Lipid bilayer stress-activated IRE-1 modulates autophagy during endoplasmic reticulum stress

Lipid content within cells is tightly regulated to maintain cellular homeostasis, and is crucial in many physiological processes including energy storage, signalling and membrane formation. The disruption in lipid homeostasis has been strongly associated with obesity and non-alcoholic fatty liver disease (NAFLD) (Puri et al., 2007; Tiniakos et al., 2010; Arendt et al., 2013; Doycheva et al., 2017), and is often characterised by excessive accumulation of lipids in liver, pancreatic and adipose tissues. Consequently, the inability of cells to alleviate such lipotoxic conditions results in a dysfunctional stress response and the induction of apoptosis, which, ultimately, leads to a disease state (Hotamisligil and Erbay, 2008; Rinella and Sanyal, 2015). Altered hepatic phospholipids levels have been observed in murine models of non-alcoholic hepatosteatosis (NASH), a more aggressive form of NAFLD, which can develop into hepatic fibrosis and hepatocellular carcinoma (HCC) (Li et al., 2006; Fu et al., 2011). Furthermore, the ratio of phosphatidylcholine (PC) to phosphatidylethanolamine (PE), the two major phospholipid species, has been associated with the survival and prognosis of liver function after partial hepatectomy in mice (Ling et al., 2012). As both PC and PE are the main lipids within the endoplasmic reticulum (ER) membrane, simultaneously blocking the two PC biosynthesis pathways, through ablation of the PE N-methyltransferase (Pemt) gene and the lack of dietary choline, leads to hepatic ER stress in mice, which correlates with steatosis progression (Li et al., 2005) (Fig. 1A). These observations link lipid bilayer stress (LBS), ER stress and metabolic disease progression.

The ER is the hub of protein folding for proteins targeted to the secretory pathway (Schroder and Kaufman, 2005; Braakman and Bulleid, 2011). In addition to its role in protein homeostasis, the ER is the site of lipid metabolism and provides majority of membrane lipids to the cell (Lagace and Ridgway, 2013; Wu et al., 2014). Long-term disruption of lipid homeostasis triggers chronic ER stress, which is associated with metabolic diseases (Fu et al., 2011; Han and Kaufman, 2016). To counter ER stress and maintain ER functionality, eukaryotes have evolved transcriptional and translational regulatory pathways collectively termed the unfolded protein response (UPR) (Cox and Walter, 1996; Schroder and Kaufman, 2005; Shen et al., 2004). The highly conserved UPR programme exerts homeostatic control by sensing accumulation of misfolded proteins or LBS at the ER (Walter and Ron, 2011; Volmer et al., 2013). As part of the programme, a subset of genes is upregulated to remodel various cellular pathways to ease stress and maintain viability, and the failure to reach ER homeostasis may result in cell death through apoptosis (Oyadomari et al., 2002; Wu et al., 2014; Mota et al., 2016).

In metazoa, the UPR consists of three conserved ER stress transmembrane transducers namely ATF6, IRE1 (also known as ERN1 in mammals) and PERK (also known as EIF2AK3 in mammals). Upon ER stress, each sensor activates their cognate downstream effectors resulting in general translational shutdown and the upregulation of target genes to restore cellular homeostasis (Walter and Ron, 2011; Wu et al., 2014). In the event of an acute response to excessive ER stress, the UPR programme leads to cell death through the IRE1 and PERK branches (Harding et al., 2003; Novoa et al., 2001). On the other hand, chronic ER stress triggers an adaptive response of the UPR programme that leads to differential expression of pro-survival signals and stabilisation of the UPR machinery, thereby bypassing apoptosis induction (Rutkowski et al., 2006; Rubio et al., 2011; Kim et al., 2017).

Previously, we have demonstrated that changes in membrane lipid composition through the ablation of the de novo PC biosynthesis gene OPI3 (the PEMT orthologue) activate the essential intervention of the UPR to remodel the protein homeostasis network in budding yeast (Thibault et al., 2012; Ng et al., 2017 preprint). The UPR is directly activated from ER membrane LBS independently from the accumulation of misfolded proteins (Promlek et al., 2011; Volmer et al., 2013; Ho et al., 2018; Tam et al., 2018). Similar to the NAFLD mouse model lacking the Pemt gene (Li et al., 2006, 2005) and our yeast model deleted for the OPI3 gene (Thibault et al., 2012; Ng et al., 2017 preprint), Caenorhabditis elegans models have been developed to block PC synthesis via loss-of-function (herein denoted lof) mutations in the PEMT orthologues pmt-1 and pmt-2 or the upstream precursor S-adenosylmethionine synthetase sams-1 (Brendza et al., 2007; Li et al., 2011; Walker et al., 2011; Ding et al., 2015) (Fig. 1A).

Excessive protein accumulation during proteotoxicity-induced ER stress has been shown to activate autophagy because the accumulation of misfolded proteins exceeds the capacity of the ER-associated degradation for their clearance (Kouroku et al., 2007; Senft and Ronai, 2015), but little is known on the role of autophagy during the LBS-induced UPR. In this study, we characterised the role of the UPR during chronic LBS-induced ER stress. We proposed that LBS may elicit a chronic ER stress response that is distinct from that resulting from acute ER stress. In C. elegans, we silenced pmt-2 to attenuate PC synthesis in the UPR mutant animals atf-6(lof), ire-1(lof) and pek-1(lof). As expected, genetic ablation of pmt-2 resulted in the perturbation of lipid homeostasis with significantly decreased PC levels, which then correlated with UPR activation. Although conventionally seen as a linear response to ER stress, our findings demonstrate a strikingly different outcome of the UPR programme when activated by LBS (hereafter denoted UPR^LBS) as opposed to proteotoxic stress (hereafter denoted UPR^PT). More importantly, we reveal that autophagy is modulated by the IRE-1–XBP-1 axis during the LBS-induced UPR, suggesting the importance of autophagy to maintain cellular homeostasis during LBS.

PMT-1 and PMT-2 are both required for the synthesis of PC from PE in C. elegans (Palavalli et al., 2006; Brendza et al., 2007; Li et al., 2011) (Fig. 1A). In the absence of dietary choline, both genes are essential for the development of C. elegans, and silencing either gene from stage-one (L1) larvae onwards leads to sterility (Brendza et al., 2007). PC cannot be obtained from the standard laboratory worm diet as the conventionally used E. coli strains OP50 and HT115(DE3) lack PC (Morein et al., 1996; Oursel et al., 2007). Thus, PC levels in worms can be altered by genetically manipulating pmt-1 or pmt-2 to induce LBS thereby activating the UPR (Fig. S1C). To better understand the role of the UPR during LBS, we subjected synchronised L1 worms to pmt-2 RNA interference (RNAi) for 48 h. Two-day RNAi feeding was sufficient to decrease pmt-2 mRNA in wild-type (WT) animals to close to the background signal of pmt-2(lof) (Fig. 1B). As previously reported, pmt-2(RNAi) animals showed a developmental defect characterised by reduced body size, which could be rescued by choline supplementation (Fig. S1A,B) (Palavalli et al., 2006). PC is synthesised from choline through the Kennedy pathway (Fig. 1A). Supplementing pmt-2(RNAi) animals with 30 mM choline was sufficient to prevent UPR activation while the growth defect was alleviated with 60 mM choline (Fig. S1A–C). To further characterise pmt-2(RNAi) worms, PC was separated from the total lipid extract of worms by thin layer chromatography (TLC). This was followed by a transesterification reaction to derive fatty acid methyl esters (FAMEs) specifically from PC, which were further quantified via gas chromatography with a flame ionisation detector (GC-FID). We found that the PC level in pmt-2(RNAi) worms was markedly reduced to 36% of that in vector control worms (Fig. 1C). Attenuating pmt-2 expression was not sufficient to fully eliminate PC in WT animals due to the large phospholipid reserve in stage L1 worms in addition to the long half-lives of phospholipids (Dowd et al., 2001). As LBS can lead to ER stress, we measured the transcriptional levels of the UPR-induced ER-resident molecular chaperone Hsp70 family (Urano et al., 2002). The mRNA levels for both of the C. elegans human Hsp70 family orthologues, hsp-3 and hsp-4, were upregulated transcriptionally in pmt-2(RNAi) worms compared to WT (Fig. 1D). The mRNA level of hsp-3 in pmt-2(RNAi) worms was similar to that of WT worms incubated with the strong UPR inducer tunicamycin (Tm) for 4 h. Tm inhibits protein N-glycosylation leading to a severe accumulation of unfolded proteins in the ER (Ericson et al., 1977). In contrast, the hsp-4 mRNA level was remarkably higher in Tm-treated animals compared to those with LBS. This result suggests that hsp-4 expression might be modulated differently by the UPR programme depending on the nature of the stress. As the UPR is activated by low PC, C. elegans subjected to pmt-2 RNAi mimics NAFLD because altered PC:PE ratios and UPR activation are interconnected (Ozcan et al., 2004; Fu et al., 2011; Thibault et al., 2012; Ng et al., 2017 preprint).

To investigate the regulatory role of the three ER stress transducers during LBS (Shen et al., 2005), atf-6(lof), ire-1(lof) and pek-1(lof) mutant worms were subjected to pmt-2 RNAi as described above (Fig. S2A). The simultaneous ablation of two or three UPR branches is not possible as any combination is lethal (Shen et al., 2005). Treatment with pmt-2 RNAi resulted in a reduction of PC across the three UPR mutants that was comparable to that seen in WT worms (Fig. 2A; Fig. S2B). The UPR transducers regulate lipid metabolism where XBP-1 downstream of IRE-1 is responsible for phospholipid synthesis (Sriburi et al., 2004). Hence, it is unsurprising that ire-1(lof) worms have decreased PC compared to WT in the untreated condition (Fig. 2A). To quantify the relative levels of the fatty acids (FAs) derived from PC and its precursor PE during LBS, both phospholipids were separated from total lipid extracts by TLC and quantified by GC-FID as described above. As expected, PC depletion in the UPR mutants and WT worms caused a significant reduction of all FAs derived from PC except for margaric acid (C17:0), α-linolenic acid (C18:3n3) and eicosadienoic acid (C20:2), indicating a general disturbance in the PC metabolic pathways during LBS (Fig. 2B). However, levels of FAs derived from PE were largely unmodified across the strains as pmt-2 RNAi leads to the accumulation of the downstream phospholipid intermediate monomethyl-phosphatidylethanolamine (MMPE) (Fig. 1A). As a reduction in PC level leads to lipid droplet (LD) accumulation in eukaryotes (Li et al., 2011; Walker et al., 2011; Thibault et al., 2012; Hörl et al., 2011), Sudan Black staining of fixed worms treated with pmt-2 RNAi was carried out to visualise lipid droplets by Nomarski microscopy (Fig. 2C; Fig. S3A). To compare LD expansion in pmt-2 RNAi-treated worms, the diameter of LDs was measured and classified into small (0.8–3 µm), medium (3.1–6 µm) and large (>6 µm) groups. We observed accumulation of large LDs but fewer total LDs in WT, atf-6(lof), ire-1(lof), and pek-1(lof) worms treated with pmt-2 RNAi, corresponding to previous study in sams-1(lof) worms, where PC is also reduced (Ding et al., 2015).

To ensure that pmt-2 RNAi treatment is sufficient to induce ER stress in the UPR mutants, we monitored the mRNA expression of two downstream target genes of IRE-1, hsp-3 and hsp-4, by quantitative RT-PCR (qPCR). As expected, the levels of both the ER-resident chaperones hsp-3 and hsp-4 levels were significantly increased upon pmt-2 RNAi and Tm treatments in WT and mutant strains with the exception of ire-1(lof) (Fig. 2D,E). Activation of the UPR from LBS was further validated in vivo by immunoblotting using the phsp-4::gfp reporter animal (Calfon et al., 2002). Consistent with the increase in hsp-4 mRNA levels, pmt-2 RNAi treatment resulted in an almost 2-fold increase in GFP in phsp-4::gfp animals, while Tm induced stronger upregulation (Fig. 2F). As expected, an increase in GFP was not detected in xbp-1;phsp-4::gfp animals treated with pmt-2 RNAi nor with Tm, as HSP-4 is specifically upregulated from the IRE-1–XBP-1 axis upon ER stress (Acosta-Alvear et al., 2007).

Several studies suggest that the essential role of the UPR in maintaining metabolic and lipid homeostasis is highly conserved across species (for reviews, see Volmer and Ron, 2015; Han and Kaufman, 2016). A novel ER stress-sensing mechanism has been proposed in which the UPR is activated by lipid bilayer stress independently of unfolded protein accumulation in the ER (Promlek et al., 2011; Volmer et al., 2013; Halbleib et al., 2017). Thus, ER stress triggered by proteotoxic stress or LBS might differentially modulate the UPR to reach cellular homeostasis. To address whether this occurs, DNA microarray analysis was performed using RNA extracted from WT, atf-6(lof), ire-1(lof) and pek-1(lof) animals treated with pmt-2 RNAi. WT worms incubated with Tm for 4 h were included in the analysis to identify genes modulated by UPR^PT and subsequently uncouple those specifically modulated by UPR^LBS. To validate the quality of microarray data, qPCR was performed on a subset of genes (Fig. S4A,C). Overall, 2603 and 1745 genes were upregulated and downregulated, respectively, in pmt-2(RNAi) animals compared to WT (Fig. 3A, Table S2, data deposited in the Gene Expression Omnibus database under code GSE99763). In addition, Tm-treated WT worms showed upregulation of 1258 genes and downregulation of 1473 others. Only 492 and 420 genes were similarly upregulated and downregulated, respectively, from both UPR^LBS and UPR^PT, thereby suggesting these groups of genes are commonly modulated from the UPR regardless of the source of ER stress.

To explore how the UPR elicits a differential stress response during LBS and proteotoxic stress, we filtered the 2111 gene candidates that were upregulated only in pmt-2(RNAi) animals and excluding genes upregulated in Tm-treated animals, respectively. Genes with unaltered expression in at least one of the UPR mutants subjected to pmt-2 RNAi are considered to be modulated by UPR^LBS. From these criteria, 1069 genes were upregulated in a UPR^LBS-dependent manner (Fig. 3B; Table S3). We identified 181, 417 and 25 genes that are specifically upregulated from the ATF-6, IRE-1 and PEK-1 branches of the UPR, respectively, while 446 genes were modulated from at least two of the three UPR branches, suggesting compensatory roles of one or more UPR transducers in the absence of the other branches. In addition, we grouped genes with at least a 1.5-fold change in expression level by hierarchical clustering (Fig. 3C; Table S4). This allowed us to visualise genes that were similarly regulated throughout the array from UPR^LBS. Manual inspection of our array data demonstrates that the upregulation of known UPR target genes are in agreement with previous reports (Fig. 3D) (Travers et al., 2000; Shen et al., 2005; Thibault et al., 2011).

To further understand the role of the UPR^LBS programme, we performed functional annotation for the 1069, 181, 417 and 25 upregulated genes identified from pmt-2(RNAi), atf-6(lof); pmt-2(RNAi), ire-1(lof); pmt-2(RNAi), and pek-1(lof); pmt-2(RNAi) animals, respectively, using the gene ontology (GO) tool DAVID (Table S5) (Huang et al., 2007). Immune regulatory genes were found to be enriched in the upregulated categories of WT and ire-1 animals (Fig. 4A,C). This is in agreement with a previous report where innate immunity was found to be modulated by sams-1, the methyl donor to pmt-1 and pmt-2 (Ding et al., 2015). Our data suggest that innate immune response is positively and negatively regulated by IRE-1 and PEK-1, respectively (Fig. 4C,D). As expected, GO terms related to ER stress were found to be enriched in the upregulated categories of WT and mutant animals when PC is depleted (Fig. 4A–D) (Ding et al., 2015). We also identified an enrichment of downregulated genes in WT related to translational initiation factors including eif-1.A, eif-3.C and eif-3.E, a characteristic effect of UPR activation (Ling et al., 2009; Long et al., 2002).

Protein tyrosine phosphatase activity is significantly regulated by ATF-6 upon LBS (GO ID 0035335, n=6, Benjamini P-value of 0.047) (Fig. 4B). This class of genes is activated in response to ER stress (Agouni et al., 2011). Transcriptional regulation (GO ID 0006355, n=49, Benjamini P-value of 3.3×10⁻⁹) is enriched among IRE-1-dependent genes and supports the evidence that they alleviate ER stress through a widespread transcriptional modification process (Fig. 4C) (Ng et al., 2000). In addition, genes involved in lipid and fatty acid processes were also enriched by IRE-1 and PEK-1 (Fig. 4C,D). Finally, protein tyrosine phosphatase activity (GO ID 0004725, n=4, Benjamini P-value of 0.039) is enriched by PEK-1, suggesting its involvement in protein modifications and the cell signalling cascade during LBS (Fig. 4D) (Bettaieb et al., 2012). Interestingly, we observed upregulation of autophagy-related processes that are IRE-1-dependent, as shown in the volcano plot [where autophagy-related genes with at least a 1.5-fold change and false discovery rate (FDR) P<0.05 were displayed on the plot; Fig. 4C,E]. Crosstalk between the UPR and autophagy in the context of protein clearance has been well documented (for a review, see Mizushima and Komatsu, 2011).

To investigate whether the autophagic process is activated during the UPR^LBS, we carried out an RNAi screen of autophagy-related genes. To decrease PC levels, WT worms were first subjected to 36 h of pmt-2 RNAi followed by 5 days of autophagy-related gene RNAi treatment (Fig. 5A). Phenotypes following RNAi treatment were classified as 0 (little difference in growth and brood size), 1 (smaller brood size, sick), and 2 (sterile, very sick) compared to vector control. The screen was carried out twice, and scores were designated for both WT animals pre-treated with vector and pmt-2 RNAi. An RNAi candidate was classified as a positive hit if the sum of the scores in our phenotype scoring was equal to or exceeded 3 (Table S6). From the 40 RNAi clones, seven were found to induce development defects upon RNAi treatment compared to vector control (Fig. S5). As positive controls, we incorporated ero-1 and pmt-2 RNAis, as either lead to developmental defects and sterility (Rual et al., 2004; Palavalli et al., 2006). The screen revealed that some autophagy-related genes are required during UPR^LBS and their absence contributes to the observed detrimental phenotypes. Identified genes include atg-7 (the orthologue of human ATG7, involved in autophagosome conjugation), atg-13 (the orthologue of human ATG13, involved in autophagosome formation), bec-1 (the orthologue of human Beclin1, involved in vesicle nucleation) and wdfy-3 (the orthologue of human WDFY3, an autophagy adaptor) (Melendez et al., 2003; Takacs-Vellai et al., 2005; Jia et al., 2007; Wang et al., 2018). Additionally, the screen identified potential autophagy-related genes that are less characterised and may prompt further investigation. These include rsks-1 (the orthologue of human RPS6KB1, a negative regulator of autophagy), sepa-1 (no human orthologue identified, involved in P-granule-associated autophagy) and trpp-8 (orthologue of human GSG1, involved in autophagic processes) (Silvestrini et al., 2018; Zhang et al., 2009; Meiling-Wesse et al., 2005).

To better understand the crosstalk between UPR^LBS and autophagy, we monitored the transcription levels of key genes involved in the regulation of autophagy. Both bec-1 and lgg-1 (an orthologue of human MAP1LC3 proteins, involved in autophagosome formation) genes were found to be significantly upregulated in WT, atf-6(lof) and pek-1(lof) animals, but not in ire-1(lof) animals, compared to the control upon pmt-2 RNAi treatment (Fig. 5B,C). These results suggest that IRE-1 modulates bec-1 and lgg-1 expression during LBS. We validated that bec-1 and lgg-1 are both upregulated by Tm as previously reported (Ogata et al., 2006). Once IRE-1 is activated, it cleaves xbp-1 mRNA through an unconventional splicing process. This is followed by the translation of the XBP-1 transcription factor, which regulates a downstream signalling cascade (Walter and Ron, 2011; Ho et al., 2018) that modulates the expression of target genes, including hsp-3, during UPR^LBS (Fig. S6A). Thus, we monitored bec-1 and lgg-1 mRNA levels in xbp-1(lof) animals treated with pmt-2 RNAi. The upregulation of bec-1 and lgg-1 was abolished in xbp-1(lof) animals suggesting that XBP-1 modulates autophagy during UPR^LBS (Fig. 5D,E). Similar increases in the mRNA level of atg-18 (an orthologue of human WIPI genes) and epg-4 (an orthologue of human EI24) were observed in pmt-2(RNAi) but not in xbp-1(lof);pmt-2(RNAi) animals (Fig. 5F,G) (Tian et al., 2010; Devkota et al., 2016). On the other hand, no significant transcriptional variations in atg-4.1 (an orthologue of the human ATG4 genes) and atg-9 (orthologue of human ATG9) were observed, while atg-16.2 (an orthologue of human ATG16L1) was upregulated upon LBS in a XBP-1-independent manner (Fig. S6B–D). Taken together, these findings suggest that the UPR programme transcriptionally regulates a subset of autophagy genes during LBS.

To gain insight into autophagy flux, we used the autophagy reporter strain plgg-1::gfp::lgg-1 crossed to the RNAi-sensitive eri-1 worms for simultaneous treatment with pmt-2 and ire-1 RNAis (Melendez et al., 2003). We detected an increased number of GFP::LGG-1 puncta in intestinal tissues of pmt-2(RNAi) animals, but not in ire-1 and pmt-2 double RNAi-treated animals (Fig. 5H,I, Fig. S6E–G). These puncta are indicative of autophagosomes, because spermidine treatment increased their number (Jia et al., 2009). Next, we separated GFP::LGG-1 from its PE-conjugated form, GFP::LGG-1–PE, an autophagosomal marker (Kang et al., 2007). When entering the lysosome, GFP::LGG-1–PE is hydrolysed, releasing stable and free GFP (Hosokawa et al., 2006). Significant increases in GFP::LGG-1–PE as well as free GFP were observed in pmt-2(RNAi) but not in ire-1(RNAi); pmt-2(RNAi) animals (Fig. 5J,K). Taken together, these results indicate that autophagy is modulated by the IRE-1–XBP-1 axis upon UPR^LBS.

Lipid perturbation refers to excessive accumulation of lipids in tissues, including liver, pancreas and adipose tissue (Hotamisligil and Erbay, 2008; Rinella and Sanyal, 2015). Dysfunctional UPR and apoptotic pathways resulting from this lipotoxicity ultimately lead to disease outcomes. To better understand the role of the UPR during LBS and the consequence of a compromised UPR programme, several studies have been conducted, focusing on their interconnection. As it is required for normal FA synthesis, as well as the regulation of very low-density lipoprotein (VLDL) assembly and its secretion, Xbp1 ablation leads to hypolipidemia in mice owing to an abnormal decrease in plasma levels of TG and cholesterol (Lee et al., 2008; So et al., 2012; Wang et al., 2012). High dietary carbohydrate is sufficient to increase FA and cholesterol synthesis through XBP1 (Lee et al., 2008). Consequently, XBP1 is required to channel excess carbohydrate into lipids, as its absence leads to insulin resistance in obese mice (Ozcan et al., 2004). XBP1 is also required to modulate phospholipid synthesis in order to expand the ER membrane network during proteotoxic stress, a process that is proposed to accommodate the increased load of misfolded proteins (Sriburi et al., 2004). The UPR modulates lipid metabolism-related genes, and the absence of its sole regulator in yeast, Ire1, confers auxotrophy of inositol, a building block of phospholipids (Cox et al., 1993). We have previously shown that Ire1 is essential for cell survival during LBS, thereby highlighting the important role of the UPR to overcome lipotoxicity in yeast (Thibault et al., 2012). The UPR sensor PERK also plays a role in pathogenesis from lipotoxicity. Lipotoxicity-induced CHOP (also known as DDIT3), the downstream target gene of PERK, promotes hepatic inflammation by activating the NK-κB pathway, thus promoting NASH and type 2 diabetes (Cunha et al., 2008; Willy et al., 2015). The ablation of ATF6, the third UPR sensor, induces NASH owing to dysregulated lipid biosynthesis in mice upon Tm treatment (Yamamoto et al., 2010). Thus, the three branches of the UPR are intimately linked to lipid homeostasis but their respective roles during lipid bilayer stress in comparison to proteotoxic stress remain elusive. Here, we took a systematic global approach to determine genes that are regulated by the UPR- specifically induced by LBS.

To introduce LBS in C. elegans, we opted to genetically attenuate pmt-2, which is required for de novo PC biosynthesis (Fig. 1A). A similar approach has been used by other groups to mimic the physiological conditions associated with NAFLD in C. elegans (Walker et al., 2011; Ding et al., 2015; Smulan et al., 2016). Because both are required for de novo PC biosynthesis, pmt-2 and sams-1 depletion leads to enlarged lipid droplets in worms (Li et al., 2004). Generally, perturbing PC levels affects the abundance and size of lipid droplets, serving as a compensatory response to LBS that results in the channelling of excess neutral lipids, triacylglycerol and sterol into lipid droplets (LDs) (Guo et al., 2008; Li et al., 2011; Walker et al., 2011).

Decreased hepatic PC in mice (Walkey et al., 1998; Ozcan et al., 2004; Li et al., 2006; Fu et al., 2011) and dietary deficiency of choline in humans are both associated with hepatic steatosis (Buchman et al., 1995; Gao et al., 2016). Initially, we subjected young adult worms to pmt-2 RNAi for 2 days. However, no significant decrease in PC level was observed (data not shown). This could be due to the slow turnover of phospholipids in C. elegans. The absence of cell division in adult worms might not require the rapid synthesis of new membrane lipids, thus genetic ablation of pmt-2 may have little or no effect on PC levels (Kipreos, 2005). Thus, L1 stage worms treated with pmt-2 RNAi were utilised as the UPR^LBS model. These latter conditions were sufficient to drastically induce LBS, lipid storage and to strongly activate the UPR, all hallmarks of NAFLD (Figs 1 and 2, Figs S1–S3). By using this approach, we interrogated the role of each UPR branch during LBS-induced ER stress.

We employed C. elegans for its well-conserved UPR pathways and relative simplicity for genetic analysis. We examined the individual effects of atf-6, ire-1 and pek-1 deficiency in vivo. Interestingly, ER stress induced by unfolded protein accumulation and LBS were found to be distinct from each other (Hou et al., 2014; Lajoie et al., 2012). A global transcriptomic analysis of UPR mutants subjected to LBS in comparison to what was seen with proteotoxic-induced ER stress in WT animals allowed us to identify genes that were specifically regulated by UPR^LBS but not UPR^PT (Fig. 3). To our knowledge, this is the first report identifying specific UPR-regulated genes induced by LBS but not proteotoxic stress. Our data show that a number of genes regulated by the UPR transducers are specific to LBS, while a smaller number of genes are commonly modulated under proteotoxic and lipid stress. As expected, LBS-induced ER stress leads to altered gene regulation and protein modification processes through ATF-6, IRE-1 and PEK-1 (Figs 3 and 4). IRE-1 is the most-conserved UPR transducer from yeast to mammals, and it regulates the largest number of genes among the three UPR transducers (Fig. 3B).

Our autophagy screening revealed that autophagy is essential during LBS, suggesting that it has an important role in regulating lipid metabolism. The change in cellular lipid landscape is at least partially mediated by autophagy through the IRE-1–XBP-1 axis (Figs 4 and 5). Generally considered a cytoprotective response, autophagy can be modulated by ER stress. PERK has been reported to modulate autophagy by phosphorylating eIF2α, resulting in a general translational inhibition (Matsumoto et al., 2013; Avivar-Valderas et al., 2011; Fujita et al., 2007; Kouroku et al., 2007). In parallel, PERK has also been reported to regulate autophagy through the transcription factor ATF4 (an orthologue of ATF-5) (Carra et al., 2009; Dever, 2002; Talloczy et al., 2002). Likewise, IRE1 modulates autophagy independently of XBP1, by activating the Jun N-terminal kinase (JNK) pathway (Younce and Kolattukudy, 2012; Vidal et al., 2012; Pattingre et al., 2009; Wei et al., 2008a,b; Ogata et al., 2006). Autophagy has additionally been reported to be activated (Younce and Kolattukudy, 2012) or inhibited (Vidal et al., 2012) by the IRE1–XBP1 axis (Adolph et al., 2013; Zhao et al., 2013; Hetz et al., 2009).

Muting the UPR during LBS revealed that the IRE-1–XBP-1 axis specifically modulates autophagy (Fig. 6). Our findings also demonstrate that a subset of autophagy genes is essential for organismal health during LBS (Fig. 5A; Fig. S5, Table S6). Considering the important role of lipid homeostasis and how its impairment contributes to the pathology of metabolic diseases, our data revealed the important role of a functional UPR programme to regulate autophagy and, consequently, maintain cellular homeostasis. As increasing evidence suggests that reduced levels of autophagy contribute to a myriad of pathological outcomes (Mizushima et al., 2008; Singh et al., 2009), the autophagy process might be a useful target for therapeutic strategies in metabolic diseases.

All strains were grown at 20°C using standard C. elegans methods as previously described (Brenner, 1974; Stiernagle, 2006). Nematode growth medium (NGM) agar plates were seeded with E. coli strain OP50 for normal growth and with HT115 bacteria for RNAi feeding. RNAi feeding was performed as previously described (Timmons and Fire, 1998), and the RNAi library was obtained from the Fire laboratory (purchased through Dharmacon, Lafayette, CO) (Fire et al., 1998). The plasmids were sequenced to confirm their identity. Wild-type N2 Bristol, atf-6(ok551), ire-1(ok799), pek-1(ok275), pmt-2(vc1952), phsp4::gfp(sj4005), xbp-1(lof);phsp-4::gfp(sj17), eri-1(mg366) and plgg-1::gfp::lgg-1(adIs2122) strains were obtained from Caenorhabditis Genetic Center (CGC). The eri-1;plgg-1::gfp::lgg-1 reporter strain was obtained by crossing eri-1 worms to plgg-1::gfp::lgg-1 as previously described (Fay, 2006).

RNAi was carried out as previously described (Timmons and Fire, 1998). Briefly, HT115 bacteria harbouring pL4440 plasmids were grown in LB medium containing 100 µg/ml ampicillin at 37°C until log phase [optical density at 600 nm (OD₆₀₀) of 0.6] and seeded onto NGM agar plates containing 50 µg/ml carbenicillin and 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Gravid adult worms were treated with hypochlorite and eggs were hatched overnight in M9 medium at 20°C to obtain L1 synchronised worms. Hatched L1 larvae were transferred to RNAi agar plates and grown until L4 larval to young adult stages. L4/young adult worms were harvested and incubated with 25 µg/ml Tm in M9 medium for 4 h at 20°C followed by washes with M9 when indicated. To measure body length, worms were transferred to 6 cm NGM plates without bacteria. Bright-field images were acquired with a dissecting microscope (Nikon SMZ1500) fitted with a JVC digital camera at 100× magnification. Length measurements were performed in Fiji software with the WormSizer plugin (Moore et al., 2013). For double-RNAi feeding, worms were fed with equal amount of HT115 bacteria harbouring pL4440 plasmids and the efficacy of gene silencing was assessed by qPCR.

L4/young adult worms (∼10,000) were collected, resuspended in water and lysed with a motorised pestle homogeniser. Total RNA was isolated using TRIzol reagent (Thermo Fisher, Waltham, MA) and subsequently purified using RNeasy Mini (Qiagen, Venlo, Netherlands) columns following the manufacturer's protocols. DNase treatment in columns was carried out with RNase-free DNase (Qiagen, Venlo, Netherlands) following the manufacturer's protocol. cDNA was synthesised from 2 μg of total RNA using RevertAid reverse transcriptase (Thermo Fisher, Waltham, MA) following manufacturer's protocol. SYBR Green qPCR experiments were performed following the manufacturer's protocol using a QuantStudio 6 Flex Real-time PCR system (Applied Biosystems, Waltham, MA). cDNA (30 ng) and 50 nM of paired-primer mix were used for each reaction. Relative mRNA was determined with the comparative Ct method (ΔΔCt) normalised to housekeeping gene act-1. Oligonucleotide primers used are listed in Table S1.

L4/young adult worms (∼10,000) were harvested and washed thoroughly with M9 buffer, lysed with 1 mm silica beads by bead beating and subsequently lyophilised overnight (Virtis). All subsequent steps were carried out at 4°C. Total lipids were extracted from dried samples with chloroform:methanol (2:1) and concentrated. Total lipid extracts and POPE (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine; 16:0-18:1n9 PE; Avanti Polar Lipids, Alabaster, AL)/DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine; 18:1n9 PC; Avanti Polar Lipids, Alabaster, AL) standard mix were spotted on HPTLC Silica gel 60 plates using Linomat 5 (CAMAG) and separated with chloroform:methanol:acetic acid:acetone:water (35:25:4:14:2). Phospholipids were visualised under long-wave ultraviolet light (λ=340 nm) by spraying 0.05 mg/ml of Primuline dye in acetone:water (80:20) onto the dried plates. Spots corresponding to PE and PC were scraped off the silica plates and transferred into 2 ml glass tubes. A 100 μl volume of 1 mM C15:0 (pentadecanoic acid) was added to the tubes containing silica-bound phospholipids as an internal standard. The phospholipids were hydrolysed and esterified to fatty acid methyl esters (FAME) with 300 μl of 1.25 M HCl-methanol for 1 h at 80°C. FAMEs were extracted three times with 1 ml of hexane. Combined extracts were dried under nitrogen, and resuspended in 100 μl hexane. FAMEs were separated by gas chromatography with a flame ionisation detector (GC-FID) (GC-2014, Shimadzu, Kyoto, Japan) using an ULBON HR-SS-10 50 m×0.25 mm column (Shinwa, Tokyo, Japan). Supelco 37 component FAME mix was used to identify corresponding FAs (Sigma-Aldrich, St Louis, MO). Data were normalised using the internal standard C15:0 and worm dry mass.

Following RNAi feeding, lipid droplets of L4 worms were stained with Sudan Black B (Sigma-Aldrich, St Louis, MO) as described previously with a few modifications (Ogg and Ruvkun, 1998). Briefly, worms were fixed in 1% paraformaldehyde in M9 buffer for 30 min at room temperature, followed by three freeze-thaw cycles using liquid nitrogen. Worms were washed once with M9 and gradually dehydrated with 25%, 50% and 70% ethanol. Subsequently, fixed worms were stained with 50% saturated Sudan Black B in 70% ethanol (filtered with a 0.22 µm membrane) for 30 min at room temperature with rocking. Stained worms were washed once with 25% ethanol for 30 min with rocking. Worms were mounted on a 2% agarose pads for imaging. Bright-field images of worms were taken with a DMi8 inverted epifluorescence microscope (Leica, Wetzlar, Germany) with 20× and 63× objective lenses. To quantify the number and size of lipid droplets, TIFF images taken at 63× magnification were converted to 8-bit grayscale images, followed by background subtraction and thresholding with Fiji imaging software. Lipid droplets were divided into three size groups based on the diameter of LDs: small (0.8–3 µm), medium (3.1-6 µm) and large (>6 µm). The percentage of LDs per size group as a proportion of the number of measured LDs per condition was presented.

Worms at L4 stage were collected in M9 buffer and treated with 1 mM spermidine (Sigma-Aldrich) for 16 h at 20°C with shaking. Worms were then washed three times with M9 buffer and harvested for downstream experiments.

Worms were collected, washed in M9 buffer and subsequently lysed in RIPA buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40, 0.1% SDS, 2 mM EDTA, and 0.5% sodium deoxycholate) with protease inhibitor cocktail (Roche, Basel, Switzerland) by bead beating three times for 30 s at 6500 rpm with the samples chilled on ice between the homogenisation steps. Samples were then centrifuged at 10,000 g for 5 min at 4°C to remove debris. Cleared lysate protein concentration was measured by using a BCA assay kit (Thermo Fisher, Waltham, MA). Either 60 µg or 80 µg of total proteins were loaded into 10% SDS-PAGE gels to detect free GFP and GFP::LGG-1, respectively; proteins were transferred to nitrocellulose membranes and stained with REVERT total protein stain (Li-COR Biosciences, Lincoln, NE) for normalisation. Membranes were blocked for 1 h with Odyssey blocking buffer TBS (Li-COR Biosciences, Lincoln, NE) at room temperature, and incubated with 1:1000 of monoclonal anti-GFP antibody overnight at 4°C (Roche, catalogue number 11814460001), washed, and incubated with 1:10,000 of IRDye 800CW anti-mouse IgG antibody (Li-COR Biosciences, Lincoln, NE, catalogue number 925-32210). Membranes were washed and scanned with an Odyssey CLx imaging system (Li-COR Biosciences, Lincoln, NE).

To quantify autophagic vesicles, eri-1;lgg-1p::gfp::lgg-1 worms were immobilised in M9 containing 0.5 M NaN₃ and mounted on 2% agarose pad and imaged using LSM Zeiss 710 scanning confocal microscope (Zeiss, Oberkochen, Germany). GFP excitation and emission wavelengths were adjusted to 493 and 517 nm, respectively, to reduce autofluorescence. Z-stacks were acquired with a 63× objective of 0.6 µm thickness. Line average scanning was set to eight times to increase the signal-to-noise ratio. Maximum intensity projections were acquired from the z-stack images with ZEN software (Zeiss, Oberkochen, Germany). The number of GFP-positive puncta were quantified in one 1000 µm² area around the anterior intestines with ZEN software.

Three independent populations of WT, atf-6(lof), ire-1(lof), pek-1(lof) and pmt-2(lof) worms were synchronised via hypochlorite treatment. L1 stage animals were treated with pmt-2 RNAi or pL4440 empty vector for 48 h. Next, total RNA from the treated worms was isolated as described above. RNA quality control for microarray analysis was carried out using a Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). RNA samples with a RIN score >9.5 were deemed suitable for microarrays. The cDNAs were then synthesised from 100 ng of total RNA, purified, fragmented and hybridised to GeneChip C. elegans Gene 1.0 ST arrays. Differentially expressed genes were identified by using Affymetrix Transcriptome Analysis Console (TAC) 3.0 software. The threshold for selecting differentially expressed genes was set at a difference of more than 1.5 fold with P-values lower than 0.05 obtained by one-way ANOVA testing. GOrilla (http://cbl-gorilla.cs.technion.ac.il/) (Eden et al., 2009), REViGO (http://revigo.irb.hr/) (Supek et al., 2011) and DAVID (https://david.ncifcrf.gov/) (Huang et al., 2007) were used for GO terms analysis. Heat maps in the figures were generated using R Studio. Venn diagrams were generated using the following generator (http://bioinfo.genotoul.fr/jvenn/example.html). For gene expression analysis, normalised and log-transformed array data were imported to Cluster 3.0 for fold cut-off and hierarchical clustering. Genes were filtered to obtain those with a significant change in gene expression (fold change >1.5 between RNAi-treated and untreated samples at P<0.05). The filtered data set was hierarchically clustered based on average linkage and the Pearson correlation method, and the output was displayed in TreeView. qPCR was performed to verify mRNA expression of selected gene targets.

RNAi screen was carried out as previously described with minor modifications (Lehner et al., 2006). Briefly, L1 larval stage animals were synchronised by hypochlorite treatment and exposed to pmt-2 RNAi on NGM agar plates for 36 h. Thereafter, worms were washed three times with M9 and five to ten worms were seeded into a 96-well plate containing RNAi clones to inhibit genes with autophagy-related functions. Control RNAi plates comprised worms exposed to pL4440 empty vector for 48 h and subsequently seeded into 96-well plates containing the same RNAi clones as above. Phenotypes of the worms were monitored over a 5-day period. Phenotypes were compared to control RNAi plates where the worms were scored for sterility and reduced body size semi-quantitatively on a scale from 0 (wild-type) to 2 (100% sterility or stunted growth) (Lehner et al., 2006).

Error bars indicate standard error of the mean (s.e.m.), calculated from at least three biological replicates, unless otherwise indicated. P-values were calculated using a one-way ANOVA with Tukey's post hoc test, unless otherwise indicated and reported as P-values with four significant digits in the figures. All statistical tests were performed using GraphPad Prism 7 software.

We are grateful to Drs Fumio Motegi, Jean-Claude Labbé and Ronen Zaidel-Bar for providing reagents and technical support to introduce C. elegans as a new model system in the laboratory. Some strains were provided by the CGC, which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440). We thank Peter Shyu Jr. for critical reading of the manuscript.