TAGs are neutral lipids composed of a glycerol backbone conjugated to three fatty acyl chains and serve as the main unit of stored energy in a range of organisms, including oleaginous bacteria, algae and mammals. In humans, most (possibly all) cell types are able to synthesize triglycerides. High levels of TAGs (hypertriglyceridaemia) are associated with obesity and metabolic syndrome, and the mobilization of TAGs from adipose tissue can contribute to cachexia, a multiorgan wasting disorder1,2,5,6,7. In cells, TAGs can be stored in dedicated organelles known as lipid droplets, which provide energy to mitochondria through direct interorganellar contact sites8. TAG synthesis in humans is carried out by DGAT enzymes, which catalyse the CoA-dependent acylation of diacylglycerol (DAG) produced by the Kennedy pathway3,5,9,10,11. DGAT1 and DGAT2 reside at the endoplasmic reticulum (ER) and are considered to be therapeutic targets for metabolic disease12,13,14,15. DGAT-dependent TAG formation is typically limited by the availability of free fatty acids3,5,16. Inspired by observations of alternative mechanisms of TAG synthesis in algae, yeast and mice, we used a haploid genetic approach in human cells to identify an unexpected route for catalysing TAG synthesis.

TMX1 restricts TAG accumulation

We first characterized TAG synthesis in haploid human HAP1 cells. Acute inhibition of DGAT1 and DGAT2 (Extended Data Fig. 1a) in HAP1 cells severely decreased baseline TAG levels, as measured by using thin-layer chromatography (TLC) to separate lipid extracts from these cells (Extended Data Fig. 1b,c). We next deleted DGAT1 and DGAT2 (DGAT double-knockout (DKO); Extended Data Fig. 1d,e) in HAP1 cells, making them resistant to lipid droplet accumulation through the loading of free fatty acids (in the form of oleic acid; Extended Data Fig. 1f) as this is a DGAT-dependent process5.

We designed a genetic screen in haploid human cells to identify regulators of alternative TAG accumulation, reasoning that carrying out this screen in cells lacking DGAT1 and DGAT2 would allow us to identify regulators of an alternative pathway. Applying gene-trap mutagenesis coupled with fluorescence-activated cell sorting (FACS) and deep sequencing17, we used lipid droplets as a fluorescent surrogate for TAG levels (Fig. 1a). Notably, mutations in transmembrane thioredoxin 1 (TMX1) led to the accumulation of lipid droplets independently of DGAT molecules and also in the absence of free fatty acid loading (Fig. 1b). This result was further validated by the genetic disruption of TMX1 in either a wild-type (WT) or DGAT DKO background in HAP1 cells (Fig. 1c–e and Extended Data Fig. 2a), as well as in 293T cells (Fig. 1f). TLC analysis of TMX1-null HAP1 and 293T cells revealed robust accumulation of TAG in the absence of DGAT activity, with an amount similar to that of oleic acid-induced TAG in DGAT-competent cells (Fig. 1g,h). We observed a similar induction of DGAT-independent TAG accumulation after TMX1 disruption in A549 and U2OS cells (Extended Data Fig. 2b). TMX1 encodes a transmembrane, ER-resident oxidoreductase (Extended Data Fig. 2c,d) that has a cysteine-containing thioredoxin domain located in the lumen of the ER18. Mutation of these redox cysteines correlated with a substantial reduction in TMX1 protein abundance (Extended Data Fig. 2e). All TMX family members are expressed in HAP1 cells17, but TMX1 transcripts are the most abundant (Extended Data Fig. 2f). We therefore introduced other TMX family members into cells that lacked TMX1 (Extended Data Fig. 2g,h). Unlike TMX1, TMX3 and the evolutionarily similar TMX4 failed to suppress TAG and lipid droplet accumulation in these cells (Extended Data Fig. 2i–k); TMX2 was expressed at lower levels. The loss of TMX1 therefore activates TAG accumulation in human cells by a mechanism that is independent of the known TAG biosynthetic process.

Fig. 1: TMX1 suppresses alternative TAG accumulation.
figure 1

a, Schematic of a haploid genetic screen in DGAT DKO cells using the stain BODIPY 493/503. Low and high represent the 5% of cells with the lowest and highest fluorescent signal, respectively. b, Fishtail plot depicting genetic regulators of lipid droplets in a screen of DGAT DKO HAP1 cells. Significant positive and negative regulators are coloured light blue and orange, respectively. The mutational index (MI) represents the ratio of inactivating gene-trap mutations per gene recovered from each (high and low) population (see Methods for a complete description). c, Immunoblot of TMX1 levels in HAP1 cell lines. WB, western blot; PDI, protein disulfide isomerase; LDHA, lactate dehydrogenase A. d, Quantitative increase in lipid droplets (visualized by BODIPY 493/503) in TMX1-knockout (ΔTMX1) HAP1 cells, as measured by flow cytometry. e, Lipid droplets, visualized by BODIPY 665/676 (green in the overlay), in HAP1 cell lines, including a double-DGAT and TMX1 triple knockout (3KO). Blue, Hoechst 33342. Scale bar, 10 μm. f, Lipid droplets (visualized by BODIPY 665/676) in ∆TMX1 293T cells. Blue, Hoechst 33342. Scale bar, 10 μm. g, TLC of neutral lipids in HAP1 (left) and 293T (right) cell lines. Cells were pulsed with 50 µM oleic acid (OA) for 24 h where indicated. 293T cells were additionally treated with 10 µM (each) DGAT inhibitor (DGATi) where indicated. h, Quantification of the increase in TAG induced by TMX1 deletion in HAP1 and 293T cell lines, normalized to WT cells. Data are mean ± s.e.m. of n = 3 independent experiments (two-way ANOVA, Bonferroni correction; each cell line was analysed separately).

Source Data

Alternative TAG accumulation by DIESL

To find out how TMX1-deficient cells accumulate TAG (Fig. 2a), we carried out a genetic suppressor screen in this condition to identify the underlying mechanism, again using lipid droplet accumulation as a readout. We then compared the results of this screen with a screen in WT cells loaded with oleic acid (Fig. 2b). Taken together, these two genetic maps (Fig. 2c) revealed lipogenesis to be a major driver of lipid droplet accumulation common to both backgrounds, with sphingolipid metabolism and mitochondrial electron transport as minor positive and negative regulators, respectively (Extended Data Fig. 3). Moreover, both screens identified SOAT1, which encodes sterol O-acyltransferase, as a positive regulator of lipid droplet biogenesis (Extended Data Fig. 3). Cholesteryl esters co-accumulate with TAG in lipid droplets3, although SOAT1 activity by itself did not explain the accumulation of TAGs. Comparison of the mutational biases of regulators in both screens identified DGAT1 as an important driver of lipid droplets in cells loaded with oleic acid5 (Fig. 2c). Notably, our study revealed a functionally uncharacterized gene, TMEM68, as a selective driver of lipid droplet accumulation in the absence of TMX1 (Fig. 2c). For reasons discussed below, we renamed this gene DGAT1/2-independent enzyme synthesizing storage lipids (DIESL). DGAT1 and DIESL were the strongest selective drivers of lipid droplet accumulation in response to oleic acid loading and TMX1 loss, respectively (Fig. 2d). DIESL-knockout cells were resistant to lipid droplet accumulation when TMX1 was disrupted (Fig. 2e–g) but were phenotypically normal when loaded with oleic acid (Fig. 2g), which confirmed the independence of the TMX1–DIESL genetic pathway from the canonical DGAT pathway9. In agreement with these observations, the loss of TMX1 induced TAG accumulation that was carried out by DIESL, rather than by DGAT1 or DGAT2 (Fig. 2h). These genetic studies demonstrate that DIESL and the two DGAT enzymes promote TAG accumulation independently in different cellular contexts.

Fig. 2: DIESL drives TAG accumulation in the absence of TMX1.
figure 2

a, Schematic representation of the DGAT pathway (green box) and the putative TMX1-inhibited pathway (orange box). FA, fatty acid or fatty acyl. b, Set-up of the modifier screens used to identify the regulators of each pathway. c, Fishtail plots of lipid droplet screens in WT HAP1 cells treated with oleic acid (OA) (left) and in ∆TMX1 HAP1 cells (right). Significant positive and negative regulators are coloured light blue and orange, respectively; larger dots indicate the genes of interest. d, Difference in mutational index (log2-transformed) between the two screens for every gene with at least 30 insertions in each screen. e, Immunoblot of TMX1 in WT, ∆DIESL and DGAT DKO HAP1 cells transduced with a synthetic guide RNA targeting TMX1 (sgTMX1). f, Ultrastructural analysis of lipid droplets (red arrowheads) in WT and ∆DIESL cells after loss of TMX1. Scale bar, 1 μm. g, Lipid droplets, visualized by BODIPY 665/676, in WT, ∆DIESL and DGAT DKO cells transduced with the indicated sgRNA or treated with 200 µM oleic acid for 24 h. sgCTRL is a control sgRNA. Scale bars, 10 μm. h, TLC analysis of neutral lipids (left) and quantification of TAG (right) in DGAT DKO HAP1 cells also lacking DIESL and/or TMX1. Bars represent mean ± s.e.m. of n = 3 independent experiments (two-way ANOVA, Bonferroni correction).

Source Data

DIESL is a TMX1-bound acyltransferase

DIESL encodes a protein that is localized to the ER membrane19 (Extended Data Fig. 4a). We observed that genetic disruption of TMX1 was associated with altered abundance of DIESL protein (Extended Data Fig. 4b). When protein synthesis was inhibited, TMX1 deficiency led to the increased turnover of DIESL (Extended Data Fig. 4c), which suggested the existence of a stable TMX1–DIESL complex. Analysing the binding partners of DIESL by chemical crosslinking, we found a single, crosslinked band at around 70 kDa by SDS polyacrylamide gel electrophoresis (SDS–PAGE) and subsequent immunoblotting (Fig. 3a). We determined that this 70-kDa band was a complex composed of both TMX1 and DIESL because it was detected by an antibody raised against TMX1 and was absent from TMX1-knockout cells (HAP1 cells in Fig. 3a and HeLa cells in Extended Data Fig. 4d). Moreover, we were able to co-immunoprecipitate TMX1 with DIESL (Fig. 3b and Extended Data Fig. 4e) in a detergent-dependent manner (Extended Data Fig. 4f). These data show that both TMX1 and DIESL are proximal (accessible by formaldehyde crosslinking) and can interact in a membrane-dependent manner.

Fig. 3: The TMX1–DIESL enzymatic complex drives alternative triglyceride synthesis.
figure 3

a, Immunoblot analysis of TMX1 and haemagglutinin (HA)-tagged DIESL (3×HA–DIESL) in HAP1 cells lacking endogenous DIESL (and TMX1), with or without crosslinking by 1% paraformaldehyde (PFA). Red arrowheads indicate the DIESL–TMX1 heterodimer, and the asterisk indicates a non-DIESL band. b, Co-immunoprecipitation of TMX1 with DIESL from rescued HAP1 cells (the asterisk indicates antibody chains). CANX, calnexin; PDI, protein disulfide isomerase; EIF4G, eukaryotic translation initiation factor 4G. c, Sequence conservation of catalytic dyads composed of a histidine (H, blue asterisk) and an aspartate (D, red asterisk) across acyltransferases (Uniprot accession numbers in parentheses). Species are Homo sapiens, Caenorhabditis elegans, Arabidopsis thaliana, Saccharomyces cerevisiae, Mycobacterium leprae, Chlamydia trachomatis, Escherichia coli and Thermotoga maritima. d, Western blot of HAP1 cells rescued with WT or catalytic-dead (H130A) DIESL (the asterisk indicates a non-specific band). FASN, fatty acid synthase. e, Analysis of lipid droplets (BODIPY 493/503 fluorescence intensity) in DIESL-rescued HAP1 cells by flow cytometry.

DIESL has previously been characterized broadly as an ER-resident acyltransferase of otherwise-unknown function19. Homology modelling of DIESL revealed that its C terminus adopts an acyltransferase fold similar to that of a bacterial acylglycerol-phosphate acyltransferase, which belongs to an enzyme family that can carry out diverse forms of acylation reactions20. The DIESL N terminus contains a conserved N-glycosylation site on Asn5 that could be validated experimentally (Extended Data Fig. 4g–i), implying that this extremity is exposed to the ER lumen. These glycans could be elongated by treating cells with brefeldin A (Extended Data Fig. 4j,k), which allows Golgi glycosylases to access ER-resident proteins by merging these two organelles21, supporting the localization of DIESL to the ER. Loss of glycosylation did not affect the ability of DIESL to drive TAG accumulation in the absence of TMX1 (Extended Data Fig. 4l). AlphaFold22 predicts an extended helical N terminus that would pass through the membrane (Extended Data Fig. 4m). The expression of truncated DIESL lacking this N-terminal luminal domain was not detected by immunoblot analysis (Extended Data Fig. 4n,o). By sequence alignment, we could identify the active site as a catalytic dyad composed of His130 and Asp136 (Fig. 3c) contained within a catalytic pocket facing the membrane (Extended Data Fig. 4p). Overexpressing DIESL induced the accumulation of lipid droplets only in the absence of TMX1, and this was abolished by mutation of the DIESL active site (H130A; Fig. 3d,e). These findings indicate that DIESL resides in the ER membrane with TMX1 and catalyses DGAT-independent TAG accumulation in an enzymatic manner.


We then studied the effects of the activated DIESL acyltransferase on the cellular lipidome. We began by generating ‘TAG-null’ HAP1 cells that lacked DIESL and the DGATs (∆DGAT1DGAT2DIESL; Extended Data Fig. 5a) and reintroduced either catalytic active or inactive DIESL in the presence or absence of TMX1, observing robust TAG accumulation under the control of both the DIESL active site and its regulator, TMX1 (Extended Data Fig. 5b). We then analysed the lipidome in DIESL-reconstituted quadruple knockouts (∆DGAT1DGAT2DIESLTMX1 (4KO)) and compared the effects of reintroducing catalytic active and inactive DIESL (Fig. 4a). The most pronounced change was observed for TAGs, which were not detected in control DIESL(H130A) cells but made up 20.5% of the lipidome in cells expressing active DIESL (Extended Data Fig. 5c). We also observed an increase in cholesteryl esters, which was not surprising because we had identified SOAT1 in our genetic screens (Extended Data Fig. 3). DIESL activity reduced the fractional abundance of only two lipid species (Fig. 4b,c and Extended Data Fig. 5d–g): DAG and the major membrane phospholipid phosphatidylcholine (PC), as well as its ether-linked form (ePC). This unbiased lipidomic analysis demonstrates that DIESL primarily stimulates TAG abundance, possibly by the acylation of DAG using a phospholipid (or a phospholipid precursor) as an acyl donor.

Fig. 4: TAG synthesis by DIESL.
figure 4

a, Lipidomic analysis of 4KO (∆DGAT1DGAT2DIESLTMX1) HAP1 cells. b, Relative abundance of TAG, DAG and (e)PC in 4KO HAP1 cells reconstituted with 3×HA–DIESL. Bars represent mean ± s.e.m. of n = 3 independent samples (two-way ANOVA, Bonferroni correction). c, Change in abundance of detected lipids (1,183 species) between 4KO HAP1 cells expressing WT or H130A DIESL. PC (red), ePC (orange) and TAG (blue) are indicated. Circle size depicts the relative, scaled abundance of the indicated molecule. d, Reconstitution of TAG synthesis in E.coli by DIESL (top) and schematic representation of human (h) DIESL constructs for expression in E. coli (bottom). Blue, pectate lysate B (pelB) signal sequence (ss); black arrowhead, cleavage site. e, TLC separation of neutral lipids and immunoblot analysis from E. coli expressing the indicated construct. f, Absolute abundance of TAG and DAG in E. coli expressing either ss-hDIESL or empty vector. Bars represent mean ± s.e.m. of n = 3 independent samples (two-way ANOVA, Bonferroni correction). g, Ultrastructural analysis of E.coli expressing ss-hDIESL H130A (control) or WT. Red arrowheads indicate lipid-rich inclusions. Scale bar, 500 nm. h, Left, cell-free reconstitution of TAG synthesis by DIESL. Lysates from control 4KO HAP1 cells or those expressing 3×HA-DIESL were incubated with 50 µM 14C-DAG for the indicated time period at 37 °C unless otherwise indicated. Lipid extracts were separated by TLC and analysed by phosphorimaging. Right, intensities were quantified for n = 3 independent experiments. The line connects the mean of each time point for both conditions (two-way ANOVA, Bonferroni correction). NS, not significant. i, Acylation assay of NBD–DAG. HAP1 cell lines were treated with 25 µM NBD–DAG for 1 h before TLC analysis of polar lipids (to accommodate the charge on the NBD group). NBD-tagged lipids were identified by NBD fluorescence.

Source Data

To demonstrate that DIESL is an autonomous TAG synthase, we reconstituted DIESL-dependent TAG production in E.coli, a non-oleaginous organism that lacks the genes required for TAG synthesis23. Expression of human DIESL in E.coli (preceded by a bacterial membrane-targeting sequence; Fig. 4d) conferred TAG synthesis to this organism, as determined by TLC (Fig. 4e) and mass spectrometry (MS) (Fig. 4f), which came at the expense of DAG levels. We additionally observed the formation of lipid inclusion bodies in these bacteria (Fig. 4g). We then examined DIESL-dependent DAG-to-TAG conversion using a cell-free assay. Incubation of HAP1 4KO (∆DGAT1DGAT2DIESLTMX1) lysates with isotope-labelled DAG demonstrated its time-dependent conversion to TAG, which was stimulated by the presence of DIESL and was sensitive to heat inactivation and low temperature (Fig. 4h). We also measured this conversion in intact cells using fluorescent DAG (nitrobenzoxadiazole-labelled DAG (NBD–DAG); Extended Data Fig. 6a). Whereas triple-knockout, TAG synthase-null cells showed no detectable conversion of DAG to TAG, DIESL was able to stimulate DAG acylation when TMX1 was inactivated in a manner strictly dependent on its catalytic acyltransferase site (Fig. 4i and Extended Data Fig. 6b–e). These data show that DIESL functions as a DAG acyltransferase and that its exogenous expression confers the ability to synthesize TAG to an organism inherently devoid of this capacity.

DIESL affects energy homeostasis

We next addressed the role and importance of DIESL-synthesized TAGs. During conditions of nutrient abundance (in the form of extracellular free fatty acids), the DGAT enzymes are known to synthesize TAGs, and we observed that this process was unaffected by the loss of DIESL (Fig. 2g). Accordingly, measuring the amount of TAG in several cell lines revealed that DIESL could noticeably affect the cellular TAG pool in the absence of DGAT activity (Extended Data Fig. 7). We therefore focused on the function of DIESL when the DGAT enzymes would be inactive. Because both TMX1 (as previously shown24) and DIESL reside at the mitochondria-associated membrane of the ER (Fig. 5a), we reasoned that TMX1–DIESL might support mitochondrial function during periods of lipid starvation, when decreased DGAT-dependent TAG formation would necessitate a reliance on an alternative means of TAG synthesis. To examine mitochondrial activity, we analysed the activation of AMP kinase (AMPK), a master metabolic sensor that directly detects the energy state of the cell25,26. Four DIESL-deficient cell lines (RPE1, U251, HT29 and 293T) were sensitized to AMPK phosphorylation, which occurred when extracellular lipids were depleted or the DGAT enzymes were inhibited (Extended Data Fig. 8a,b). Focusing on untransformed RPE1 cells, we measured increased mitochondrial reactive oxygen species in the absence of DIESL, a feature that was exacerbated during lipoprotein starvation (Extended Data Fig. 8c). This starvation response did not cause widespread autophagy (Extended Data Fig. 8d,e). DIESL deficiency in RPE1 cells also decreased the mitochondrial membrane potential and ATP levels when cells were cultured in the absence of lipoproteins, but this could be rescued with oleic acid (Fig. 5b). This rescue was sensitive to the inhibition of carnitine palmitoyltransferase 1A (CPT1A) by etomoxir, demonstrating that it was dependent on fatty acid uptake by mitochondria (Extended Data Fig. 8f). RPE1 cells that lacked DIESL showed impaired cellular fitness when cultured under lipoprotein-depleted conditions (Fig. 5c). Culturing more resilient cell lines, such as 293T and U251 cells, under more severe starvation conditions also sensitized them to DIESL loss (Extended Data Fig. 8g). Thus, during various starvation stimuli across several cell types, DIESL was shown to contribute to mitochondrial function, energy levels and cell fitness.

Fig. 5: DIESL deficiency in cells and mice.
figure 5

a, Subcellular fractionation of HAP1 cells. Hom., homogenate; P100k, pellet obtained by ultracentrifugation at 100,000g; S100k, 100,000g supernatant; Mito., mitochondria; MAM, mitochondria-associated membrane of the ER. Markers of light membranes and cytosol (CLTC), as well as mitochondria (MT-ND2), are included. b, Mitochondrial membrane potential (∆ψ, left) and ATP levels (right) in RPE1 cells, cultured in either complete medium (control), medium lacking lipoproteins or lipoprotein-deficient medium supplemented with 50 µM oleic acid for 24 h. Bars represent mean ± s.e.m. of n = 3 (∆ψ) or n = 4 (ATP) independent experiments, respectively (two-way ANOVA, Bonferroni correction). c, Bright-field images (left) and quantification of viability (right) of the RPE1 cells from b. Bars represent mean ± s.e.m. of n = 3 independent experiments (two-way ANOVA, Bonferroni correction). Scale bar, 50 μm. d, Body weight of adult (22–28-week-old) mice. Bars represent mean ± s.e.m. of n = 8–15 mice (one-way ANOVA, Bonferroni correction). Het., heterozygous mice. e, Postnatal growth curves of male (left) and female (right) mice (Control designates both WT and heterozygous mice). Bars represent mean ± s.e.m. of n = 5–17 mice (two-way ANOVA, Bonferroni correction); male and female mice were analysed separately. f, Hepatic TAG levels in male mice, quantified at 3.5 and 6 weeks (arrows in e), expressed as a percentage of total lipids quantified (Extended Data Fig. 9l). Bars represent mean ± s.e.m. of n = 4–6 livers per condition (two-way ANOVA performed on the entire dataset, Bonferroni correction). g, Respiration exchange ratio (RER). Left, an increased RER indicates preferential carbohydrate oxidation (instead of lipids). Right, quantification of RER in male mice. Bars represent mean ± s.e.m. of n = 3 mice (two-way ANOVA, Bonferroni correction). h, The DGAT pathway (green) acylates DAG using exogenously derived fatty acyl-CoA (FA-CoA) whereas the TMX1–DIESL pathway (orange) uses endogenous fatty acyl chains derived from membrane phospholipids or their precursors.

Source Data

To study the role of DIESL in vivo, we generated knockout mice (Extended Data Fig. 9a–c). Diesl-knockout mice were viable, born at the expected Mendelian ratio (Methods) and displayed a reduction in male (17%) and female (19%) body weight at adulthood (more than 5 months old) (Fig. 5d). The reduction in body weight was accompanied by a decrease in body length, a roughly 50% reduction in fat mass (Extended Data Fig. 9d,e) and a reduction in TAG levels in serum and brain (Extended Data Fig. 9f,g). The organs of Diesl-knockout mice were of normal size (when corrected for body weight) and morphology (Extended Data Fig. 9h,i). It thus follows that the TAG synthase DIESL is required to attain normal body weight and size in mice.

We studied the consequences of Diesl deficiency in mice in more detail. Notably, knockout mice were indistinguishable from their WT counterparts one week after birth but grew at a much slower rate between weeks two and four (Fig. 5e). The weight difference we found after four weeks was maintained throughout adulthood (Fig. 5d). Knockout mice displayed a small decrease in food intake that was absent when adjusted for body weight (Extended Data Fig. 9j). Thus, Diesl is required to support rapid growth at a specific time point after birth, which notably coincides with a dietary switch from milk to chow. By MS, we detected a roughly 40% reduction in liver TAG at 3.5 weeks of age, but this was no longer present several weeks later when WT and knockout mice grew at similar rates (Fig. 5f and Extended Data Fig. 9k,l). Because we measured low TAG levels, we reasoned that DIESL could provide the energy required for rapid growth at this stage of life because fuel sources were limiting. We used indirect calorimetry in fasted adult mice to measure energy homeostasis under conditions of nutrient shortage. We observed an increased respiratory exchange ratio in response to fasting in Diesl-deficient mice (significant by ANOVA, although the number of animals was too low for reliable regression analysis; Fig. 5g), indicating an attenuated switch from carbohydrate to lipid oxidation compared with WT mice. Taken together, our data show that TMX1–DIESL is essential to support the rapid organismal growth of young animals and regulates a response to dynamic changes in nutrient availability.


The enzymatic synthesis of triglyceride from fatty acyl-CoA precursors was described more than 60 years ago9. Using haploid genetics, we here identify a regulated protein module, composed of TMX1 and DIESL within the ER, that carries out cell-autonomous alternative TAG synthesis. In contrast to known TAG biosynthetic routes that are activated by high extracellular lipid levels, DIESL converts endogenous acyl chains into TAGs, altering the composition of cellular membranes. In cells, TMX1–DIESL may enable the supply of fatty acids to mitochondria during times of lipid or nutrient scarcity. TAG synthesis by TMX1–DIESL at the expense of membrane phospholipids during nutrient limitation is reminiscent of an autophagy-like mechanism and thus would require tight regulation by TMX1 (Fig. 5h). How the activity of the TMX1–DIESL complex is precisely tuned to the needs of the mitochondrion (perhaps by oxidative signals, as TMX1 contains a thioredoxin domain) remains to be elucidated.

Loss of Diesl leads to metabolic phenotypes in mice, indicating the importance of alternative TAG synthesis in normal physiology. In mice two to four weeks old, we observed a significant delay in rapid postnatal growth that coincides with a dietary switch from milk (20% fat27) to chow (4% fat). Furthermore, fasting of adult mice lacking Diesl suggests an impaired switch from carbohydrates to lipids for nutrient oxidation. These observations point to a role for DIESL during fluctuations or limitations in nutrient supply.

Studies of DGAT-dependent TAG synthesis during conditions of high lipid accumulation have led to them being examined as drug targets in disease (DGAT1 in obesity6 and DGAT2 in fatty liver disease14). Conversely, TAG synthesis during nutrient restriction has received less attention but may also have a prominent biological role. Here, we ascribe such a function to TMX1–DIESL. It is conceivable that organisms would have evolved such a pathway to mitigate the consequences of food scarcity or extreme physical endurance, and these activities may have important roles in pathologies that are seriously affected by cellular energy availability, such as cancer and neurodegeneration.


Cell culture, reagents and antibodies

HAP1 cells were maintained in IMDM medium supplemented with 10% FBS, L-glutamine, penicillin and streptomycin, and cultured at 37 °C in 5% CO2. 293T, A549, HeLa, HT29, RPE1, U251 and U2OS cells (purchased from ATCC) were maintained in DMEM under the same conditions. Cell lines used in this study were routinely monitored for mycoplasma contamination. BODIPY 493/503 and 665/676 neutral lipid dyes, TMRM and MitoTracker Red CM-H2XROS were from ThermoFisher. Brefeldin A, CCCP, cycloheximide, etomoxir, lipoprotein-deficient fetal calf serum, oleic acid, PF-06424439 (a DGAT2 inhibitor) and phosphatase-inhibitor cocktail were from Sigma-Aldrich. A-922500 (a DGAT1 inhibitor) was from Selleck. Protease inhibitor cocktail was from Roche. TAG (triolein) and DAG (diolein) were from Avanti Polar Lipids. 14C-DAG (1,2,-dioleoyl-rac-glycerol) was from American Radiolabeled Chemicals. Antibodies used in this study were anti-ACTB (Abcam, ab6276, WB 1:10,000), anti-α-tubulin (Santa Cruz Biotechnology, sc-32293, WB 1:1,000), anti-AMPK (Cell Signaling Technology, 2532, WB 1:1,000), anti-AMPK pT172 (Cell Signaling Technology, 2535, WB 1:1,000), anti-CANX (Abcam, ab22595, immunofluorescence (IF) 1:100, WB 1:1,000), anti-CLTC (Thermo Fisher, PA5-17347, WB 1:1,000), anti-EIF4G (Cell Signaling Technology, 2498, WB 1:1,000), anti-FASN (Santa Cruz Biotechnology, sc-55580, WB 1:1,000), anti-HA (Biolegend, 901503, IF 1:200, WB 1:1,000), anti-HSPA5 (Cell Signaling Technology, 3177, WB 1:1,000), anti-LAMP1 (Santa Cruz Biotechnology, sc-19992, WB 1:1,000), anti-LC3B (Cell Signaling Technology, 2775, WB 1:1,000), anti-LDHA (Cell Signaling Technology, 3582, WB 1:5,000), anti-PDI (Abcam, ab2792, IF 1:500, WB 1:20,000), anti-S6 (Cell Signaling Technology, 2137, WB 1:1,000), anti-S6 pS235/pS236 (Cell Signaling Technology, 4856, WB 1:1,000), anti-TMX1 (Atlas Antibodies, HPA003085, IF 1:100, WB 1:1,000 and Origene, TA507042, WB 1:1,000), anti-TOMM20 (Abcam, ab186735, WB 1:10,000) and anti-V5 (ThermoFisher, 14-6796-82, IF 1:500, WB 1:1,000).

Plasmids and cloning

We purchased sgRNAs as short single-stranded DNA with sticky ends. These oligonucleotides were then annealed and cloned into pX330 and pLentiCRISPRv2 (with puromycin, blasticidin or mCherry selection markers) that had been cut with BbsI or BsmBI (New England BioLabs), respectively. As a non-targeting control sgRNA (sgCTRL), a sequence targeting the zebrafish tia gene was used28. A complete list of sgRNAs used in this study, generated using the Broad Institute’s Genetic Perturbation Platform (, is provided in Supplementary Table 1. 3×HA-tagged DIESL (Q96MH6-1, WT and N0) and V5-tagged TMX1 (Q9H3N1), TMX2 (Q9Y320-1), TMX3 (Q96JJ7-1) and TMX4 (Q9H1E5) were purchased as linear DNA flanked by NheI and AgeI restriction sites from Integrated DNA Technologies. These were then digested and cloned into pLEX305 backbones. DIESL(N5Q) and DIESL(H130A) mutants were generated by site-directed mutagenesis using the QuikChange II kit (Agilent). For recombinant expression of DIESL, human DIESL was codon-optimized for E.coli expression and pET24 constructs encoding DIESL, pelB–DIESL and pelB–3×HA–DIESL were purchased as plasmids from Twist Bioscience.

Generation of clonal knockout cell lines

HAP1 and 293T cells were transfected with pX330 encoding the sgRNA of choice, along with a plasmid either carrying an integration cassette28 or encoding a blasticidin or puromycin resistance gene, using Xfect (Takara), according to the manufacturer’s instructions. Transfected cells were selected with 25 µg ml−1 blasticidin (HAP1) or 1 µg  ml−1 puromycin (HAP1 and 293T). After selection, the medium was replaced with complete medium and clones were allowed to form colonies, which were eventually picked by micropipette and transferred to 24-well plates. Genetic modification of individual clones was detected by PCR, in some cases using a primer annealing to the sequence of the blasticidin (5′-CCGACATGGTGCTTGTTGTCCTC-3′) or puromycin (5′-GCAACCTCCCCTTCTACGAGC-3′) resistance gene. See Supplementary Table 2 for a list of primers used to amplify genomic loci. Disruption of the locus was confirmed by Sanger sequencing, either directly or, in the case of compound heterozygotes, using TIDE analysis29 for 293T cell lines. The targeted genetic modifications in cell lines used in this study are summarized in Supplementary Table 3.

Lentiviral transduction

Lentivirus was produced in 293T cells transfected with p∆VPR, pVSVg and either pLEX305 or pLentiCRISPRv2 lentiviral plasmids, as well as pAdVAntage. Then, 2 days after transfection, virus was collected from conditioned medium passed through a 40 µm filter. Viral supernatants, supplemented with 8 µg ml−1 protamine sulfate, were applied directly to cells for 24 h. In some cases, a second collection and transduction was done to improve efficiency. Transduced cells were selected with puromycin (in HAP1, 1 µg  ml−1; A549, 0.5 µg  ml−1; HeLa, 2 µg  ml−1; HT29, 3 µg  ml−1; U251, 2 µg  ml−1; U2OS, 2 µg  ml−1) or transduction efficiency was assessed by expression of a fluorescent marker (mCherry). RPE1 cells were transduced with a large viral titre and were not selected. Transduced cells were analysed between 7 and 28 days after transduction.

Mutagenesis screening

All genetic screens reported here used gene-trap mutagenesis in haploid HAP1 cells, as described previously17. Typically, 2–3 × 109 gene-trapped HAP1 cells of the indicated genotype were collected by trypsinization and fixed in Fix Buffer I (BD Biosciences) for 10 min at 37 °C. For the oleic acid-loaded screen, cells were first cultured for 24 h in complete medium supplemented with 200 µM oleic acid and then chased in medium lacking oleic acid for another 24 h before collection. Cells were treated with 1 mg ml−1 RNase A (Qiagen) diluted in FACS buffer (10% FBS in PBS) at 37 °C for 30 min before staining with 1 µg ml−1 BODIPY 493/503 and 10 µg ml−1 propidium iodide (Sigma-Aldrich), diluted in FACS buffer, for 1 h at room temperature. Cells were washed twice in FACS buffer before being passed through a 40 µm cell strainer. Cell sorting was done using an S3 Sorter (Bio-Rad), collecting 107 cells (per population) representing the lowest and highest 5% of BODIPY signal from haploid cells in G1 phase. The isolation of genomic DNA, preparation of sequencing libraries and the analysis were done as described17. Reads were aligned to the reference genome (GRCh37 genome assembly), tolerating one mismatch, and sense insertions in the ‘low’ and ‘high’ populations were compared using the Fisher’s exact test to determine significant differences (P < 0.05). The mutational index represents the ratio of the occurrence of unique, disruptive (that is, insertions (ins.) of the gene-trap in the sense orientation) mutations in the body of a given gene (5′ untranslated region, exon and intron) in the high compared with the low population, normalized by the total of other unique, disruptive mutations in each population17: (sense ins. of gene X in high/(total sense ins. in high − sense ins. of gene X in high))/(sense ins. of gene X in low/(total sense ins. in low − sense ins. of gene X in low)). The number of recovered mutations for each population in each screen is reported in Supplementary Table 4.

SDS–PAGE and immunoblot analysis

Unless otherwise specified, cells were lysed in RIPA buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% SDS and protease-inhibitor cocktail) on ice. Lysates were sonicated twice (40% amplitude for 4 s), cleared by centrifugation and protein concentrations were determined by BCA assay (Thermo Scientific). Equivalent amounts of protein were separated by SDS–PAGE over Bolt 4–12% Bis–Tris gels (Invitrogen) then transferred to nitrocellulose membranes. Membranes were blocked with 5% milk–TBST and incubated overnight at 4 °C in primary antibody diluted in 3% BSA–TBST. The next day, membranes were washed three times in TBST and then incubated in HRP-conjugated secondary antibodies (Invitrogen 65-6120 and Bio-Rad 1706516), diluted 1:10,000 in blocking buffer for 1 h at room temperature. After three more washes in TBST, signal was developed using Clarity Western ECL substrate (Bio-Rad) and imaged on a Universal Hood II GelDoc system (Bio-Rad).

Recombinant expression in E. coli

BL21 E.coli were transformed with pET24 or pET28a plasmids expressing the indicated construct. Single clones were cultured up to a 5 ml volume in LB broth at an optical density of 0.5 at 37 °C. Cultures were then transferred to 25 °C for 1 h and expression was induced with 0.2 mM IPTG for 16 h at 25 °C. For SDS–PAGE and immunoblot analysis, 100 µl of culture was pelleted for 5 min at 3,500g before lysis in SDS–PAGE sample buffer.

TLC and DAG acylation assays

Equal numbers of human cells grown to confluence in six-well plates were washed with PBS. Cells were then incubated twice in 400 µl extraction buffer (hexane:isopropanol at 3:2) for 10 min30. E.coli cell pellets from 2 ml of the overnight culture were incubated in 500 µl extraction buffer (only once) for 30 min while being rotated. Extractions were pooled and dried under a stream of nitrogen to a volume of around 10 µl and then spotted on silica gel 60 TLC plates (Sigma-Aldrich). Plates were placed in a TLC tank containing a mobile phase to separate neutral lipids (hexane:diethyl ether:acetic acid at 80:20:1) or polar lipids (chloroform:methanol:water:acetic acid at 60:50:4:1). After separation, plates were air dried and stained with 0.2% Amido Black 10B (Sigma-Aldrich) dissolved in 1 M NaCl for 15–30  min (ref. 31). After staining, plates were rinsed with water and then washed several times with 1 M NaCl before drying overnight. Plates were imaged the next day on a Universal Hood II or EZ Imager GelDoc system (Bio-Rad).

NBD–DAG acylation in HAP1 cells

Unless otherwise specified, 25 µM NBD–DAG (Avanti Polar Lipids) was first conjugated to 0.125% fatty acid-free BSA in serum-free IMDM for 1 h at 37 °C. Cells were then incubated in this medium for 1 h before lipid extraction. TLC was done as described above. After drying, NBD fluorescence was monitored on a Universal Hood GelDoc system (Bio-Rad) using the fluorescein channel.

Cell-free 14C-DAG acylation

4KO HAP1 cells (naive or expressing 3×HA-DIESL) were washed in ice-cold PBS, collected in ice-cold buffer (20 mM HEPES pH 7.4, 250 mM sucrose, 2 mM MgCl2) and lysed by passing through a glass homogenizer 60 times on ice. Lysates were cleared by centrifugation at 600g for 10 min at 4 °C. Protein content was assayed and lysates were diluted to 2 mg ml−1 and stored at −80 °C. Reconstitution assays were based on assays described previously32 and were performed at the Radionuclides Centre of the Netherlands Cancer Institute according to its guidelines. First, 1.4 µl of 1.8 mM 14C-DAG, 3.6 µl buffer and 5.0 µl of 6.6 mg  ml−1 fatty acid-free BSA were incubated on a heat block for 1 h at 37 °C with shaking at 850 rpm. We then added 100 µl of the indicated lysate and assays were incubated at 37 °C with shaking at 850 rpm for the indicated time (the final 14C-DAG concentration was 50 µM and the total radioactivity per assay was 0.14 µCi). For some controls, the reaction was incubated instead on wet ice or the lysate was heat-inactivated for 20 min at 90 °C before being added to the reaction. Reactions were quenched by performing a modified Bligh and Dyer extraction33; 110 µl chloroform and 110 µl methanol were added to each reaction, mixed and centrifuged at 20,000g for 2 min at room temperature. The bottom organic layer was transferred to a new tube, dried and reconstituted in 8 µl ethanol and separated by neutral-lipid TLC as described above. Plates were dried and imaged using a Typhoon FLA 9500 phosphorimager (General Electric, software v. after exposure to a BAS-TR2949 imaging plate (Fuji) for 3 days.

Immunoprecipitation and crosslinking

Formaldehyde crosslinking

DIESL and ∆TMX1DIESL HAP1 cells reconstituted with 3×HA-DIESL or WT HeLa cells expressing 3×HA-DIESL were washed with PBS and incubated in 1% paraformaldehyde for 20 min at room temperature. Crosslinking reactions were quenched by incubation in 0.2 M glycine for 10 min. Cells were then lysed in RIPA buffer on ice as described above, sonicated, and protein concentrations were determined. Samples were separated by SDS–PAGE and immunoblotted as described above.


DIESL HAP1 cells reconstituted with 3×HA-DIESL and grown to confluence in a six-well plate were lysed in 400 µl IP buffer (25 mM HEPES pH 7.0, 150 mM NaCl, 1% detergent (Tween-20 unless otherwise indicated) and protease- and phosphatase-inhibitor cocktails) on ice, sonicated and cleared by centrifugation. Then 250 µl of the lysate was incubated with 10 µl PBS-washed anti-HA magnetic beads (Pierce) overnight while being rotated at 4 °C. The next day, beads were washed three times in IP buffer before elution in SDS–PAGE sample buffer at 90 °C for 20 min. One-quarter of this eluate (along with the corresponding amount of input sample) were separated by SDS–PAGE and subjected to immunoblot analysis as described above.

Deglycosylation assay

DIESL HAP1 cells reconstituted with 3×HA-DIESL were lysed in buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40, protease-inhibitor cocktail) on ice, sonicated and protein levels were quantified as described above. N-glycans were removed using PNGase F (New England BioLabs) according to the manufacturer’s instructions. Denatured lysates (1 µg µl−1) were incubated for 15 min at 37 °C while being shaken at 850 rpm, in the presence or absence of 25 units µl−1 PNGase F. Reactions were separated by SDS–PAGE and analysed by immunoblotting as described above.

Subcellular fractionation and MAM purification

The MAM of the ER was purified from six confluent 15 cm plates of HAP1 cells, as described previously34, with all steps performed on ice or at 4 °C. Cells were washed in ice-cold PBS, collected in 750 µl per plate ice-cold homogenization buffer (10 mM HEPES pH 7.4, 250 mM sucrose) and passed through a glass homogenizer 100 times. The homogenate was centrifuged for 5 min at 600g. The pellet was resuspended in 2 ml homogenization buffer, homogenized and centrifuged again. Both supernatants were pooled and centrifuged at 10,300g for 20 min. The supernatant was put aside and the pellet (crude mitochondria) was resuspended in 1 ml isolation buffer 1 (5 mM HEPES pH 7.4, 250 mM mannitol, 0.5 mM EGTA). Next, 500 µl of resuspended mitochondrial pellet was layered on top of 1.5 ml Percoll solution (25 mM HEPES pH 7.4, 225 mM mannitol, 1 mM EGTA, 30% (v/v) Percoll) in two ultracentrifuge tubes. Mitochondria were fractionated by centrifugation at 95,000g for 30 min using a TLS-55 swinging-bucket rotor (Beckman). Pure mitochondria and crude MAM layers were collected using a syringe and resuspended in four volumes of isolation buffer 1 and isolation buffer 2 (25 mM HEPES pH 7.4, 225 mM mannitol, 1 mM EGTA), respectively. Purified mitochondria were centrifuged at 10,500g for 10 min. The crude MAM was cleared by centrifugation at 6,300g for 10 min and then both the MAM and the postmitochondrial supernatant from earlier were centrifuged for 1 h at 100,000g. The pellets (microsome and pure MAM) were collected and the cytosolic supernatant was concentrated by centrifugal filtration (10 kDa filter, Amicon). Protein concentration was determined by BCA assay (Thermo Scientific) and fractions were stored at −80 °C before immunoblot analysis.

Electron microscopy

Cells were prepared for transmission electron microscopy as previously described35. Grids were imaged on a Tecnai 12 G2 (ThermoFisher).

Immunofluorescence and confocal microscopy

Cells were grown on glass coverslips and fixed in 4% formaldehyde in PBS for 15 min at room temperature and then washed three times in PBS. Cells were permeabilized with 0.1% Triton X-100 in PBS for 10 min and then washed three times in PBS and blocked in 1% BSA in PBS for 20 min. Cells were stained with primary antibodies, diluted in blocking buffer, for 1 h at room temperature. After three washes in blocking buffer, cells were stained with Alexa Fluor 488-, 568- and 647-conjugated secondary antibodies (Invitrogen A10042, A11011, A11004, A11008, A21245 and A31571) and/or neutral lipid dyes, diluted 1:500 in blocking buffer, for 1 h at room temperature. Cells were washed three times in PBS, counterstained with Hoechst 33342 (Invitrogen) diluted in PBS, and then washed three more times in PBS, before being mounted on glass slides using Aqua Poly/Mount (Polysciences). Cells were imaged by confocal laser-scanning microscopy using a Leica SP5 microscope with a 60× (1.4 NA) objective lens. Images were analysed using ImageJ (NIH).

Flow cytometry

BODIPY 493/503 measurements in fixed HAP1 cells

HAP1 cells grown on 10 cm plates were collected by trypsinization and fixed in Fix Buffer I (BD Biosciences) for 10 min at 37 °C. Cells were pelleted, washed with FACS buffer (10% FBS in PBS), resuspended in FACS buffer and counted. Next, about 10 million cells were stained with 1 µg ml−1 BODIPY 493/503 and 5 µg ml−1 DAPI (Invitrogen), diluted in FACS buffer, for 1 h at room temperature. Cells were washed once in FACS buffer then passed through a 35 µm nylon mesh cell strainer into a FACS tube. Fluorescence was analysed on an LSR Fortessa (BD Biosciences) analytical flow cytometer, using lasers with wavelengths of 405 nm and 488 nm to detect DAPI and BODIPY 493/503, respectively. Data were analysed using FlowJo (BD Life Sciences). Fluorescence plots represent the fluorescent signal measured in single haploid G1 cells, as determined by DAPI fluorescence intensity.

Mitochondrial measurements in live RPE1 cells

RPE1 cells were cultured in the presence or absence of lipoproteins, supplemented with 50 µM oleic acid where indicated, and pulsed for 30 min with either 600 nM TMRM or 250 nM MitoTracker Red CM-H2XROS in medium depleted of lipoproteins. In TMRM experiments, cells were then incubated in 150 mM TMRM for an additional 30 min, using 20 µM carbonyl cyanide m-chlorophenylhydrazone (CCCP) as a positive control for depolarization. Cells were collected by trypsinization and stored on ice in FACS buffer. Fluorescence (using a 561 nm laser) and data were analysed as described above. Fluorescence plots represent the fluorescent signal measured in single cells. Membrane potential was calculated by subtracting the fluorescence mean of depolarized (CCCP-treated) cells.

ATP measurements

RPE1 cells were plated in 96-well plates and ATP levels were measured using CellTitre-Glo (Promega) according to the manufacturer’s instructions. ATP levels were normalized by measuring protein content in a parallel plate by BCA assay (Thermo Scientific).


Generation and genotyping

Animal experiments were performed with the approval of the National Ethics Committee for Animal Experiments of the Netherlands, in accordance with the relevant guidelines and regulations (including laboratory and biosafety regulations). Frozen embryos (morula stage, from C57/BL6J mice) carrying a single disruptive allele in Diesl (also known as Tmem68) tm1a(EUCOMM)Wtsi (herein referred to as gt(lacZ-neo)), were purchased from the European Conditional Mouse Mutagenesis Program (EUCOMM). After thawing, embryos were developed into blastocysts overnight in KSOM medium in an incubator at 37 °C, and were then implanted into C57/BL6N foster female mice and carried to term. Mice were genotyped by PCR, using primer combinations to detect the WT (5′-GCTCCCTTCCATTTACTCTG-3′ and 5′-CCGGTGAGATAGCTAACAAG-3′) and mutant (5′-CTTATCATGTCTGGATCCGG-3′ and 5′-CCGGTGAGATAGCTAACAAG-3′) alleles. Because C57/BL6J mice have a deletion in the Nnt gene that influences DIESL protein levels36, only mice with 6N alleles at the Nnt locus were used in this study, and this was monitored by PCR, using primer combinations to detect the 6N (5′-GGGCATAGGAAGCAAATACCAAGTTG-3′ and 5′-GTAGGGCCAACTGTTTCTGCATGA-3′) and 6J (5′-GTGGAATTCCGCTGAGAGAACTCTT-3′ and 5′-GTAGGGCCAACTGTTTCTGCATGA-3′) alleles. To generate mice for analysis, mice heterozygous for the mutant Diesl allele were intercrossed to generate knockout and control mice. Mice were born at the expected Mendelian ratio (P = 0.4279, chi-squared test, n = 341 mice) and maintained on a diet of RM3-SAFE (4.2% fat, Special Diet Services). Mice were maintained in a certified animal facility at 21 °C and 55% humidity in 12 h:12 h light:dark cycles. No sample-size calculations were performed. No randomization was performed because animals were assayed based on genotype. In experiments where subjectivity could be introduced, the experimenter was blinded to the details.

PCR with reverse transcription

RNA from mouse liver tissue was isolated using the Qiagen RNeasy Mini Kit, and cDNA libraries were synthesized with SuperScript III reverse transcriptase (Invitrogen) using random hexamers. The DIESL transcript was detected by PCR using a primer pair spanning exons 4 and 5 (5′-GAGCCATCCCCATAGACTTTTACTACTTC-3′ and 5′-CCCGGTGAGATAGCTAACAAGTGAC-3′). In the knockout, the disruptive cassette was integrated between these two exons. The ACTB transcript was used as a control and was amplified using the following primer pair: 5′-ATCCTGACCCTGAAGTACCCCA-3′ and 5′-CCTCTCAGCTGTGGTGGTGAAGCTGTAGCCACGCT-3′. Although DIESL has previously been reported to be a brain-specific protein19, we readily detected DIESL expression in the mouse liver, in agreement with proteomic data36. Furthermore, expression data from the Human Protein Atlas indicates that DIESL expression could be detected in all tissues that were tested37.


Tissues and organs were collected and fixed in EAF fixative (ethanol:acetic acid:formaldehyde:saline at 40:5:10:45 v:v) and embedded in paraffin. Sections were prepared at 2 µm thickness from the paraffin blocks and stained with haematoxylin and eosin according to standard procedures. The sections were reviewed with a Zeiss Axioskop2 Plus microscope (Carl Zeiss Microscopy) and images were captured with a Zeiss AxioCam HRc digital camera and processed with AxioVision 4 software (both from Carl Zeiss Vision).

Indirect calorimetry

Indirect calorimetry was done using fully automated metabolic cages (LabMaster, TSE systems). Mice were housed individually in a 12 h:12 h light:dark cycle (light, 07:00–19:00 hours) with ad libitum access to water and a standard chow diet (10% fat, 23% protein and 67% carbohydrate; V1554-703, Ssniff Spezialdiäten). After at least 24 h of acclimatization, O2 consumption (\({V}_{{{\rm{O}}}_{2}}\)), CO2 production (\({V}_{{{\rm{CO}}}_{2}}\)) and caloric intake were measured for at least three consecutive days followed by 12 h fasting and refeeding. The RER was calculated as \({V}_{{{\rm{CO}}}_{2}}/{V}_{{{\rm{O}}}_{2}}\). Glucose oxidation was calculated using the formula \(((4.585\times {V}_{{{\rm{CO}}}_{2}})-(3.226\times {V}_{{{\rm{O}}}_{2}}))\times 4\) and fat oxidation was calculated as \(((1.695\times {V}_{{{\rm{O}}}_{2}})-(1.701\times {V}_{{{\rm{C}}{\rm{O}}}_{2}}))\times 9\). Values were corrected for the lean mass of each individual mouse.

Body composition

Fat mass and lean tissue mass were assessed at several time points in non-anaesthetized mice using NMR (MiniSpec LF90 BCA-analyser).

Shotgun lipidomics

Sample preparation

4KO (∆DGAT1DGAT2DIESLTMX1) HAP1 cells, reconstituted with WT or H130A DIESL, were cultured in complete medium. Prior to collection, cells were cultured in serum-free IMDM for 30 min. Cells were then collected by trypsinization, washed twice with PBS and 3 million cells (in triplicate) were pelleted and stored at −80 °C. For E.coli samples, 2 ml of induced culture (as described above, in triplicate) was pelleted and incubated overnight at −80 °C. The next day, bacterial pellets were thawed and incubated in 25 µg ml−1 lysozyme in PBS for 30 min. Cells were pelleted, washed in PBS and pelleted again before resuspension in distilled water. Cells were sonicated at 60% amplitude for 5 min, in continuous cycles of 5 s on and 25 s off. The insoluble fraction was pelleted by centrifugation and the lysate (supernatant) was quantified for protein and stored at −80 °C. For mouse tissue samples, tissues were collected from mice fed ad libitum and stored at −80 °C. Tissues were then thawed on ice, weighed and homogenized in PBS (50 mg ml−1 tissue) using a glass homogenizer with 20 and 40 strokes for brain and liver tissue, respectively. Homogenates were diluted 10-fold further in PBS and homogenized again with 20 strokes. Homogenates (5 mg ml−1) were stored at −80 °C. Lipid abundancies are expressed either in weight (where indicated, corrected for the amount of protein in the sample) or as a molar percentage of the total lipid species in the sample (mol%).

Lipid extraction for MS lipidomics

MS-based lipid analysis was performed by Lipotype as described38. Lipids were extracted using a two-step chloroform–methanol procedure39. Samples were spiked with internal lipid standard mixture containing: cardiolipin 14:0/14:0/14:0/14:0 (CL), ceramide 18:1;2/17:0 (Cer), DAG 17:0/17:0, hexosylceramide 18:1;2/12:0 (HexCer), lyso-phosphatidate 17:0 (LPA), lyso-phosphatidylcholine 12:0 (LPC), lyso-phosphatidylethanolamine 17:1 (LPE), lyso-phosphatidylglycerol 17:1 (LPG), lyso-phosphatidylinositol 17:1 (LPI), lyso-phosphatidylserine 17:1 (LPS), phosphatidate 17:0/17:0 (PA), phosphatidylcholine 17:0/17:0 (PC), phosphatidylethanolamine 17:0/17:0 (PE), phosphatidylglycerol 17:0/17:0 (PG), phosphatidylinositol 16:0/16:0 (PI), phosphatidylserine 17:0/17:0 (PS), cholesterol ester 20:0 (CE), sphingomyelin 18:1;2/12:0;0 (SM) and TAG 17:0/17:0/17:0. Chain composition is given as length:saturation;oxidation. After extraction, the organic phase was transferred to an infusion plate and dried in a speed vacuum concentrator. The first-step dry extract was resuspended in 7.5 mM ammonium acetate in chloroform:methanol:propanol at 1:2:4 (v:v:v) and the second-step dry extract in 33% ethanol solution of methylamine:chloroform:methanol at 0.003:5:1 (v:v:v). All liquid-handling steps were performed using the Hamilton Robotics STARlet robotic platform with the anti-droplet control feature for organic solvents pipetting.

MS data acquisition

Samples were analysed by direct infusion on a QExactive mass spectrometer (Thermo Scientific) equipped with a TriVersa NanoMate ion source (Advion Biosciences). Samples were analysed in both positive and negative ion modes with a mass-to-charge ratio resolution (Rm/z) of Rm/z = 200 = 280,000 for MS and Rm/z = 200 = 17,500 for tandem MS (MS/MS) experiments, in a single acquisition. MS/MS was triggered by an inclusion list encompassing corresponding MS mass ranges scanned in 1-Da increments40. Both MS and MS/MS data were combined to monitor CE, DAG and TAG ions as ammonium adducts; PC, PC O- as acetate adducts; and CL, PA, PE, PE O-, PG, PI and PS as deprotonated anions. MS alone was used to monitor LPA, LPE, LPE O-, LPI and LPS as deprotonated anions; Cer, HexCer, SM, LPC and LPC O- as acetate adducts.

Data analysis and postprocessing

Data were analysed by Lipotype with in-house lipid-identification software based on LipidXplorer41,42. Data postprocessing and normalization were performed using an in-house data management system. Only lipid identifications with a signal-to-noise ratio greater than 5 and a signal intensity 5-fold higher than in corresponding blank samples were considered for further data analysis.

Homology modelling and data analysis

The DIESL AlphaFold22 model was accessed via the EMBL–EBI portal ( and visualized using PyMOL. Data wrangling, statistical analyses and plot generation were done using Prism (GraphPad Software) and RStudio (

Statistics and reproducibility

The number of replicates is indicated in the figure legends. All attempts at replication were successful.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.