication and persistence will have a profound impact, beyond HCV, for the appraisal of the pathogenesis of this family of viruses. Each step of the replication cycle of HCV is tightly dependent on the host lipid metabolism, and liver steatosis associated with the accumulation of LDs is a hallmark of chronic HCV infection 7,8 . HCV and LD interactions are crucial for persistent viral propagation and virion production 9 .
HCV is an enveloped positive single-stranded RNA virus from the Flaviviridae family. HCV genomic RNA encodes for a polyprotein precursor of B3,100 amino acids post-translationally cleaved by cellular and viral proteases into the structural (core, E1 and E2), non-structural (NS) proteins (NS2, NS3, NS4A, NS4B, NS5A and NS5B) and p7 protein 8 . The core protein plays a major role in the virion assembly through interaction with LDs 10 . The proteins covering the LD surface and their link with HCV life cycle have been characterized, whereas the potential role of phospholipids in these events remains unclear. The composition of the phospholipid monolayer of LDs depends on the cell type, but is mostly composed of phosphatidylcholine, phosphatidylethanolamine, and, to a lesser extent, phosphatidylinositol (PtdIns) and lysophospholipids 1,11,12 . The phosphorylated derivatives of PtdIns (PIs) are important signalling molecules that are essential for a variety of cellular functions, including membrane remodelling and trafficking 13 . Although recent studies revealed the role of PtdIns 4-kinases (PI4K-IIIa and PI4K-IIIb) and their lipid products, PtdIns4P, as critical regulators of the HCV life cycle 8 . The role of class II phosphoinositide 3-kinase (PI3K) was also demonstrated 14 . Furthermore we demonstrated that PIs control epithelial morphogenesis and polarity 15,16 and represent HCV cellular targets 17 .
Septins form a family of GTP-binding proteins composed of 13 members in mammals and conserved from yeast to humans 18 . The complexity of this gene family is increased by the existence of alternative splicing, which markedly increases the number of potential isoforms. Septins assemble into complexes variable in size and composition between organisms 19,20 . In drosophila and mammalian septins, complexes contain in general three to four different septins present in two copies. The core heterooligomeric units of six to eight subunits serve to build higher order structures, including filaments and rings structures, which are important for septin in vivo biological function 20 . Septins are implicated in multiple cellular functions, including cytokinesis, ciliogenesis, cell migration, vesicle trafficking, and cell polarity [21][22][23] . Septins associate with the cytoskeleton and are known as a fourth type of cytoskeletal structure 23 . Septins also bind to membranes, specifically to PIs, providing membrane stability and serving as diffusion barriers for membrane proteins 20,21,23 . Nevertheless, these septin/PI interactions have been scarcely investigated [24][25][26] and their link with HCV life cycle and their potential role in LD biogenesis and growth remain unclear. Septins are also involved in infection by pathogens, including bacteria, fungi 23,27,28 and HCV virus 29 . Alteration in expression profile of septins are reported in different cancers and we determined that septin 9 is specifically upregulated in liver cancer 30 .
Despite being involved in carcinogenesis, septin 9 expression and function have not been investigated in precancerous diseases such as cirrhosis. In this study, we investigate the expression of septin 9 and the differential expression of the other septins in HCV-induced cirrhosis. We also demonstrate that septin 9 regulates LD growth through binding to PtdIns5P in HCVinfected cells. Of interest, we show that this new mechanism is involved in lipid homoeostasis independently of HCV infection.

Results
Transcriptomic analysis of septins in human cirrhosis. To determine the expression profile of septin 9 in cirrhosis, we used the GSE14323 data set 31 , including transcriptomes of normal liver (n ¼ 19) and HCV-induced cirrhosis samples (n ¼ 49). Data revealed a significant upregulation (two-sided Student's t-test, P ¼ 0.012) of septin 9 transcriptional expression in cirrhosis samples compared to that in normal liver samples (Fig. 1a).
Septin 9 occupies a terminal position in an octameric septin complex involved in the formation of higher order structures such as filaments and rings (Fig. 1b) 32 . According to the importance of hetero-oligomerisation in the assemblies and functions of septins 23 , we assessed the differential expression of septins in HCV-induced cirrhosis using the GSE14323 data set by unsupervised principal component analysis ( Fig. 1c-f). Sample projection on the first factorial map showed a significant separation between the normal liver and HCV-induced cirrhosis samples according to the first axis (P ¼ 2.67E À 15) (Fig. 1c). Learning machine algorithm of Random Forest confirmed a good separation of the normal liver and cirrhosis samples. The classification error by class was low and especially null for the cirrhosis class after building 500 classification trees and multidimensional Scaling plot (MDSplot), confirming the good separation of the samples after learning machine analysis ( Fig. 1d; Supplementary Fig. 1). Random Forest analysis predicted that most septins are affected in HCV-induced cirrhosis when compared to normal liver, while septin 10 was less affected. This was confirmed by individual boxplot performed on each septin ( Fig. 1e, P value were calculated by two-sided Student t-test). These analyses showed that the expression of septin 7 and 2 is massively affected in these samples (Fig. 1d,e). A weak error on misclassification by class (confusion of 1.67%) was detected when using the learning machine algorithm ( Supplementary Fig. 1c). Thus, we performed an unsupervised classification on these samples with previously used septin expression (classification tree with Euclidean distance and complete linkage (Fig. 1f)). This unsupervised algorithm performed on 58  Normal liver HCV-cirrhosis the discrimination of two major groups of experiments with no error of misclassification between normal liver and HCV-induced cirrhosis samples. This result suggests that the expression of septin protein family is deregulated in HCV-induced cirrhosis as compared with that in the normal liver.
Following the above described transcriptomic study, we hypothesized that HCV could be responsible of the upregulation of septin 9 in cirrhosis. To explore this possibility, Huh7.5 cells were inoculated for 24, 48 and 72 h with HCV JFH-1 (Japanese fulminate hepatitis 1), a genotype 2a strain, which produces infectious particles in culture and recapitulates all steps of the HCV life cycle 33 . Immunoblot data revealed a non-significant, time-dependent increase of septin 9 expression in non-infected cells, while septin 9 expression was significantly enhanced in JFH-1-infected cells (Fig. 2a), with the highest expression at 72 h of infection (Fig. 2a) when septin 9 formed bundles of filaments surrounding the clusters of HCV core (Fig. 2b). Treatment of Huh7.5 cells with two different septin 9 siRNAs (si1 and si3) before their infection strongly reduced core expression (Fig. 2b,c). These data suggested that HCV-regulated septin 9 assembly and expression was required for core protein expression. Thus, to further characterize the septin 9 filamentous structure, we determined the presence of other septins. We focussed on septin 2, the major ubiquitously expressed septin and the most deregulated in HCV-induced cirrhosis (Fig. 1). In non-infected cells, septin 2 formed short filaments that co-localized with septin 9 (Fig. 2d,e). These filamentous structures containing septin 2 and their co-localization with septin 9 were significantly increased upon JFH-1 infection (Fig. 2d-f). Depletion of septin 9 using si3 disrupted septin 2 filaments and reduced its staining when compared with that in control cells in both non-infected and JFH-1-infected cells (Fig. 2d,e). These data were confirmed by immunoblot analysis (Fig. 2g) and suggested the role of septin 9 in the formation of septin filamentous structure, consistent with previous reports 32 .
Next, we assessed the effects of septin 9 expression on HCV-modulated microtubule (MT) network. Tubulin staining indicated a strong MT network in HCV-infected cells when compared with that of non-infected cells (Fig. 2h,i; Supplementary Fig. 2a). Importantly, MTs co-localized with septin 9 filaments and si3 severely destroyed MT filaments ( Fig. 2h-j), indicating a regulatory role of septin 9 in HCV-dependent MT organization.
Septin 9 regulates LD behaviour and HCV replication. The accumulation of LDs is an important feature of HCV infection 4 . Thus, to understate the importance of septin 9 in HCV life cycle, we investigated LDs. JFH-1 infection sharply increased LDs that formed clusters co-localized with the core protein around the nucleus (Fig. 3a), while si3 treatment markedly decreased LDs (Fig. 3b) and their co-localization with core ( Fig. 3c,d). A typical profile of LD distribution was presented (Fig. 3e,f). In noninfected cells, about 70% of LDs were present in the perinuclear area and si3 treatment decreased the percentage to 55%. JFH-1 infection increased the perinuclear LDs up to 92% and septin 9 depletion decreased this level to 62% (Fig. 3g). These changes in the level and distribution of LDs were accompanied by a 2.5-fold increase of LD size in JFH-1-infected cells, while si3 decreased the LD size by half (Fig. 3h). Similar changes were observed for LD number (Fig. 3h). PLIN2 is among the best studied protein associated with LDs and is considered as a marker of LDs 34 . The strong increase in PLIN2 expression and its co-localization with core observed in JFH-1-infected cells were abolished by si3 treatment (Supplementary Fig. 3) and emphasized the role of septin 9 in HCV-induced LD accumulation. Finally, we revealed a strong decrease of HCV genomic RNA after septin 9 knockdown (Fig. 3i), suggesting a critical role of septin 9 in HCV replication. We also analysed the transcript level of the five well-characterized variants of septin 9 (i1 to 5) after JFH-1 infection. Isoform 1 mRNAs were the most abundant, while isoform 3 mRNAs were almost undetectable ( Supplementary Fig. 2b). Accordingly, the following studies focused on septin 9 isoform 1 (Septin 9_i1) and we used Huh7 cells stably expressing HCV full-length genomic replicon from genotype 1b (Huh7R) 35 . This system permitted the efficient expression of all viral proteins and is a convenient and valuable tool for molecular studies. A decrease in both the core and envelope E2 proteins was observed in Huh7R cells transfected with si3 ( Supplementary Fig. 4a). By opposite both core and E2 expression increased in cells expressing septin 9_i1 compared with control cells (Supplementary Fig. 4b). Later, septin 9_i1 and the LD characteristics were assessed by immunofluorescence analysis. Septin 9_i1 assembled in filaments surrounding the large clusters of LD and transfection of si3 inhibited LD accumulation (Fig. 4a). Overall, profiles of LD distribution and changes in LD morphology (Fig. 4b,c) were similar to the results obtained using the JFH-1 strain (Fig. 3). Septin 9_i1 filaments co-localized with endogenous septin 2 ( Supplementary Fig. 5a), indicating that septin 9_i1 is integrated in a hero-oligomeric structure composed of other septins consistent with data in Fig. 2. Moreover, the ratio of the intensity of exogenous septin 9_i1 revealed with an anti-V5 tag antibody to the intensity of endogenous septin 9 detected with septin 9 antibody was 1.1 ( Supplementary Fig. 5b). Thus, we concluded that septin 9_i1 expression was similar to that of endogenous septin 9.
Although the first evidence for HCV interaction with LDs was obtained from a study on the core protein, recent studies indicated that the interaction of the HCV non-structural protein NS5A with PLIN3 is essential for the recruitment of NS5A on LDs to promote the release of HCV particles 36,37 . Interestingly, increases in PLIN3 and NS5A expression were observed in Huh7R cells expressing septin 9_i1 ( Supplementary Fig. 6a). These results were confirmed by immunoblot ( Supplementary  Fig. 6b) indicating that septin 9_i1 regulated LDs growth and HCV protein expression.
Septin 9 regulates neutral lipid metabolism. LD growth can be mediated by several mechanisms, but, typically, its size expansion involved an increase in the neutral lipid content, which essentially consists of TAG and cholesterol ester 5 . Therefore, we analysed the neutral lipid content of septin 9_i1-expressing cells. Transfection of septin 9_i1 in Huh7R cells induced a threefold increase of TAG and correlated with a decrease of diacylglycerol (DAGs) (Fig. 4d). As a consequence, the ratio of TAG to DAG was greatly increased (Fig. 4e). The fatty acid composition of TAG and DAG was also analysed. The composition of these two lipids was very similar and consisted mostly of C16 and C18 long chains (Fig. 4f). Similar data were obtained using septin 9-depleted Huh7R cells ( Fig. 4g-i). A significant TAG decrease and an increase in DAG were observed (Fig. 4g), suggesting a conversion of the TAG pool into DAGs.
Diacylglycerol acyltransferase-1 (DGAT1) is one of the two known DGAT enzymes that catalyse the final step in triglyceride biosynthesis 38 . The role of DGAT1 in the trafficking of the HCV core protein to LDs and its involvement in the production of infectious HCV particles were reported 39,40 . Thus, we analysed DGAT expression and we observed a significant increase in the total intensity of the DGAT1 signal in septin 9_i1-expressing cells. Moreover, DGAT1 co-localized with septin 9_i1 filaments, as shown at the highest magnification (Fig. 4j). Together, these data indicated that septin 9 could also participate in LD growth by regulating lipid biosynthesis.
MTs are required for septin 9-induced LD behaviour. Different reports indicated that MTs play a role in the dynamics and growth of LDs 41,42 . Furthermore, septins interact with MTs and are essential for MT-dependent cell processes 12,23,43    ARTICLE we postulated that MTs could play a role in septin 9-induced LD growth. Consequently, we performed a nocodazole washout (Fig. 5). Treatment of Huh7R cells expressing septin 9_i1 with nocodazole resulted in the dispersion of LDs throughout the cell and decreased their size. One hour after nocodazole removal, LDs started to gather around the nucleus and increased in size. These changes were accompanied by the partial recovery of the microtubule filamentous structure (Fig. 5a). Finally, LDs were restored to their initial size and relocated around the nucleus 3 h after nocodazole removal (Fig. 5b-d). Taken together, these results indicate that MT dynamics are required for septin 9 to induce LD growth and perinuclear clustering.
PtdIns5P binds to septin 9 and promotes LD growth. To further identify the mechanism underlying septin 9 assembly and its role on LD, we assessed the importance of phosphoinositides (PI). The phospholipid monolayer that surrounds LDs plays a role in the regulation of LDs morphology. Phosphatidylcholine stabilizes the tension and reduces LD growth, while phosphatidic acid contributes to LD expansion 44 . PtdIns was also detected on LD surface 11 , however, the presence and the role of its phosphorylated derivatives, PIs, on LD growth have not been investigated. Therefore, we identified the PIs that bind to septin 9 by performing a protein-lipid overlay assay using a glutathione-S-transferase (GST)-tagged recombinant protein (Fig. 6a). Data from the assay revealed that septin 9 bound specifically to monophosphorylated PtdIns (PtdIns3P, PtdIns4P and PtdIns5P) and the strongest interaction was observed with PtdIns5P while no signal was detected using GST protein alone as a negative control (Fig. 6a), supporting the specificity of the observed binding signals. Furthermore, we showed that PtdIns3P, PtdIns4P and PtdIns5P promoted a significant increase in LD size compared with non-treated cells, whereas PtdIns(4,5)P2 evaluated as a non-binding septin PI, had no significant effect (Fig. 6b,c). PtdIns5P was the most efficient (Fig. 6b,c), consistent with the results in Fig. 6a. Moreover, septin 9 assembled surrounding the LDs that became more visible upon the addition of PtdIns5P (Fig. 6b). Overall, these data revealed the crucial role of monophosphorylated PIs in the expansion of LDs and the organization of septin 9 high-order structures. Thus, we focused on PtdIns5P, which was the most effective on LDs. PtdIns5P specifically binds to the plant homeodomain (PHD) finger found in many chromatin-remodelling proteins, and the PHD domain of the tumour suppressor inhibitor of growth (ING2) was proposed to function as a receptor of PtdIns5P (ref. 45). We used the biotinylated GST-tagged PHD domain of ING2 as a probe to visualize cellular PtdIns5P, as previously described 46 . As expected, compared with untreated cells, PtdIns5P addition strongly increased both the PHD signal and the LD size and confirmed that addition of exogenous PtdIns5P increased its intracellular level ( Supplementary Fig. 7a-c). To validate the role of PtdIns5P on LDs, we used another approach based on IpgD, the virulence factor of Shigella flexneri, which is a PtdIns(4,5)P2-4-phosphatase responsible for the profound increase of PtdIns5P in S. flexneri-infected cells 47 . Transfection of Huh7R cells with GFP-tagged-IpgD cDNA (GFP-IpgD) increased PtdIns5P signal (Supplementary Fig. 7d) and LD size and accumulation (Fig. 6d).
PIKfyve regulates septin 9 filaments and LD accumulation. To further study the interaction between septin 9 and PtdIns5P and its importance in LD accumulation, we used YM201636, a selective inhibitor of FYVE finger containing phosphoinositide kinase (PIKfyve), a major contributor to the intracellular production of PtdIns5P (ref. 48). First, the effect of YM201636 was verified by PtdIns5P immunostaining in Huh7R cells (Fig. 7a). Subsequently, the cells were transfected either with an empty vector or a septin 9_i1 construct and treated with YM201636. Moreover LD size and level increased in cells expressing septin 9_i1 and were surrounded by septin 9_i1 filaments as described in Fig. 4. YM201636 treatment profoundly reduced the LD size (Fig. 7b,c and Supplementary Fig. 8a) and decreased TAGs concomitantly with an increase of DAGs ( Supplementary Fig. 8b-d). Interestingly, YM201636 treatment disrupted septin 9_i1 ( Fig. 7b; Supplementary Fig. 8a) and MT filaments and prevented both co-localization (Fig. 7c). Furthermore, PtdIns5P addition to Huh7R cells transfected with septin 9 siRNA partly rescued the disruption of MT filaments and LD clustering caused by septin 9 depletion ( Supplementary  Fig. 9). Thus, these data demonstrated that PtdIns5P is a crucial regulator of septin 9 structural and functional features.
Interestingly our data revealed that both septin 9 and HCV-regulated PtdIns5P levels. Indeed infection of Huh7.5 cells using JFH-1 particles increased PtdIns5P ( Supplementary  Fig. 10a). Moreover PtdIns5P increased in Huh7R cells compared with naive Huh7 cells and transfection of the later cells with septin 9_i1 cDNA significantly enhanced the PtdIns5P signal ( Supplementary Fig. 10b,c).
Septin 9 PBR is required for its assembly and LD growth. Septins bind PIs in vitro via the polybasic region (PBR) near their N terminus 26,49 . The conserved sequence encompassing residues aa289 to aa294 is boxed in Fig. 8a. Thus, as another approach to assess the importance of PI for septin 9 structural and functional features, we generated a PBR-deleted mutant (septin 9_del1) from the cDNA of V5-tagged septin 9_i1. The recombinant proteins from the entire septin 9_i1 and the deleted construct were produced in Escherichia coli, purified to near homogeneity, and analysed by SDS gel electrophoresis (Fig. 8b). Coomassie blue staining revealed a major band of the expected molecular weight (70 kDa) for both septin 9_i1 and the deleted mutant, while some lower molecular weight species co-purified with septin 9_i1 (Fig. 8b). Immunoblot (Fig. 8c) confirmed that all detected protein bands were recognized by a V5-tag antibody. Purified recombinant proteins were used to perform a lipid overlay assay. Septin 9_i1 bound PtdIns3P, PtdIns4P, and PtdIns5P as expected, although a signal was observed with phosphatidic acid and PtdIns(3, 5)P2. The lipid-binding signal was strongly reduced for septin 9_del1 (Fig. 8d). Subsequently Huh7R cells 35 were transfected with the septin 9_i1, septin 9_del1 cDNAs, or empty vector (EV). Septin 9_i1 formed filaments whereas septin 9_del1 could not form filaments (Fig. 8e). Expression of septin 9_i1 increased LD in the perinuclear region compared to EV-transfected cells (Fig. 8e,f). In contrast, septin 9_del1 expression significantly reduced the total level of LDs and their perinuclear distribution (Fig. 8e,f). These changes in the LD distribution were accompanied by significant decrease of the LD size and number (Fig. 8g). Furthermore septin 9_i1 filament structures co-localized with MTs. By contrast, the mutant septin 9_del1 lost the filamentous structure and co-localization with MTs. The Pearson co-localization coefficient (Rr) between septin 9 and MTs dropped from 0.56±0.01 in cells expressing septin 9_i1 to 0.17 ± 0.03 in cells expressing septin 9_del1 and the RGB line profile showed a similarity between MTs and septin 9_i1 cellular distribution (Fig. 8h). These results indicate that PBR is necessary for the formation of septin 9 filamentous structures and LD growth. Moreover, PBR is involved in MTs organization and highlighted the critical role of septin 9 in cytoskeleton organization.
Septin 9 regulates LD growth in naive cells. Unravelling this new role of septin 9 in LD behaviour led us to examine its relevance independently of HCV infection. We performed a set of experiments using naive Huh7 cells. Septin 9_i1 expression significantly increases LD size comparing to EV-transfected cells ( Supplementary Fig. 11a,b). Additionally, neutral lipids analysis revealed an increase TAG in septin 9_i1-transfected cells (Supplementary Fig. 11c,d). Expression of GFP-IpgD  Fig. 11e,f) or the addition of exogenous monophosphorylated PIs to huh7 cells increase LD size ( Supplementary  Fig. 12a). In contrast, septin 9_del1 transfection or treatment with YM201636 in septin 9_i1-transfected cells significantly decrease LD size ( Supplementary Fig. 12b), disrupted septin 9 filamentous structure and its co-localization with MTs ( Supplementary  Fig. 13a-c). Similar effects were observed on actin filaments in septin 9_del1 expressing Huh7 cells (Supplementary Fig. 13d). Taken together, these results indicate the fundamental role of septin 9 and PtdIns5P in LD accumulation and cytoskeleton dynamic independently of HCV infection.
To emphasize the function of septin 9 and PtdIns5P in lipid  homoeostasis, Huh7 cells were treated with sodium oleate (0-200 mM), commonly used to induce LD formation. Neutral lipid analysis revealed a concentration-dependent increase of TAG and a decrease of DAG due to the addition of sodium oleate, validating the effect of sodium oleate (Supplementary Fig. 14). The immunofluorescence data indicated a dose-dependent enlargement of the LDs and an increase of endogenous septin 9. Remarkably, in cells supplemented with 200 mM sodium oleate, septin 9 filaments were visible around the very large clusters of LDs (Fig. 9a). Immunoblot data confirmed the gradual increase of septin 9 expression with the sodium oleate concentration and a concomitant increase of PLIN2 expression (Fig. 9b).
To further investigate the association of septin 9 with LD formation, we performed membrane flotation assays 37 (Fig. 9c,d) using Huh7 cells and the same cells treated with 100 mM sodium oleate as described in methods. Twenty-five fractions were collected and assessed by western blotting for septin 9, septin 2. PLIN2 and calnexin were analysed as markers of LDs and ER, respectively. The lowest-density membrane fractions (fractions 1-8) are called 'LD-rich' and the fractions (9-15) represented the 'ER/endosome-rich'fractions, while the fractions (16)(17)(18)(19)(20)(21)(22)(23)(24)(25) contained soluble and aggregated proteins. In non-treated Huh7 cells, septin 9 was mainly found in the ER/endosome-rich and the soluble/aggregated protein fractions, but not within low-density LD-rich fractions (Fig. 9c, upper panels). Similar distribution profiles were observed for septin 2 and calnexin, while PLIN2 was detectable in 'LD-rich' fractions. These data suggested that, in non-treated cells, septins are mainly associated with LDs in the ER. This distribution changed in the presence of sodium oleate, septin 9 and septin 2 increased in low-density LD-rich fractions with high expression of PLIN2 and decreased from the ER/endosome-rich fractions totally depleted of PLIN2. There was no visible change in the distribution profile of the different proteins among the soluble/aggregated protein fractions. Overall, these results suggested that septin 9 and septin 2 are associated with the ER site of LD biogenesis and sodium oleate promoted LD biogenesis and enhanced septins association with LDs. Subsequently, we explored the role of septin 9 and PtdIns5P on oleate-induced LD accumulation. Transfection with septin 9_i1 cDNA of Huh7 cells treated with 100 mM sodium oleate increased the LD size, while transfection with the septin 9_del1 construct or with si3 septin 9 (Fig. 10a,b) and YM201636 treatment (Fig. 10c), produced the opposite effect, indicating that septin 9 and PtdIns5P regulated oleate-induced LD biogenesis.

Discussion
In this study, we showed a deregulation of septin expression in HCV-induced cirrhosis compared with that in the normal liver. Subsequent studies using HCV-infected cells revealed an unexpected role for septin 9 in the control of cellular lipid homoeostasis through its binding to PtdIns5P required for microtubules stabilization and LD perinuclear accumulation. As depicted in our model in Fig. 10d this new role of septin 9 was also observed independently of the presence of HCV in naïve cells and in sodium oleate-induced LD accumulation. Additionally, septin 9 regulates HCV replication and assembles around the perinuclear clusters of the core in HCV-infected cells. Septin 9 promotes the perinuclear recruitment of proteins coating the LD surface such as PLIN2 and PLIN3 as well as the HCV core and the non-structural NS5A. Furthermore, PLIN3 playing an essential role in the HCV replication process 37 .Thus, one could speculate that HCV recruits septin 9 to control LD growth and creates a lipid-enriched environment, facilitating HCV replication. Septins are crucial cytoskeletal components that are exploited by many pathogens such as S. flexneri, Listeria monocytogenes 23 , Candida albicans 50 and Chlamydia trachomatis 27 . Moreover, septin 6 controls HCV replication via a direct interaction with a RNA binding protein, acting as a scaffold protein to facilitate the positioning of the HCV replication complex 29 . However, the roles of the LDs were not investigated in these studies related to septin/pathogen interactions. Here we demonstrated that IpgD, the virulence effector of S. flexneri increased the cellular content of PtdIns5P and induced LD growth. Even though this result confirmed the role of PtdIns5P in LD growth and accumulation, its enabled speculation on the potential role of LDs in infection by S. flexneri. Therefore, the findings of this study can be extended to other types of pathogens beyond HCV.
Despite the numerous publications about LD biology, several questions related to their biogenesis, growth and dynamics remain unanswered 1,4,44 . LD formation likely occurs at the ER, and different mechanisms are involved in LD growth, including growth from nascent LDs in the ER or in the vicinity of the ER via the incorporation of TAGs and phospholipids locally synthesized into LDs. These proposed mechanisms arose from the observations that key enzymes of phospholipid and neutral lipid synthesis are present on the LD surface 44 . LDs might also expand via fusion of existing LDs. These fusion processes include the absorption of small LDs by large LDs mediated by Fsp27 (a 27 kDa fat-specific protein) enriched at the contact site between LDs 51,52 . Fusion mechanisms can also be mediated via phospholipid metabolizing enzymes that produce fusogenic compounds at the LD surface 53,54 .This fusion between LDs involves the SNARE proteins 55 . In this study, we demonstrated that septin 9 significantly regulates the cellular content of TAG, a neutral lipid component essential for LD biogenesis. Furthermore, septin 9 promotes perinuclear accumulation of neutral lipid synthesis enzymes such as DGAT1 and regulates the expression of PLIN2 and PLIN3, two PLINfamily proteins crucial for LD formation. These data support a role of septin 9 in LD growth and biogenesis.
Nonetheless, we cannot rule out a role for septin 9 in LD fusion mechanisms. LD fusions are MT-dependent processes 42 . In this study, we demonstrated the requirement for dynamic MTs for septin 9-induced LD growth. Furthermore, LD fusion and size is related to the composition of the surrounding monolayer of phospholipids, including PtdIns. Interestingly, we demonstrated that septin 9 binds to mono-phosphorylated PtdIns, which is consistent with previously published data for Saccharomyces cerevisiae 49 . These PIs promote LD growth and are reminiscent of the role of PtdIns4P in the homoeostasis of LDs in yeasts 56,57 .
PtdIns5P is mainly present in the plasma and intracellular membranes, including the smooth ER and Golgi 58 , and remains the least characterized among the eight different PIs 59 . Notably, PtdIns5P regulates septin 9 assemblies in high-order structures. PtdIns5P partly rescued the loss of microtubule filaments induced by septin 9 knockdown. These results are consistent with the reported role for PIKfyve in microtubule-dependent vesicular trafficking 60 and strongly indicate a new role for PtdIns5P as a partner of septin 9 in the regulation of MT and LD dynamics, providing new insights on PtdIns5P biological role. Consequently, we proposed that septin 9 could act as a scaffold that establishes contacts between MTs and LDs via binding to PtdIns5P that coats the surface of LDs and promotes LD growth and intracellular movement.
These results revealed an unprecedented role for PIs in LD biology in mammals, although a detailed mechanism awaits further investigations. PIs are important regulators of membrane trafficking in coordination with small GTPase proteins such as ARF GTPase, Rab GTPases, and the subunits of the COPI complex. These different components are present on the LD surface 61,62 . In addition, LDs interact with different cellular organelles, including the ER, Golgi and mitochondria 63 . Therefore, the existence of a different pool of LDs with a specific PI and Rab GTPase composition could be postulated and might help define LD identity as for endosomal compartments 64 . Thus, a study of the contribution of PI signalling to LD biogenesis and growth is of great interest in LD biology.  In conclusion, this study provides new insights for research on the increasing pathologies associated with lipid metabolism disorders and cancer.
Cell treatments. Cells treatment with sodium oleate: complexes of sodium oleate with BSA were prepared as previously reported 65 . Briefly, the solution containing 20 mM sodium oleate and 2.4 mM bovine serum albumin-fatty acid free (BSA FFA) was heated to 55°C. The obtained complex was diluted in a pre-warmed culture medium at indicated concentration before addition to the cells. The treated cells were maintained in culture for 24 h and collected either for immunoblot or immunofluorescence experiments.
Cells treatment with YM201636: cells were incubated for 1 h at 37°C in the pre-warmed culture medium containing 160 mM of YM201636 and collected for analyses.
Cells treatment with PIs: PIs as a lyophilized powder were solubilized at 500 mM in water by vortex. The solution was then diluted in Dulbecco's phosphate-buffered saline (DPBS) to the concentration indicated in the legends of the experiment. Cells were washed with DPBS and the PIs solution was added to the cell for indicated time.
JFH-1/HCVcc production and inoculation. Huh7.5 were infected with cell culture-grown HCV JFH-1 at a multiplicity of infection (MOI) of 0.5 TCID 50 cell À 1 (50% tissue culture infective doses). After incubation at 37°C for 16 h, the inoculum was removed then cells were washed three times with DMEM and were further maintained in culture for the indicated time in the legend and text.
Reverse transcription and real-time PCR analysis. Total RNA was isolated using RNAble, solution (Eurobio) and HCV RNA copy number was quantified using the SuperScript III Platinium One-Step quantitative RT-PCR system (Invitrogen) and a Ligth Cycler Fast Start DNA MasterPlus SYBR Green I mix (Roche). PCR was carried on a Light Cycler 480 Real-Time PCR System (Roche). The sequence of the Primer used in real-time RT-PCR analysis was presented in Supplementary Tables 1 and 6-FAM-CCTTGTGGTACTGCCTGA-MGB (Applied Biosystems, Foster City, CA, USA) was used as internal probe. To analyse the septin 9 transcript variants, total RNA was isolated using RNeasy Mini Kit 50 (Cat# 74104 QIAGEN) and applied to reverse transcription using RevertAid First Strand cDNA Synthesis Kit (Cat#K1622 Fermentas). The cDNA was analysed by qPCR using QuantiTect SYBR Green PCR Kit (Cat#204143 QIAGEN) and a 7500 Fast Real-Time PCR System (Applied Biosystems). Reaction parameters were 30 min at 50°C, 15 min at 95°C, followed by 45 cycles of 15 s at 94°C, 30 s at 55°C and 30 s at 72°C. The triplicate mean values were calculated using GAPDH gene transcription as reference for normalization. Used qRT-PCR primers sequences are presented in the Supplementary Table 1. siRNA, plasmids and transfection reagents. siRNAs from septin 9 were a Stealth RNAi siRNA (si1: Cat#SEPT9HSS173895, si3: Cat# SEPT9HSS173897) from Invitrogen and non-targeting siRNA (Cat#sc-37007) was from Santa Cruz. The sequences targeting septin 9 (si1 and si3) are presented in the Supplementary Table 2.
cDNA septin 9_i1 vector was used as a template to generate pcDNA V5 septin 9_del1 using the QuikChange II XL Site-Direced Mutagenesis Kit (Cat#200521) from agilent following the manufacturer's recommendations. Therefore the septin 9_i1 V5/His tag and septin 9_del1 V5/His sequences have been amplified by PCR from their pcDNA3.1/V5-His-TOPO vector. The resulting DNA fragments have been inserted in a pET21d vector using In-Fusion HD Cloning System Cat#639648 (Clontech) pET21d in-frame with the C-terminal 6-histidine coding sequence of the vector and according to manufacturer recommendations to obtain pET21d septin 9_i1 V5/His and PET21d septin 9_del1 V5/His. All the primers sequences were presented in the Supplementary Table 1.
The transfection of cDNA and siRNA were performed using X-tremeGENE 9 DNA Transfection Reagent (Roche Diagnostics) and Lipofectamine RNAiMAX reagent (Invitrogen) respectively following the manufacturer's protocol. The transfection was performed for 48 h unless indicated otherwise in the text.
Protein production and purification. E. coli BL21(DE3) Rosetta cells (#70953-3 Novagen) were transformed with pET21d septin 9_i1 V5/His and PET21d septin 9_del1 V5/His plasmids and all culture media contained 100 mg ml À 1 ampicillin, 34 mg ml À 1 chloramphenicol. Cells from a single colony were used to seed an overnight 20 ml preculture of LB medium. One litre LB medium cultures from these precultures were grown at 37°C with shaking to an optical density (OD 600 nm ) of 0.9. The temperature was brought down to 28°C and protein expression was then induced with 1 mM Isopropyl b-D-1-thiogalactopyranoside (IPTG) from Sigma (Cat# I6758) for 4 h. The cells were harvested by centrifugation and the pellet was stored at À 80°C until used. The cell pellet was suspended in 15 ml of 50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, 10% glycerol, 20 mM imidazole, 0.1% Triton X-100, and one protease inhibitor cocktail EDTA free tablet (Roche). Cell lysis was performed by sonication on ice and the cell lysate was clarified by centrifugation for 30 min at 4°C to 40,000g. After filtering the supernatant the protein was isolated on a 1 ml metal affinity column (HisTrap HP, GE Healthcare) pre equilibrated in 50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, 10% glycerol, 20 mM imidazole and eluted with 50 mM sodium phosphate pH 7.4, 300 mM sodium chloride, 300 mM imidazole. The fractions containing the protein of interest were further purified by cation exchange chromatography. The pooled fractions were diluted with a 1:5 ratio to a final solution of 30 mM tris pH8, 100 mM NaCl, 1 mM EDTA and then incubated with 1 ml of a strong cation exchange resin (Macro-Prep 25S, BioRad). The resin was packed in a column and eluted with a 100-600 mM NaCl linear gradient. Fractions containing the majority of protein were obtained B350 mM NaCl. Septins were then flash-frozen and stored at À 80°C. GST-2XPHD preparation. The GST-2XPHD probe was produced in BL21 RIL þ bacteria strain (Stratagene) and purified with glutathione Sepharose 4B (Amersham). After dialysis in PBS overnight and concentration with Vivaspin 500, MCO 10000, the recombinant protein was biotinylated using the BiotinTag Micro biotinylation kit (Sigma-aldrich) according to the manufacturer.
Immunofluorescence. Cells were grown on coverslips, fixed with paraformaldehyde 4% for 20 min and permeabilized for 20 min at 37°C using the permeabilizing buffer (PFS): DPBS containing saponin (Cat#10294440 Fisher scientific) 0.025% m.v À 1 , gelatin from cold water fish skin (Cat#G7041 Sigma 0.7% m.v À 1 ). Then cells were incubated with primary antibody for 2 h washed three times for 5 min with PFS and incubated with the appropriate secondary antibodies or with the dye for 90 min. The coverslips were mounted using Prolong Gold (Cat#P36934 Invitrogen). For PtdIns5P staining the cells on coverslips were fixed and permeabilized as described above and then incubated with recombinant Bio-GST-2XPHDING2 (prepared as describe above) in 1% BSA-PBS for 1 h. Following incubation, cells were washed and incubated for another hour with streptavidin Alexa-488 (1/300) (Invitrogen) and Hoechst for nuclei staining, mounted and analysed for immunofluorescence.
Images acquisition and analysis. Images acquired with a Leica TCS SP5 AOBS tandem confocal microscope were analysed by Icy bioimage analysis software for 3D reconstruction. For colocalization analysis, images were treated by ImageJ software, the plugin 'Intensity Correlation Analysis' was used to generate the Pearson's correlation coefficient (Rr) which range from À 1 (perfect exclusion) to þ 1 (perfect correlation).
To determine LD distribution in cells, the plugin 'Radial profil' was used. For analysis, a circle was defined at the periphery of each cell and the plugin produces a profile plot of normalized integrated intensities around concentric circles as a function of distance from a point in the image, considered here as the centre of the cell. The concentric circles were assembled in three circle bands, the first corresponding to the area of the nuclei and the rest corresponding to the cytoplasm was divided in two equal bands (the band near the nuclei is considered as the 'perinuclear' and the other the 'periphery'). The intensity in each band was calculated from the total integrated intensities around concentric circles present in the band.
To calculate the size and number of LDs, images obtained by confocal microscopy were processed by background subtraction and standardized thresholding (default), cell ROI was made by free hand selection tool, then LDs size and number was obtained by ImageJ analyse particles function (particle area less than 0.01 mm 2 were excluded).
Immunoblot. Cells were washed with ice-cold DPBS and lysed on ice using the following buffer: 20 mM Tris, HCl, 100 mM NaCl, 1% Triton X-100 at PH 7.4 containing protease inhibitors (cOmplete ULTRA Cat#05892970001 Roche). The proteins were separated on SDS-PAGE and electrotransferred onto nitrocellulose membrane. After transfer, the membrane was saturated in DPBS containing 0.1% Tween 20 and 5% milk. Primary antibodies were added overnight at 4°C or for 2 h at room temperature depending on the antibody. The membranes were washed with DPBS and incubated for 1 h at room temperature with appropriate secondary antibody coupled with peroxidase. ECL plus kit (Cat#32132 Thermo Scientific) was used for protein detection. Chemiluminescent signal was detected by G:BOX Chemi Fluorescent & Chemiluminescent Imaging System from SYNGENE. Blot quantification was done using ImageJ software.
Blot Scans are presented in the figures and the uncropped scan are supplied in Supplementary Fig. 15.
PIP strip overlay assay. Membranes PIP Strips (Echelon Biosciences) were blocked with 3% BSA FFA dissolved in phosphate-buffered saline (PBS) containing 0.1% Tween 20 (3% BSA FFA PBS-T) at room temperature for 60 min, then incubated overnight at 4°C with the same buffer containing the purified protein of interest at a concentration of 0.5 mg ml À 1 or the purified corresponding tag (V5 or GST) at equivalent molar concentration used as controls. The membranes were washed and bounded proteins were detected with suitable antibody.
Neutral lipids analysis. This analysis was performed using the facility service of Metatoul in Toulouse (http://www.metatoul.fr). One million of cells were used for lipid extraction according to Bligh and Dyer 68 using dichloromethane/methanol/ water (2.5:2.5:2.1, v/v/v) mix, in the presence of the internal standards: 4 mg of stigmasterol, 4 mg of glycerid dimyristoleate, 4 mg of cholesteryl heptadecanoate, 8 mg of glyceryl trinonadecanoate. Organic phases were collected and evaporated to dryness and neutral lipids were separated through a SPE cartridge. Briefly SiOH cartridge (200 mg, Macherey Nagel) was equilibrated with 2 ml of dichloromethane and the lipid extract was put down in 20 ml of a solution of dichloromethane containing 10% omethanol and loaded. Neutral lipids were eluted with 2 ml of the same mixture. The final extract were concentrated and dissolved in 20 ml of ethyl acetate. One microlitre of the lipid extract was analysed by gas-liquid chromatography on a FOCUS Thermo Electron system using an Zebron-1 Phenomenex fused silica capillary columns (5 m Â 0,32 mm i.d., 0.50 mm film thickness) 69 . The temperature of the Oven was programmed from 200 to 350°C at a rate of 5°C per min and the carrier gas was hydrogen (0.5 bar). The injector and the detector were at 315 and 345°C, respectively.
Membrane flotation assay. The assay was performed as recently reported 37 . Huh7 cells treated with 100 mM sodium oleate for 24 h were trypsinized, washed and counted. Cells (15 Â 10 6 ) total for each sample were resuspended in 1.75 ml PBS-containing 0.25 M sucrose (PBS/sucrose) supplemented with protease inhibitor cocktail (Sigma). The cells were then lysed with 200 passages in a tight fitting dounce homogenizer to ensure B90% lysis. The cell lysate was then spun at 2,500g for 10 min at 4°C to pellet cellular debris and nuclei. Equal amounts of protein (2.5 mg) for each cell type was adjusted in volume to 1 ml with PBS/sucrose and mixed with 1 ml of 60% iodixanol (Sigma) resulting in a 30% iodixanol concentration. A discontinuous iodixanol gradient (10%, 20%) was layered on top of the lysate/iodixanol mixture, and the gradient was spun at 200,000g for 16 h at 4°C in a MLS 50 Rotor. A total of 25 fractions were collected from top to bottom. The fractions were mixed 1:1 with Laemmli buffer for immunoblot.
Transcriptome analysis. Transcriptome data set GSE14323 (ref. 31) was downloaded from NCBI website (http://www.ncbi.nlm.nih.gov/geo/query/ acc.cgi?acc=GSE14323): normalized matrix by users (RMA method in the R affy package) was employed to realize analysis. This data set is composed of 19 control samples of normal liver and 49 samples of cirrhosis. R software version 3.2.2 was employed to perform, principal component analysis, boxplots, heatmap, two-sided Student t-test, also algorithm of random forest (RF). The learning machine application RF was performed by initializing the function tuneRF to determine the mtry parameter with the optimal error of bag. The algorithm was implemented with a learning based on 500 classification trees with the optimal mtry parameter previously determined 70 .
Data availability. The data that support the findings of this study are available from the corresponding author upon request. Transcriptome data set GSE14323 (ref. 31) was downloaded from NCBI website (http://www.ncbi.nlm.nih.gov/geo/ query/acc.cgi?acc=GSE14323).