The Inner Nuclear Membrane Is a Metabolically Active Territory that Generates Nuclear Lipid Droplets

1The endoplasmic reticulum (ER) synthesizes phospholipids for membrane growth and cell proliferation, and triacylglycerol (TAG) to store energy. TAG and phospholipids have common precursors, and the flow of lipids into storage or growth is a key cellular decision that is balanced with nutrient availability (Figure 1A). Phosphatidic acid (PA) is a lipid precursor at the branch point of the lipid storage and growth pathways (Figure 1A). PA is either converted into cytidyldiphosphate diacylglycerol (CDP-DAG) by yeast Cds1 (human CDS1) to promote new phospholipid synthesis or is dephosphorylated by Pah1 (human Lipin) to produce diacylglycerol (DAG). DAG, a signaling lipid, is phosphorylated by the Dgk1 kinase to regenerate PA, or is metabolized further into TAG. TAGs and steryl esters are stored in lipid droplets, which are synthesized on the ER and outer nuclear membrane (ONM). Although lipid droplets have sporadically been observed inside the nucleus of some mammalian cells, lipid droplet synthesis is considered to be the task of the ER/ONM (Figure 1B).

2The lipid composition of cellular organelles is customized to their functions, and the lipid composition of membranes can vary, including lipid asymmetry across the bilayer as well as in the lateral dimension. The ER, the cell’s main lipid factory, is contiguous with the ONM, which is joined with the inner nuclear membrane (INM) at membrane openings occupied by nuclear pore complexes (NPCs). Given that lipids can laterally diffuse within membranes (Figure 1B), it is generally assumed that the INM passively receives lipids from the ONM/ER and is itself metabolically silent.

3The INM is enriched in proteins that are required for genome regulation, inheritance, and protection, including integral and peripheral membrane proteins. Recent evidence suggests that specific lipids interact with the transmembrane domain of particular proteins. Hence, the specific INM proteome might reflect a functionally tailored lipid environment. It is therefore important to know what the INM lipids are and where they come from.

4We have developed tools to directly examine the INM lipid composition in living cells and combine a lipid biosensor strategy with specific perturbations of lipid metabolism. Our study argues that the INM is a metabolically active territory, whose chemical reactions could influence numerous aspects of genome function.

5The INM and ONM are only 10–50 nm apart and biochemical fractionation fails to separate them into pure fractions that would be suitable for lipidomics. In order to explore the lipid composition of the INM, we designed a set of genetically encoded, fluorescently labeled biosensors that visualize PA and DAG in living yeast cells. By appending or omitting a nuclear localization signal (NLS) to the sensors, we can assess PA and DAG levels in the nuclear as well as the cytoplasmic membrane compartment. The Q2 domain of the S. cerevisiae transcription factor Opi1 specifically recognizes high PA levels at the plasma membrane with a consistent pattern across a cell population (Figure 1C) confirming earlier reports. When increasing the sensor concentration about 10-fold, the fluorescence intensity at the plasma membrane increases correspondingly, but no other membrane compartments become stained (Figures S1A and S1B). In contrast to this cytoplasmic sensor, an NLS version of the PA sensor showed a diffuse intranuclear signal (Figure 1C; see Figures S1C for sensor specificity, S1D for expression levels, and S1E and S1F for the import mechanism). Consistent results were obtained by using the PA-sensing domain of the S. cerevisiae Spo20 protein (Figures S2A and S2B). These data suggest that PA is present at lower levels at the INM and ONM/ER compared to the PA-rich plasma membrane under the conditions tested. To detect the downstream lipid DAG, we used the DAG-specific recognition domains of R. norvegicus protein kinase C (PKCβ C1a+C1b). We detected DAG predominantly at the vacuolar membrane, but not at the ONM and ER (Figure 1D; see also Figures S2C for sensor specificity and S1D for expression levels). This specific distribution was retained when we overexpressed the sensor (Figures S2D and S2E). Both 10-fold and approximately 40-fold overexpression strongly increased the signal at the vacuole, yet little DAG signal was observed at the ONM/ER or the plasma membrane. This suggests a major difference in DAG levels between the vacuolar membrane and the ONM/ER/plasma membrane. To test whether the sensor can in principle detect DAG in membrane compartments other than the vacuole, we conditionally targeted Pah1 to the PA-rich plasma membrane in order to ectopically convert PA into DAG. Upon tethering a constitutively active variant of Pah1 (Pah1 7A) to the plasma membrane protein Pma1, the DAG sensor stained the plasma membrane in addition to the vacuole, with about equal intensity (Figure S2F). This indicates that the DAG sensor is able to detect newly synthesized DAG at an ectopic location, and that enrichment of the sensor on the vacuole does not prevent it from recognizing other DAG-containing membranes.

6Given the lack of DAG detection at the ONM/ER, it was striking to see that the NLS-bearing version of the DAG sensor targeted the INM (Figure 1D). The NLS-DAG sensor is expressed at similar levels as the non-NLS version and the sensor fluorescent intensities at the INM and vacuolar membrane are in a similar range (Figures 1D and S1D). This suggests that the difference in DAG detection between ONM/ER and INM is not a result of different sensor concentrations. Moreover, a version of the NLS-DAG sensor (Q63/Q128 > E) that is deficient in DAG binding distributes throughout the nucleoplasm (Figure S2C). Hence, the nuclear rim localization of the sensor is most likely a consequence of DAG recognition at the INM rather than exclusion by chromatin or unspecific interactions with INM proteins. Finally, DAG recognition per se does not appear to be influenced by appending an NLS to the sensor, since the NLS-DAG sensor was bound to the vacuole when its nuclear import was blocked (Figure S1F). In summary, our findings indicate a potential enrichment of DAG at the INM.

7To validate the INM localization of the DAG sensor, we performed bimolecular fluorescence complementation (BiFC) with lipid sensors fused to the C-terminal half of the Venus fluorophore (VC) and two nuclear proteins, Pus1 and Nup60, fused to the N-terminal half of the fluorophore (VN) (Figure 1E). Pus1 is a pseudouridine synthase in the nucleoplasm; Nup60 is a basket nucleoporin, exclusively localized on the nuclear face of the NPC. The NLS version of the DAG sensor interacted with the nucleoplasmic Pus1-VN specifically at the nuclear rim consistent with the presence of DAG at the INM (Figure 1F). Also, Nup60 interacted with the NLS-DAG sensor confirming its INM localization. In contrast, the DAG sensor without NLS failed to interact with Pus1-VN and Nup60-VN. The NLS-PA sensor showed fluorescent complementation with Pus1-VN in the nucleoplasm and Nup60-VN at the nuclear rim consistent with its nucleoplasmic localization in wild-type cells (Figure 1F). A faint BiFC signal of the PA sensor without exogenous NLS in Nup60-VN-expressing cells likely stems from an intrinsic NLS that is present in the Q2 domain of the PA sensor and leads to weak nuclear import. The PA and DAG sensors did not interfere with cell growth under the conditions tested, making it unlikely that they cause major alterations of lipid metabolism (Figure S1G).

8If DAG is indeed enriched at the INM, we should be able to convert this pool into PA by experimentally increasing Dgk1 activity at the INM (Figure S3A). Dgk1 is an ER-resident transmembrane protein and is expected to follow a different nuclear import route than soluble cargo. We therefore appended the NLS of the INM-resident transmembrane protein Heh2 to Dgk1 and analyzed its localization. Overexpressed wild-type Dgk1 was detected both at the peripheral ER and at the nuclear envelope (NE) and induced NE proliferations consistent with previous reports (Figures S3B and S3C). In contrast, the NLS version of Dgk1 removed the enzyme from the peripheral ER while targeting the NE, suggesting its nuclear import (Figure S3B). Indeed, immunogold electron microscopy (EM) demonstrated that wild-type Dgk1 is present on both sides of the NE (i.e., INM and ONM), whereas NLS-Dgk1 is enriched mainly at the inner side of the NE (Figure S3D). Importantly, NLS-Dgk1 caused a release of the NLS-DAG sensor from the INM into the nucleoplasm with concomitant binding of the NLS-PA sensor to the INM, whereas a catalytically inactive mutant of NLS-Dgk1 failed to release the NLS-DAG sensor from the INM (Figure S3E). These results suggest that our lipid sensors can specifically detect a Dgk1-mediated interconversion of DAG into PA at the INM.

9Taken together, our data demonstrate that a DAG lipid sensor recognizes the INM differently from the ONM/ER, likely as the result of a different lipid composition (Figure 1G). This raises the important question whether the INM exhibits a unique lipid metabolic activity, which could be functionally important in cell physiology.

10To explore whether the INM is metabolically active, we measured its lipid composition upon reprogramming lipid metabolism toward lipid storage (Figure 2A). Cds1 inhibition is expected to channel PA into the Pah1-dependent branch of TAG synthesis and storage. To create this condition, we employed a temperature-sensitive mutant of Cds1 (cds1-ts; Figure S4A) and measured PA levels at the INM with the NLS-bearing PA sensor at the permissive (23°C) and restrictive (37°C) temperatures. Lipid storage was monitored by staining lipid droplet (LD) neutral lipids with the BODIPY 493/503 dye. When cds1-ts cells were grown at the permissive temperature (23°C), they showed low PA levels at the INM and small cytoplasmic lipid droplets (cLDs), similar to wild-type cells grown at 23°C or 37°C (Figure 2B). In contrast, cds1-ts cells at the restrictive temperature (37°C) showed large cLDs, consistent with a role of yeast and human Cds1 in influencing lipid droplet size. Strikingly, we detected some large BODIPY-positive structures, typically one or two per cell, that enriched the NLS-bearing PA sensor on their surface. Moreover, these structures localized to the nuclear interior as judged by the NPC marker Nup188-GFP (Figure 2B; see Figure S4B for specificity control). These features suggest that the observed structures are intranuclear and represent nuclear lipid droplets (nLDs) with a core of BODIPY-stainable neutral lipids and a PA-rich lipid layer on the surface. To understand how nLDs form, we followed their emergence after Cds1 inhibition (Figure 2C and Videos S1 and S2). PA levels at the INM began to rise after 2 hr at the restrictive temperature, PA started to accumulate at small BODIPY-positive foci at the INM after 3 hr, and prominent nLDs had formed after 4 hr, which absorbed almost the entire pool of nuclear PA on their surface. Quantification of nLD biogenesis in multiple cells confirms the redistribution of the NLS-PA sensor from the nucleoplasm to the INM and, finally, onto the nLD surface. nLDs were found in approximately one-third of cds1-ts cells at 37°C (Figure 2D) and exhibited a similar diameter as cLDs (Figure 2E). To study nLD ultrastructure, we employed transmission electron microscopy (TEM). After 4 hr of Cds1 inactivation at 37°C, we observed prominent nLDs, often close to the INM (Figure 2F), which likely correspond to the BODIPY-positive and NLS-PA-sensor-enriched structures seen by live-cell fluorescence microscopy. To exclude the possibility that nLD production merely reflects a mutant condition, we supplemented exponentially growing wild-type cells with 0.5% oleic acid, a monounsaturated fatty acid, which cells can utilize as their sole carbon source. This high-fat diet induced an increase in cellular lipid droplet size (Figure S4C). It also resulted in a PA increase at the INM in around 60% of cells and led to nLD formation in 10% of cells as detected by simultaneous analysis with the NLS-PA sensor and BODIPY (Figures 2G and 2H). Thus, nLD formation in cds1-ts mutant cells reflects the natural capacity of cells to synthesize nLDs besides the known pool of cLDs. These findings also suggest that the conserved Cds1 enzyme limits both PA abundance at the INM and nLD synthesis from the INM.

11Our results raised the question whether cells target specific lipid biosynthetic enzymes, such as Cds1, to the INM. To test this, we analyzed by BiFC whether key lipid enzymes, Cds1, Pah1, and Dgk1 (Figure 1A), localize inside the nucleus besides their known presence at the ER. Notably, BiFC detected all enzymes at the nucleoplasmic face of the NPC via Nup60-VN indicating an INM localization (Figure 3A). Pah1 additionally appeared in the nucleoplasm, as detected by Pus1-VN. The interaction pattern is consistent with Cds1 and Dgk1 being predicted transmembrane proteins, whereas Pah1 dynamically interacts with membranes via an amphipathic helix. The BiFC data on Dgk1 further confirm the immunogold localization of Dgk1 at both the ONM and INM (Figure S3D). Since Pah1 seemed to have a soluble pool in the nucleoplasm, we tested whether appending an exogenous nuclear export signal (Rna1 NES) would prevent its natural import into the nucleus. Indeed, the NES version of Pah1 caused a loss of BiFC signal, whereas a mutant NES restored BiFC with Pus1-VN (Figures S3F and S3G). To confirm these findings, we localized Pah1 and Cds1 by immunogold EM. Consistent with the BiFC data, Pah1 localized to the nucleoplasm with some gold labeling present at the NE (Figure 3B). Similar to Dgk1 (Figure S3D), Cds1 was detected on both sides of the NE but not in the nucleoplasm (Figures 3B and S3H). To further test whether nLDs could be regulated by characteristic lipid droplet enzymes involved in lipid storage metabolism, we tested whether Tgl5, Tgl4, Tgl1, and Yju3, a group of lipases involved in cLD turnover, co-localized with nLDs. Indeed, after cds1-ts inactivation, GFP-tagged versions of these enzymes localized to nLDs (Figures 3C and S4E). In sum, the localization of key lipid enzymes inside the nucleus indicates that the INM is a metabolically active lipid territory capable of generating nLDs, the end product of cellular lipid storage.

12PA and DAG are both lipids that can induce membrane curvature. This property is thought to be important for the budding of cLDs from the ER as well as the fusion of individual lipid droplets, processes that involve the bending of membranes. Since DAG is enriched at the INM (Figure 3D) and produced from PA, we sought to investigate whether DAG is also present on nLDs. Indeed, the nuclear NLS-DAG sensor enriched on nLDs upon Cds1 inhibition, while maintaining its localization at the INM (Figure 3D; specificity control in Figure S4D). To simultaneously monitor PA and DAG at the INM during nLD biogenesis, we co-expressed the NLS-DAG and NLS-PA sensors, tagged with different fluorophores in cds1-ts cells. At the permissive temperature, these sensors recapitulate the distribution of PA and DAG in wild-type cells (Figure 3E). At the restrictive temperature both sensors co-localized to ring-shaped structures that are characteristic of the nLD surface (Figure 3E). In sum, Cds1 has a major impact on the distribution of both PA and DAG at the INM and their co-enrichment on nLDs. This raises the question, by which mechanism these lipids are channeled to nLDs.

13The nLD-producing cds1-ts strain exhibited a drastic growth defect at the restrictive temperature (Figure S4A), which led us to search for mutants that allow a structural analysis of nLDs at standard growth temperature. We hypothesized that INO4 and INO2 deletion would phenocopy Cds1 inactivation because CDS1 expression is promoted by the Ino4/Ino2 transcriptional complex (Figure 4A). Indeed, ino4Δ and ino2Δ cells displayed enlarged cLDs, as well as prominent nLDs with the characteristic NLS-PA-sensor enrichment at 30°C (Figures 4B and S5A). nLDs were present in approximately 1/3 of ino4Δ cells (Figures 4C and S5D) and had a similar size as the enlarged cLDs (Figure 4D). We used this condition to further characterize the relationship between nLDs and the INM by TEM. Interestingly, nLDs were found to interact with the INM at multiple membrane bridges (Figure 4E). By TEM, we could directly visualize the two opposing leaflets of the lipid bilayer and found that the lipid monolayer surrounding the nLD is contiguous with one of the two leaflets of the INM (Figures 4F and 4G). This suggests that PA, DAG, and other lipids can move from the INM onto the monolayer surrounding the nLD and into its interior. We reconstructed the 3D architecture of the INM-nLD contact points using stacks of electron micrographs (Video S3). The INM-nLD contacts do not form a stalk-like structure with a circular base but rather consist of an extended membrane ridge (Figure 4H). This membrane geometry probably requires scaffold proteins to be formed. Seipin is a possible candidate since it is implicated in stabilizing cLD-ER contacts. A second feature of nLD-containing nuclei is the altered perinuclear space, with multiple evaginations of the INM, which measure about 70 nm at the membrane opening (Figures 4E, 4I, 4J, S5C, and S5D). INM evaginations are also abundant in ino2Δ cells (Figure S5B). Since lipid droplets are sometimes found in areas occupied by INM evaginations (Figures 4I and 4J), it is possible that these zones represent intermediate stages of nLD production. Multiple small nLDs that arise from INM evaginations may subsequently fuse with a larger nLD that remains tethered to the INM. In sum, we show that the INM directly forms nLDs and thereby contributes to cellular lipid storage.

14For mammalian cytoplasmic lipid droplets, Seipin is thought to mediate lipid droplet maturation at the ER and both PA and DAG were implicated in this process. Mutations in Seipin lead to a severe form of congenital lipodystrophy, characterized by a lack of adipose tissue. Seipin, a predicted transmembrane protein, is conserved from yeast (SEI1) to human and localizes to the ER. In yeast, SEI1 deletion results in clusters of irregularly sized lipid droplets. To test whether Seipin plays a role in nLD production, we first examined whether it is present at the INM. By performing BiFC analysis with Nup60-VN and Sei1-VC, we observed BiFC puncta (Figure 5A), which indicate the presence of Sei1 at the INM besides its known localization at the ER. Notably, after induction of nLDs by inactivation of Cds1, fluorescent puncta resulting from Sei1-VC interaction with Nup60-VN were found in the vicinity of nLDs (Figure 5B).

15To directly examine a potential effect of Seipin on nLD formation, we deleted SEI1 in ino4Δ cells (Figure 5C). Notably, we did not observe defined membrane bridges upon SEI1 deletion. Instead, the nLDs were found to adhere closely to the INM (Figures 5C, S6A, and S6C). Whereas ino4Δ cells exhibited regularly sized evaginations in the periplasmic nuclear space (Figure 4I), sei1Δ ino4Δ cells displayed irregular periplasmic cavities (Figures 5D and S6B). In sum, we demonstrate that yeast Seipin, a key lipid droplet biogenesis factor, is found at the INM, where it affects the formation of membrane bridges with nLDs and the architecture of the periplasmic space during nLD production.

16The finding of lipid storage initiating from the INM is striking. Since PA can be channeled into either lipid storage or lipid biosynthesis (Figure 1A), we asked whether INM lipids also change during membrane growth. We deleted Pah1, which increases cellular PA and induces excessive NE proliferation likely because PA is redirected toward phospholipid synthesis (Figure 1A). In contrast to wild-type cells, the NLS-bearing PA sensor relocalized from the nuclear interior to the INM in pah1Δ cells, whereas the PA sensor without exogenous NLS remained at the plasma membrane and did not significantly label the ONM and ER (Figure S7A; protein levels in Figure S7C). This suggests that the INM can be altered in its PA content in response to changes in membrane growth. The typical DAG staining of the vacuolar membrane was decreased in pah1Δ cells (Figure S7B), consistent with reduced Pah1-dependent DAG production. Using the NLS-bearing DAG sensor, we could still detect DAG at the INM in pah1Δ cells (Figure S7B), consistent with lipidomic studies, which suggested residual DAG synthesis by other enzymatic pathways in pah1Δ cells. In sum, we demonstrate an increase of PA at the INM in a situation where membrane proliferation is stimulated (Figure S7D). Together with the localization of key lipid enzymes at the INM (Figures 3A and 3B), these findings support the notion that the INM is a metabolically active lipid territory, which is remodeled in its PA content during both membrane growth and lipid storage.

17The metabolic switch between membrane proliferation and lipid storage is controlled by Opi1, the master regulator of yeast phospholipid metabolism. The production of large nLDs in cds1-ts and ino4Δ cells coincided with the production of similarly sized cLDs (Figures 2E and 4D). cLDs are produced from the ONM and peripheral ER (Figure 6A), and we show that nLDs are directly produced from the INM. These findings suggest a coordinated mechanism for lipid droplet synthesis and size control in the nuclear and cytoplasmic compartment. Opi1 is an ER-associated transcription factor that senses changes in cellular PA concentration, enabling feedback regulation of lipid metabolism. Opi1 binds PA via its Q2 domain (i.e., the domain that we used as a lipid biosensor) and at least one additional binding site. Opi1 also interacts with the ER protein Scs2 via its two phenylalanines (F) in an acidic tract motif (FFAT) domain (Figure 7A). Upon PA reduction, Opi1 translocates from the ER into the nucleus, where it represses the Ino2/4 transcriptional activator complex and hence induces expression of phospholipid metabolic genes such as CDS1 (Figure 4A). Thus, changes in PA concentration are predicted to affect the location and repressor activity of Opi1. To address whether and how Opi1 might influence or be influenced by cytoplasmic and nuclear LD production, we first determined the localization of mCherry-tagged full-length Opi1 in cds1-ts cells grown at permissive (23°C) or restrictive (37°C) temperature (Figures 6B, S7E, and S7F). Opi1 localized mostly to the NE at 23°C, consistent with previous data. Strikingly, Opi1 relocalized to the surface of both cLDs and nLDs upon CDS1 inactivation (Figure 6B), and in wild-type cells grown with oleic acid (Figure 6C). The binding of Opi1 to nLDs and cLDs reflects their high PA content, as visualized by the PA sensors with and without an NLS (Figures 2B and 6D). Unlike PA, the DAG content seems higher on nLDs than cLDs under the conditions tested (Figures 3D and 6E).

18These findings raised the question whether PA-mediated sequestration of Opi1 on lipid droplets influences Opi1 transcriptional activity. To test this, we altered Opi1 levels and localization and measured the effect on cellular lipid droplet size. We partially untethered Opi1 from the ER by deleting SCS2, such that Opi1 localization mainly depends on its ability to bind PA via its Q2 domain (Figure 7A). Accordingly, Opi1 expressed from its endogenous promoter now localized mainly to the nucleoplasm due to an endogenous NLS (Figure 7B). Interestingly, a mutant in the Opi1 Q2 domain (Q2mut), which decreases affinity toward PA, induced nLD production and caused an increase in cellular lipid droplet size (Figures 7B and 7C). We attribute this effect to increased repressor activity of Opi1 at Ino4/Ino2-occupied promoters, since the mutant Opi1 is no longer sequestered by PA-containing membranes such as the lipid droplet surface. We predicted that lipid droplet size should grow even further when the intranuclear dosage of Opi1 is increased. To this end, we expressed Opi1 variants from a strong heterologous promoter (GPD) in scs2Δ cells (Figure 7B; protein levels in Figure S7G). Notably, these cells exhibited large lipid droplets and the nLDs were coated by Opi1. We then assessed the Opi1 Q2 mutant expressed from the GPD promoter. This mutant produced even larger lipid droplets (Figure 7C) and Opi1 became more diffuse in the nucleoplasm (Figure 7B), likely because PA on the nLD surface is now poorly recognized. Ablating the Opi1 AID domain, which is required for its repressor activity, abolished nLD production in scs2Δ cells and lowered global lipid droplet size to a comparable level as the deletion of OPI1 (Figures 7B and 7C). As a further functional readout, we measured growth of the Opi1 mutants on media, which lack inositol. Inositol auxotrophy is an indicator of Opi1 function. Indeed, inositol auxotrophy was most pronounced in cells with high Opi1 repressor activity (Figure 7D) in line with the effects on lipid droplet size. Taken together, we show that Opi1 regulates both nLD and cLD formation. We propose that Opi1 absorption onto PA-rich nLDs and cLDs may reduce its intranuclear repressive activity, providing feedback control on lipid storage at the INM.

19We have discovered active lipid metabolism at the INM and demonstrate a capacity of the INM for lipid storage via nuclear lipid droplet formation. This opens avenues for investigating the function of nuclear energy storage, as well as links between INM lipid metabolism and gene regulation.

20The entry point for the identification of a distinct lipid metabolism of the INM was our finding that a DAG-specific sensor bound to the INM, but not the ONM/ER (Figure 1D). We cannot formally rule out that the sensor’s affinity is influenced by the nucleoplasmic or cytoplasmic environment. However, we were unable to find evidence that the DAG sensor could not bind the ONM/ER due to its limiting concentration (Figure S2D), or that DAG sensor binding to the INM is due to exclusion from chromatin or unspecific interactions with INM proteins (Figure S2C). Hence, a DAG asymmetry across the NE is a likely scenario. We hypothesized that the differential recognition of DAG across the NE may reflect a physiologically relevant lipid metabolic activity at the INM. Indeed, we found that the INM can be tuned for lipid storage by environmental and genetic perturbations (Figures 2G and 4B). An emerging question now is how cells establish an asymmetric lipid distribution across the NE despite the continuity of the ONM with the INM. Based on our finding of multiple lipid metabolizing enzymes at the INM, we propose that different metabolic activities between the ER/ONM and INM can create distinct lipid compositions. Hence, asymmetry may arise because the INM is topologically insulated from the ER/ONM by NPCs, which regulate which enzymes reach the INM. Moreover, all lipid traffic between ONM and INM must pass through the pore membrane beneath NPCs. NPCs may influence lipid traffic by hosting special enzymes or other “lipid gating” mechanisms that avoid mixing of the INM and ONM lipid environments.

21Our study raises questions about the potential functions of PA and DAG at the INM. Why would the INM of exponentially growing yeast cells exhibit a low PA to DAG ratio? We suggest that PA is kept low at the INM by high turnover, which is mediated by the activities of Pah1 and Cds1 (Figure 7E). High PA levels at the INM either correlate with strong NE proliferation or with strong nLD formation, depending on which branch of lipid metabolism is active. Thus, a low PA level may reflect a dynamic steady state that can be tuned toward proliferation or storage.

22Does the apparent high abundance of DAG at the INM only reflect the turnover of PA? Probably not; DAG has special physicochemical properties, which could play multiple physiologically important roles at the INM. DAG is a conically shaped lipid, and this creates lipid packing defects in bilayers, which could favor the binding of specific INM proteins. DAG also facilitates the fusion of lipid bilayers, for example, during vesicle budding. Interphase NPC biogenesis requires a fusion between the ONM and INM, which is not understood. It will be interesting to examine whether DAG at the INM influences membrane remodeling and NPC assembly. Moreover, DAG becomes enriched on nLDs, suggesting that the INM reservoir of DAG could play a role in lipid droplet biogenesis in conjunction with PA. The conically shaped DAG may not only act as a surfactant, but also as fusogenic lipid during lipid droplet coalescence and as an inducer of negative curvature, which is present at nLD-INM membrane bridges. Finally, DAG is an important second messenger in cell signaling. The DAG-dependent kinase PKCβ, from which our DAG sensor is derived, phosphorylates histone H3 to regulate gene expression. Whether DAG-dependent PKCβ signaling occurs inside the nucleus remains to be determined. Given that PA has similar effects on membrane structure to DAG, investigating DAG and PA chemistry at the INM can illuminate new biological roles of these versatile lipid precursors and their potential roles in nuclear lipid signaling.

23We identified enzymes of both branches of lipid metabolism at the INM. With regards to membrane proliferation, this could imply that NE growth during the cell cycle relies on local lipid synthesis and turnover at the INM in addition to lipid delivery from the ER/ONM. The physiologic importance of lipid storage at the INM is underscored by the fact that a surplus of fatty acids induces nLD formation (Figure 2G). nLD production appears to be triggered by mechanisms that increase PA levels at the INM and draw PA into the storage branch (Figure 7E). nLD and cLD sizes seem to be co-regulated and are governed by the amount of Opi1 inside the nucleus, where Opi1 represses the promoters of genes involved in phospholipid synthesis (Figure 7E). Hypothetically, Opi1 could turn on lipid droplet production by inhibiting CDS1 expression, thereby shunting PA into TAG synthesis. Why then is Opi1 sequestered on both cLDs and nLDs via its PA-sensing domain? This association seems to restrict access of Opi1 to target genes and thereby restrain Opi1’s ability to induce lipid droplet formation. We propose that nLDs act as signaling platforms for transcription factors in spatial proximity to target promoters. By forming a classical negative feedback loop, this could buffer short-term fluctuations of nutrients and promote lipid homeostasis.

24What is the role of nLDs beyond their involvement in Opi1-dependent gene expression? The presence of nLD-INM bridges strongly suggests that nLDs communicate with the INM. Besides receiving lipids, nLDs may return lipids to the INM in times of increased demand, for example, during resumption of the cell cycle after stationary phase. cLDs were implicated in the neutralization of cytotoxic free fatty acids as well as the storage of misfolded hydrophobic proteins and histones. It will be interesting to examine, whether nLDs act as a storage site for nuclear proteins and, more generally, as a signaling platform for proteins with phospholipid-binding domains. Based on our work, lipid droplet function can now be comprehensively addressed in the context of nuclear biology taking advantage of the experimental versatility of S. cerevisiae. We have shown that yeast Seipin is required for the formation of proper membrane bridges between the INM and nLDs. Of note, human Seipin is mutated in Berardinelli-Seip congenital lipodystrophy type 2, which is characterized by a severe loss of body fat, ectopic fat deposition, and a deranged overall metabolism. Additionally, mutations in Lipin, the human ortholog of yeast Pah1, cause lipodystrophy in mice. The presence of Sei1 and Pah1 at the yeast INM raises the question whether their INM function is conserved between yeast and humans and, if so, which role the INM plays in the pathogenesis of human metabolic diseases.

25In sum, we suggest that the INM is not merely a remote province of the ER, spatially isolated and dependent on lipid supplies from the ER. Instead, we identify the INM as a territory with its own lipid metabolism and striking metabolic adaptability. This conceptual framework is testable in more diverse genetic perturbations, additional biological settings, and across evolution.

26Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Alwin Köhler (alwin.koehler@mfpl.ac.at).

27All yeast strains used in this study are listed in the Key Resources Table. Genes in yeast were tagged/deleted by a standard one-step PCR-based technique. Microbiological techniques followed standard procedures. Cells were grown in standard yeast extract peptone dextrose (YPD) or when transformed with plasmids in selective synthetic dextrose complete (SDC) drop-out media. Oleic acid-containing growth media were prepared from standard SDC media containing 0.1% glucose, 0.5% oleic acid, 1% Tween 80. Oleic acid was omitted in the negative control. To induce protein production from the GAL1 promoter, cells were grown exponentially in media containing 2% raffinose before adding galactose to a final concentration of 2% and further growth for 4 hr (for Dgk1, Figures S3B–S3E) or 3.5 hr (for DAG sensor, Figure S2D). For rapamycin-dependent targeting of Pah1 to the plasma membrane, cells were grown in synthetic media supplemented with adenine (0.2 mg/mL). Microscopy was performed 1 hr after rapamycin (1 μg/mL) addition.

28For the Opi1 Q2 PA sensor, a fragment comprising Opi1 aa103-189 was amplified from yeast genomic DNA. This fragment corresponds to the reported PA binding domain and contains an endogenous NLS sequence (aa109-112; KRQK). To create a PA-binding deficient mutant of Opi1 Q2, the following mutations were introduced based on a previous study: L124R, Y127A, L129R, M131A, I133R, K136A, K137A, R138A. For the PKCβ C1a+C1b DAG sensor, a fragment comprising aa31-158 was amplified from cDNA. DAG recognition is predicted to depend on specific residues, which were simultaneously mutated to abolish DAG sensing (Q63E in the C1a domain and Q128E in the C1b domain). All sensors are constructed in a modular fashion allowing exchange of promoters and fluorophores, NLS-sequences corresponding to Nup60 aa1-24 were introduced by PCR.

29Exponentially growing cells were immobilized on microscope slides with agarose pads and imaged on a DeltaVision Elite microscope (GE Healthcare) in a temperature-controlled chamber. Images were acquired with a 60x oil immersion objective and recorded with a CoolSNAP HQ2 CCD camera (Photometrics). Deconvolution was carried out using softWoRx software (GE Healthcare). Images were processed with ImageJ. Cell contours were marked with a dashed white line based on brightfield imaging. To stain lipid droplets or vacuoles, BODIPY 493/503 (final concentration 5.7 μM) or CellTracker Blue (final concentration 10 μM) was added, respectively, and cells were imaged after 20 min. For time-lapse imaging, yeast cells were immobilized with Concanavalin A (Sigma-Aldrich) on μ-Slide (8-Well Glass Bottom, Ibidi).

30Cells were grown exponentially, harvested and resuspended to a final OD600 of 0.5. 10-fold serial dilutions were prepared, spotted on appropriate plates and incubated at 23, 30 or 37°C.

31Cells were fixed in 2% paraformaldehyde and 0.2% glutaraldehyde (both from EMS) in 0.1 M Sörensen phosphate buffer (pH 7.4) for 2 hr at room temperature (RT), then overnight at 4°C. The cell pellet was embedded in 12% gelatine and cut into 1 mm3 blocks, which were immersed in 2.3 M sucrose overnight at 4°C. These blocks were mounted onto a Leica specimen carrier (Leica Microsystems) and frozen in liquid nitrogen. Frozen blocks were cut into ultra-thin sections at a nominal thickness of 60 nm at −120°C with a Leica UCT/FCS cryo-ultramicrotome (Leica Microsystems). A mixture of 2% methylcellulose and 2.3 M sucrose (1:1 ratio) was used as a pick-up solution. Sections were picked up onto 200 mesh Ni grids (Gilder Grids) with a carbon coated formvar film (Tokuyasu method). For immunolabeling, grids were placed on plates with solidified 2% gelatine and warmed up to 37°C for 20 min to remove the pick-up solution. After quenching of free aldehyde-groups with glycin (0.1% for 15 min), a blocking step with 1% BSA (fraction V) in 0.1 M Sörensen phosphate buffer (pH 7.4) was performed for 30 min. The grids were incubated in primary antibody, rabbit anti-GFP (Abcam), diluted 1:250 in 0.1 M Sörensen phosphate buffer containing 0.1% BSA over night at 4°C. Subsequently, a goat anti-rabbit secondary antibody coupled with 6 nm gold (GAR 6 nm, Aurion), diluted 1:20 in 0.1 M Sörensen phosphate buffer containing 0.1% BSA, was applied for 2 hr at RT. The sections were stained with 4% uranyl acetate (Merck) and 2% methylcellulose in a ratio of 1:9 on ice. All labeling steps were performed in a wet chamber. The sections were inspected using a FEI Morgagni 268D TEM (FEI) operated at 80 kV. Electron micrographs were acquired using an 11 megapixel Morada CCD camera from Olympus-SIS.

32ino4Δ and ino2Δ cells were grown at 30°C, whereas the cds1-ts strain was grown at 37°C for 4 hr in synthetic media prior to TEM analysis. Pelleted cells were mixed 1:1 with 10% BSA, used as a filler, and transferred into the 100 μm cavity of a 3 mm aluminum specimen carrier. This carrier was sandwiched with a flat 3 mm aluminum carrier and immediately high pressure frozen in a HPF Compact 01 (both carriers and high pressure freezer from Engineering Office M. Wohlwend GmbH). The frozen samples were subsequently transferred into a Leica EM AFS-2 freeze substitution unit (Leica Microsystems). Over a period of 4 days, samples were substituted in a medium of acetone containing 2% osmium tetroxide (Agar Scientific), 0.2% uranyl acetate and 2%–5% water. Freeze substitution was performed according to the following protocol: 30-40 hr at –90°C, warm up at a rate of 2°C per hour to –54°C, 8 hr at –54°C, warm up at a rate of 5°C per hour to –24, 15 hr at –24°C, warm up at a rate of 5°C per hour to 0°C, 2 hr at 0°C. At 0°C samples were taken out and washed 3 times in anhydrous acetone (on ice) and infiltrated with Agar 100 Epoxy resin (Agar Scientific), in a graded series of acetone and resin over a period of 3 days. Polymerization took place at 60°C. Ultra-thin sections with a nominal thickness of 70 nm were cut using a Leica UCT ultramicrotome (Leica Microsystems) and post-stained with 2% aqueous uranyl acetate and Reynold’s lead citrate. Examination regions on the sections were randomly selected and inspected with a FEI Morgagni 268D (FEI) operated at 80 kV. Digital images were acquired using an 11 megapixel Morada CCD camera (Olympus-SIS).

33For room temperature electron tomography, ino4Δ cells were prepared as described in the previous section, except that 150-250 nm sections were cut on a Leica UCT ultramicrotome (Leica Microsystems). After collecting the sections on a 50mesh Cu/Pd grid (Gilder Grids), previously coated with a supporting film of formvar, the section was covered on both sides with 10 nm gold (Aurion) by incubating the grid in a drop of concentrated gold solution for 3 min. Tilt series were acquired at a Tecnai G2 20 microscope (FEI) equipped with an Eagle 4k HS CCD camera (FEI) and operated at 200 kV. Dual axis tilt series were collected with a tilting range from −60° to +60° each at 1° increment. For data acquisition, processing and modeling the IMOD software from the Boulder Laboratory for 3D Electron Microscopy of Cells was used.

34Yeast whole-cell lysates were prepared, normalized for protein concentration and analyzed by western blotting according to standard procedures. Antibodies were used according to the manufacturer’s instructions.

35To manually quantify the percentage of cells with nLDs, the following criteria were used for nLD classification: the NLS-PA sensor co-localizes with a BODIPY-stained structure and this structure exhibits a round shape (for Figure 2D at least 200 cells for each condition and for Figure 4C wild-type n = 150, ino4Δ n = 400). The LD diameter was then measured using ImageJ (Figures 2E and 4D).

36To manually quantify the NLS-PA sensor localization in cells grown with oleic acid (Figure 2H), the following criteria were used: nLDs were classified according to the definition above. The sensor was classified as exhibiting an INM localization, whenever cells had no nLD, but exhibited a higher fluorescent intensity at the NE as compared to the nucleoplasm. The peak fluorescent intensity of the NE was measured by drawing a line across the NE using ImageJ. If the peak intensity at the NE was at least 1.5 times higher than the nucleoplasmic fluorescent intensity, the sensor localization was classified as “INM,” otherwise as “nucleoplasmic” (n = 195 cells treated with oleic acid and n = 140 control cells).

37Automated quantification of cellular LD area was performed with ImageJ (Figures 7C, S4C, S7E, and S7I). The lower fluorescent intensity threshold was chosen such that BODIPY-stained LDs appeared as single, non-clustered circular particles. Identical thresholds were applied to all experimental conditions. Particles with a circularity index of 0.7-1.0, were used to compare the area of LDs of different yeast cells (at least 390 cells were analyzed for each condition).

38To quantify sensor fluorescence intensity, a line was drawn across the organelle of interest and line-scan graphs were generated to measure the intensity value of the pixels (ImageJ software). Image acquisition parameters for images of the same experiment are always identical. Number of analyzed cells (n) is indicated in each figure.

39Gold particle quantification was performed by counting particles within a 125 nm zone relative to the NE midline (INM and ONM side) or in the nucleoplasm (NP) (> 125 nm from NE midline). For Figure 3B, 381 gold particles (Pah1) and 602 particles (Cds1) from 8 nuclei were counted. For Figure S3D, 760 gold particles (Dgk1) and 877 particles (NLS-Dgk1) from 9 nuclei were counted.

40Statistical significance was evaluated by Wilcoxon test or ANOVA analysis and Tukey HSD test using the GraphPad Prism and RStudio software. ∗ represents p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001.