Nuclear lipid droplets (LDs) in hepatocytes are derived from precursors of very-low-density lipoprotein in the ER lumen, but it is not known how cells lacking the lipoprotein secretory function form nuclear LDs. Here, we show that the inner nuclear membrane (INM) of U2OS cells harbors triglyceride synthesis enzymes, including ACSL3, AGPAT2, GPAT3/GPAT4, and DGAT1/DGAT2, and generates nuclear LDs in situ. mTOR inhibition increases nuclear LDs by inducing the nuclear translocation of lipin-1 phosphatidic acid (PA) phosphatase. Seipin, a protein essential for normal cytoplasmic LD formation in the ER, is absent in the INM. Knockdown of seipin increases nuclear LDs and PA in the nucleus, whereas seipin overexpression decreases these. Seipin knockdown also up-regulates lipin-1β expression, and lipin-1 knockdown decreases the effect of seipin knockdown on nuclear LDs without affecting PA redistribution. These results indicate that seipin is not directly involved in nuclear LD formation but instead restrains it by affecting lipin-1 expression and intracellular PA distribution.
Lipid droplets (LDs) are cytoplasmic organelles that are made of lipid esters covered by a phospholipid monolayer. The process of LD formation is thought to begin with the deposition of lipid esters between the two leaflets of the ER membrane, which eventually bulges toward the cytoplasm as a globule (Gao et al., 2019; Henne et al., 2019; Olzmann and Carvalho, 2019; Walther et al., 2017). The deposition and globule formation of lipid esters can be reconstituted in liposomes without proteins (Ben M’barek et al., 2017), but the process of LD biogenesis in vivo requires orchestration by regulatory proteins, including seipin (Renne et al., 2020).
Seipin is an evolutionally conserved integral membrane protein that distributes in the ER domain where nascent LDs form (Salo et al., 2016; Szymanski et al., 2007). Seipin deficiency causes aberrant cytoplasmic LDs in both mammalian cells and budding yeast (Fei et al., 2008; Salo et al., 2016; Szymanski et al., 2007; Wang et al., 2016), illustrating its importance in normal LD formation. Interestingly, yeast lacking the seipin orthologue Fld1 harbor nuclear LDs (Cartwright et al., 2015; Grippa et al., 2015), whereas nuclear LDs are scarce in normal yeast even in the post-diauxic phase when triglyceride (TG) is synthesized in the inner nuclear membrane (INM; Barbosa et al., 2019). These results suggest that nuclear LDs in seipin-deficient yeast may occur as a result of disrupted seipin function in the ER. This interpretation, however, contradicts that of another study on yeast, which reported that seipin exists in the INM and is involved in nuclear LD formation (Romanauska and Köhler, 2018). It thus remains unclear how seipin deficiency increases nuclear LDs in yeast. Whether and how seipin is involved in nuclear LD formation in mammalian cells also remains unknown.
Recently, we found that nuclear LDs in hepatocytes are derived from apolipoprotein B100 (ApoB)–free lipid particles in the ER lumen, a precursor of very-low-density lipoprotein (VLDL; Sołtysik et al., 2019b). Briefly, the ApoB-free lipid particles accumulate at the end of the type I nucleoplasmic reticulum (NR), an inward extension of the nuclear cistern (Malhas et al., 2011), intrude into the nucleoplasm through breaches in the INM, and become nuclear LDs. This unique mechanism explains why hepatocytes harbor abundant nuclear LDs, but other mechanisms are likely to exist, because cells that do not have lipoprotein secretory function also have nuclear LDs (Soltysik et al., 2019a).
In the present study, we addressed how nuclear LDs form in nonhepatocytes and whether and how seipin is related to this process. We found that the INM harbors TG synthesis enzymes and generates nuclear LDs in situ. Importantly, seipin is not present in the INM, and thus, seipin is not directly involved in nuclear LD formation. Rather, seipin depletion increases nuclear LDs by the transcriptional up-regulation of lipin-1β and an increase of phosphatidic acid (PA) in the nucleus. Moreover, seipin overexpression decreases nuclear LDs and nuclear PA. These results indicate that nuclear LDs in nonhepatocytes form in the INM through a seipin-independent mechanism and that seipin restricts nuclear LD formation through its effect on lipin-1 expression and intracellular PA distribution.
Results and discussion
Nuclear LDs form in the INM
U2OS, an osteosarcoma cell line, frequently harbors nuclear LDs (Fig. S1 A; Ohsaki et al., 2016). As in Huh7, a hepatocarcinoma cell line, the number of nuclear LDs in U2OS increased when cells were cultured with oleic acid (OA), and the increase was suppressed by triacsin C, a long-chain-fatty-acid–CoA ligase (ACSL) inhibitor (Fig. 1 A). Nuclear LDs in U2OS were also similar to those in Huh7 in that they are associated with the promyelocytic leukemia–nuclear body (PML-NB; Fig. S1 B; Ohsaki et al., 2016).
Nuclear LDs in U2OS, however, were different from those in Huh7 in several respects. First, ER stressor (tunicamycin) and microsomal TG transfer protein (MTP) inhibitor (BAY13-9952), which increases and decreases nuclear LDs in Huh7, respectively (Sołtysik et al., 2019b), did not affect nuclear LDs in U2OS (Fig. 1 A). Second, overexpression and knockdown of PML-II, which increases and decreases nuclear LDs in Huh7, respectively (Ohsaki et al., 2016), did not have such effects in U2OS (Fig. 1 B). Third, the type I NR develops in accordance with nuclear LDs in Huh7 (Sołtysik et al., 2019b), but this did not occur in U2OS (Fig. 1 C).
These differences suggest that nuclear LDs in Huh7 and those in U2OS form through different mechanisms. In Huh7, nuclear LDs derive from ApoB-free lipid particles that are synthesized in the ER lumen by an MTP-dependent process (Sołtysik et al., 2019b), but in U2OS, which lacks MTP (Fig. S1C), the origin of nuclear LDs should be different. Live imaging in which nuclear LDs were visualized by the specific marker NLSx3-HPos (Sołtysik et al., 2019b) showed that nuclear LDs in U2OS originate in the nuclear envelope and grow toward the nucleoplasm (Fig. 1, D and E; and Videos 1 and 2). This result indicates that nuclear LDs in U2OS are generated in the INM in situ. Additionally, the time course observed here suggests that the nuclear LD formation in the INM is a slower process than cytoplasmic LD formation in the ER (Kassan et al., 2013; Wang et al., 2016).
TG synthetic enzymes in the INM
The aforementioned findings suggest that TG synthesis enzymes distribute in the INM. To test this possibility, we performed the bimolecular fluorescence complementation (BiFC) using the GFP fragment GFP1–10 appended with a nuclear localization signal (NLS; Fig. S1 D). The complementary GFP fragment GFP11, fused to glycerol-3-phosphate acyltransferase 3 (GPAT3), GPAT4, 1-acyl-sn-glycerol-3-phosphate acyltransferase 2 (AGPAT2), AGPAT4, lysophosphatidylcholine acyltransferase 1 (LPCAT1, which has lysophosphatidic acid acyltransferase activity; Shindou et al., 2009), and diacylglycerol O–acyltransferase 1 (DGAT1), showed linear fluorescence in the nuclear periphery (Fig. 2 A). In contrast, when GFP1–10 without NLS was used as the BiFC partner, GFP11-tagged proteins generated fluorescence in a cytoplasmic network pattern, validating our methodology (Fig. S1 E).
For ACSL3, BiFC did not work, but GFP-ACSL3 distributed in nuclear LDs (Fig. 2 B) and was labeled in the INM by immuno-EM (Fig. S1 F). Endogenous ACSL3 was also immunolabeled in nuclear LDs (Fig. S1 G). GFP-tagged GPAT3, GPAT4, LPCAT1, and DGAT2, all of which are distributed in cytoplasmic LDs via a hairpin domain (Wilfling et al., 2013), were also found in nuclear LDs, indicating their presence in the INM (Fig. 2 B).
The inhibitory effect of triacsin C (Fig. 1 A) indicated that ACSL is necessary for nuclear LD formation. Knockdown of individual triacsin C–sensitive ACSL isoforms (i.e., ACSL1, ACSL3, and ACSL4; Grevengoed et al., 2014) indicated that ACSL3 is essential (Fig. 2 C). Double knockdown of GPAT3 and GPAT4 also decreased the number of nuclear LDs (Fig. 2 D). To examine whether TG synthesis in the INM is critical for nuclear LD formation, we compared GFP-DGAT2 and GFP-PK-DGAT2, the latter of which contains a bulky pyruvate kinase (PK) segment that precludes its entry into the INM (Soullam and Worman, 1995). In cells depleted of endogenous DGAT2, the expression of GFP-DGAT2 induced significantly more nuclear LDs than the expression of GFP-PK-DGAT2 did (Figs. 2 E and S1 H). These results indicated that nuclear LD formation involves TG synthesized in the INM.
We next examined whether nuclear LDs are connected to the INM. First, using photoconvertible Dendra2-LPCAT1, red fluorescence provoked by laser illumination in the cytoplasm appeared in nuclear LDs several minutes later (Fig. 2 F and Video 3). Second, GFP-DGAT2 photobleached in nuclear LDs recovered its fluorescence (Video 4). These results indicated that nuclear LDs and the INM are connected at some point. However, the appearance of photoconverted Dendra2-LPCAT1 and the recovery of GFP-DGAT2 occurred only rarely, suggesting that the nuclear LD–INM connection is transient and that most nuclear LDs are detached from the INM. Consistently, GFP-ACSL3 in nuclear LDs faded away and thereafter those LDs shrank, suggesting that GFP-ACSL3 is not continuously imported from the INM to nuclear LDs (Fig. 2 G and Video 5). The nuclear LD–INM relationship will be examined further in a later section.
Lipins and nuclear LDs
The penultimate step of TG synthesis, the conversion of PA to diacylglycerol, is catalyzed by lipins (Wang et al., 2017), which are soluble proteins that bind membranes through hydrogen bonds with PA (Zhang and Reue, 2017). Quantitative real-time PCR (qRT-PCR) showed that U2OS expresses lipin-1α, lipin-1β, lipin-2, and lipin-3 (Fig. 3 A). In adipocytes and hepatocytes, lipin-1β is confined to the cytoplasm (Bou Khalil et al., 2009; Péterfy et al., 2005), but lipin-1β in U2OS also distributed in the nucleus as in HEK293 cells (Ren et al., 2010; Fig. 3 B). The nuclear distribution of lipin-1 was increased by mTOR inhibition with Torin1 as reported previously (Peterson et al., 2011); this increase was particularly prominent for lipin-1β (Fig. 3 B). Lipin-2 and lipin-3 distributed only in the cytoplasm, even after Torin1 treatment (Fig. 3 B).
Nuclear LDs increased in cells treated with OA and Torin1 (Fig. 3 C), suggesting that the nuclear translocation of lipin-1 may facilitate nuclear LD formation. This idea was supported by the observation that GFP–lipin-1, particularly GFP–lipin-1β, showed transient concentration around nuclear LDs (Video 6). Accumulation of GFP–lipin-1β in nuclear LDs was observed more clearly after lipin-1 knockdown (Fig. 3 D), probably because hetero-oligomer formation between GFP–lipin-1β and endogenous lipin-1α was decreased (Liu et al., 2010).
To examine whether lipin-1 functionality is necessary for nuclear LD formation, lipin-1β mutants that were deficient in the PA phosphatase activity alone (D714E; Harris et al., 2007) or in both PA phosphatase and transcriptional coactivator activities (LxxFF; Finck et al., 2006) were expressed after knockdown of endogenous lipin-1. The expression of wild-type lipin-1β increased nuclear LDs as expected, whereas the expression of lipin-1β (D714E) and lipin-1β (LxxFF) did not (Fig. 3 E). Similar results were obtained for wild-type lipin-1α and lipin-1α mutants (Fig. S2 A). Additionally, lipin-1β tagged with a nuclear export signal (NES) showed a smaller effect on nuclear LDs than wild-type lipin-1β (Fig. S2 B). These results indicated that the PA phosphatase activity of lipin-1 in the nucleus is involved in nuclear LD formation.
Seipin and nuclear LDs
We next asked how seipin is related to nuclear LD formation. First, we found that seipin knockdown increased nuclear LDs (Figs. 4 A and S2 C), and the reintroduction of either nontagged seipin (Fig. 4 B) or GFP-seipin (Fig. S2 D) negated this effect. Knockdown of ACSL3 suppressed the increase of nuclear LDs by seipin deficiency (Fig. 4 A), as observed for OA-induced nuclear LDs (Fig. 2 C). EM confirmed that nuclear LDs in seipin-deficient cells are located in the nucleoplasm (Fig. 4 C).
Yeast seipin was inferred to exist in the INM based on a BiFC experiment using a basket nucleoporin, Nup60, as the complementation partner (Romanauska and Köhler, 2018). To study whether seipin exists in the INM in U2OS, we also used BiFC, but with NLSx3-RFP–tagged GFP1–10, a soluble nucleoplasmic protein, as the complementation partner. NLSx3-RFP-GFP1–10 complemented GFP11-GPAT4 and generated linear green fluorescence in the nuclear periphery as expected, whereas it never did with GFP11-seipin (Fig. 4 D). The functionality of GFP11-seipin was verified by complementation with RFP-GFP1–10, which generated green fluorescence in a cytoplasmic network pattern (Fig. 4 D). Furthermore, immuno-EM of GFP-seipin showed labeling in the ER and the outer nuclear membrane, but not in the INM (Fig. 4 E; see Fig. S1 F for comparison). These results led us to conclude that seipin does not exist in the INM in U2OS. What causes the difference between yeast and mammalian results is not clear and needs further investigation.
Next, we examined the influence of seipin knockdown on nuclear and cytoplasmic LDs using FRAP of GFP-ACSL3. In agreement with the previous observation using HPos (Salo et al., 2016), the majority of control cells exhibited fluorescence recovery of GFP-ACSL3 in cytoplasmic LDs, while in seipin-deficient cells, recovery occurred more slowly and in a smaller proportion of cytoplasmic LDs (Fig. 4 F). In contrast, in nuclear LDs, the fluorescence recovery of GFP-ACSL3 was not observed in control cells (Fig. 4 F and Video 7), consistent with the low frequency of GFP-DGAT2 FRAP (Video 4). Remarkably, after seipin knockdown, recovery occurred in more than half of nuclear LDs (Fig. 4 F and Video 8), indicating that the nuclear LDs connected to the INM were increased. Consistent with this, super-resolution microscopy of mRuby3-NLSx3-HPos showed that a nuclear LD–INM association was far more common in seipin-deficient cells than in control cells (Fig. 4 G). The association of nuclear LDs with the INM was also observed by EM (Fig. S2 E).
Overexpression of seipin also caused a reciprocal change in nuclear and cytoplasmic LDs, decreasing nuclear LDs while increasing small and constant-sized cytoplasmic LDs (Fig. S2, F and G). These results indicate that seipin affects nuclear and cytoplasmic LDs differently, often in opposite ways. Considering the absence of seipin in the INM, we inferred that seipin knockdown and overexpression affect nuclear LDs indirectly through changes of seipin functionalities in the ER.
Mechanism of nuclear LD increase in seipin deficiency
Western blotting revealed that lipin-1, but not lipin-2, increased as a result of seipin knockdown (Fig. 5 A). Lipin-1α and -1β could not be separated in Western blotting, but qRT-PCR showed that lipin-1β mRNA increased while lipin-1α mRNA did not change (Fig. 5 B), suggesting that lipin-1β protein was increased in seipin-deficient cells. It is notable that seipin knockdown also decreased splicing factor TRA2Β (SFRS10; Fig. 5 A). Reduction of TRA2Β was shown to favor the generation of lipin-1β mRNA in human obesity (Pihlajamäki et al., 2011; but see Brosch et al., 2012 for an alternative result) and thus may be the cause of lipin-1β up-regulation in seipin deficiency. The mechanism linking seipin and TRA2Β warrants further investigation in relation to abnormalities caused by seipin mutation.
To examine whether lipin-1 up-regulation is essential for the increase of nuclear LDs in seipin deficiency, knockdown of seipin alone and double knockdown of seipin and lipin-1 were compared. The additional knockdown of lipin-1 suppressed the increase of nuclear LDs by seipin knockdown (Figs. 5 C and S3 A), indicating that up-regulation of lipin-1 is at least partially responsible for the increase of nuclear LDs in seipin-deficient cells.
An important change in seipin deficiency occurs in the intracellular distribution of PA (Fei et al., 2011; Han et al., 2015; Wolinski et al., 2015). We hypothesized that seipin deficiency increases nuclear LD formation by increasing nuclear PA, thereby facilitating lipin-1 recruitment (Eaton et al., 2013; Ren et al., 2010). To test this idea, we used the fluorescent biosensor NLS-Opi1 Q2-GFP, which binds to nuclear PA (Romanauska and Köhler, 2018). Although the Opi1 Q2 domain may have a preference for some PA species (Kassas et al., 2017), the accumulation of NLS-Opi1 Q2-GFP in mCherry-GPAT4–positive nuclear LDs supports its functionality as a PA sensor (Fig. S3 B). NLS-Opi1 Q2-GFP distributed diffusely in untreated U2OS, but in U2OS cultured with OA alone or with OA and Torin1, puncta besides nuclear LDs and rings around them increased, probably reflecting an increase of PA synthesis (Menon et al., 2017; Fig. S3, C and D). In live imaging, NLS-Opi1 Q2-GFP showed blinking and punctum-to-ring changes, indicating fluctuation of nuclear PA (Fig. S3 E and Video 9). NLS-Opi1 Q2mut-GFP, mutated in the PA-binding domain, distributed only diffusely in the nucleus, verifying the specificity of the PA biosensor (Fig. S3, C and D).
Remarkably, the number of PA puncta/rings significantly increased in seipin-deficient cells (Fig. 5 D) and decreased in cells overexpressing seipin (Fig. S3 F). The increase of nuclear PA in seipin-deficient cells was confirmed by enzyme-coupled fluometric assay of PA in isolated nuclei (Figs. 5 E and S3 G; Morita et al., 2009). Accumulation of GFP–lipin-1β in nuclear LDs increased in seipin-deficient cells, irrespective of its PA phosphatase activity (Fig. 5 F), indicating that increased PA facilitated lipin-1 recruitment, although a PA-independent mechanism might be also involved. The nuclear PA signal in seipin-deficient cells was not affected by additional knockdown of lipin-1 (Fig. 5 G). The result indicated that the increase of nuclear PA occurs independently from the increase of lipin-1 expression.
GPAT3 and GPAT4, when overexpressed, were mostly distributed in the cytoplasm and induced large cytoplasmic LDs, as previously reported (Pagac et al., 2016), but they did not increase nuclear LDs or nuclear PA puncta (Fig. S3 H). This result suggests that, in the presence of seipin, PA that is produced by GPAT3/4 is used efficiently for LD biogenesis in the ER and scarcely increases PA in the INM. Diffusion of PA to the INM may also be restricted by binding to seipin (Yan et al., 2018). Moreover, under normal conditions, seipin down-regulates GPAT3 and GPAT4 (Pagac et al., 2016). When these effects are lost by seipin depletion, freely diffusible PA should increase in the ER and reach the INM. We speculate that PA that is increased in the INM in this manner recruits up-regulated lipin-1, thereby enhancing TG synthesis and nuclear LD formation (Fig. 5 H), although a mechanism unrelated to PA may be also involved. The increased PA in seipin-deficient cells may also increase the INM-connected nuclear LDs by making LD budding inefficient (Ben M’barek et al., 2017) and stabilizing the connecting membrane bridge (Choudhary et al., 2018).
Nuclear LDs in hepatocytes recruit choline-phosphate cytidylyltransferase A (CCTα) and activate phosphatidylcholine (PC) synthesis (Sołtysik et al., 2019b). Nuclear LDs in U2OS also recruit CCTα (Lee et al., 2020), but knockdown of perilipin-3, which up-regulates PC synthesis in Huh7 by increasing CCTα recruitment, did not have the same effect in U2OS (Fig. S3 I), suggesting that the impact of nuclear LDs on CCTα regulation may be smaller in U2OS. On the other hand, nuclear LDs may modulate PML-NB functions (Lee et al., 2020; Soltysik et al., 2019a). Interestingly, the nuclear translocation of lipin-1 induced by mTOR inhibition causes down-regulation of sterol regulatory element–binding proteins (Peterson et al., 2011), which may distribute in PML-NBs (Zoumi et al., 2005). Increased nuclear LDs may be involved in this regulatory process. The present study provides a solid basis to probe this and other fascinating possibilities by showing how nuclear LDs form and how they can be manipulated experimentally.
Materials and methods
Rabbit anti-human perilipin-3 antibody was raised against a peptide of human perilipin-3 segment (amino acids 305–318) and affinity purified by using a peptide column (Ohsaki et al., 2006). Rabbit anti-GFP and rabbit anti-PML antibodies were donated by Dr. Masahiko Watanabe (Hokkaido University, Sapporo, Japan) and Dr. Kun-San Chang (The University of Texas MD Anderson Cancer Center, Houston, TX), respectively. Mouse anti-ACSL3 (H00002181-B01P; Abnova), mouse anti-Na+,K+-ATPase β 1 (NB300-147SS; Novus Biologicals), rabbit anti-GPAT3 (AP18013c; ABGENT), rabbit anti-GPAT4 (NB100-2390SS; Novus Biologicals), rabbit anti-DGAT1 (NB110-41487SS; Novus Biologicals), rabbit anti-seipin (NBP1-79336; Novus Biologicals; 23846; Cell Signaling), rabbit anti–lipin-1 (HPA038021; Atlas Antibodies), rabbit anti–lipin-2 (A303-703A; Bethyl Lab), rabbit anti–lamin B receptor (GWB-C7CA28; Genway Biotech), rabbit anti-PML (M041-3; Medical and Biological Laboratories), rabbit anti-actin (A2066; Sigma), rabbit anti-TRA2B (GTX114752; GeneTex), rabbit anti–lamin A/C (GTX101127; GeneTex), goat anti-MTP (sc-33116; Santa Cruz Biotechnology), and goat anti-Sec61α (sc-12322; Santa Cruz Biotechnology) antibodies were obtained from the indicated suppliers. Secondary antibodies conjugated to fluorochromes (Invitrogen and Jackson ImmunoResearch Lab) and protein A conjugated with colloidal gold (Utrecht University Medical Center) were also purchased.
Cell lines and reagents
U2OS, Huh7, and A549 cells were kindly donated by Dr. Hidemasa Goto (Aichi Cancer Center, Nagoya, Japan), Dr. Eija Jokitalo (University of Helsinki, Helsinki, Finland), and Dr. Takashi Takahashi (Nagoya University, Nagoya, Japan), respectively. Cells were cultured in DMEM (U2OS and A549) or minimum essential medium (Huh7) supplemented with 10% FBS and antibiotics at 37°C in a humidified atmosphere of 95% air and 5% CO2. In some experiments, OA (Sigma) in complex with fatty acid–free BSA (Wako) at a molar ratio of 6:1 was added. BAY13-9952 (Implitapide; Bayer Healthcare), triacsin C (Santa Cruz Biotechnology), tunicamycin, and Torin1 (Sigma) were purchased from the indicated suppliers.
Plasmids and siRNA
pEGFP-C3-PML-II was donated by Dr. P. Hemmerich (Fritz-Lipmann Institute, Jena, Germany; Weidtkamp-Peters et al., 2008). The sequences encoding human GPAT3 (NCBI RefSeq accession no. NM_032717.5), human GPAT4 (NM_178819.4), human AGPAT2 (NM_006412.4), human AGPAT4 (NM_020133.3), human LPCAT1 (NM_024830.5), human ACSL3 (NM_004457.5), human seipin (NM_001122955.4), human lipin-2 (NM_014646.2), and human lipin-3 (NM_022896.3) were amplified using standard PCR procedures with cDNA from U2OS, Huh7, or HeLa cells and subcloned into pEGFP-C1, pEGFP-N3, pmCherry-C1, or pDendra2-C vectors (Clontech). Mouse DGAT1 (NM_010046.3) and mouse DGAT2 (NM_026384.3) were subcloned from the Flag mouse DGAT1 and Venus mouse DGAT2 vectors, which were donated by Dr. Ho Yi Mak (Hong Kong University of Science and Technology, Hong Kong; Xu et al., 2012). Human lipin-1α (NM_145693.4), lipin-1β (NM_001349204.2), and lipin-1γ (NM_001349200.2) were subcloned from pET-28b(+) plasmids that were donated by Dr. George Carman (Rutgers University, New Brunswick, NJ; Han and Carman, 2010). Plasmids encoding lipin-1 mutants, lipin-1α (D678E, LxxFF) and lipin-1β (D714E, LxxFF) were generated by PCR using primers with appropriate nucleotide substitutions. Split GFPs for BiFC, GFP1–10, and GFP11 were subcloned from pcDNA3.1-GFP(1–10) (#70219; Addgene) and pEGFP-GFP11-clathrin light chain (#70217; Addgene), which were gifts of Dr. Bo Huang (University of California, San Francisco, San Francisco, CA; Kamiyama et al., 2016). A chicken muscle pyruvate kinase segment lacking the first 16 amino acids was subcloned from pEGFP-C1-PK (#24163; Addgene), which was a gift of Dr. Warner Greene (Gladstone Institutes, San Francisco, CA), to construct a plasmid for expression of GFP-PK-DGAT2. For fluorescent PA biosensor NLS-tagged Opi1 Q2-GFP, Opi1 Q2 (amino acids 103–189 of Opi1; NM_001179100.1) was amplified by standard PCR procedures from Saccharomyces cerevisiae (SEY6210) cDNA and subcloned into pEGFP-N3; NLS-Opi1 Q2mut-GFP was generated by substituting eight amino acids with alanine (L124R, Y127A, L129R, M131A, I133R, K136A, K137A, and R138A) as previously described (Romanauska and Köhler, 2018). GFP-NLSx3-Hpos, a nuclear LD marker, was constructed by inserting three NLSs in tandem before Hpos (MDVLVPLLQLLVLLLTLPLHLLALGCWQPLFEAIGKIFSNIRISTQKEI; Kassan et al., 2013) as previously described (Sołtysik et al., 2019b). NLS (KRPAATKKAGQAKKKK) and NES (NELALKLAGLDINKT) were taken from Xenopus nucleoplasmin (GenBank accession no. X04766.1) and human cAMP-dependent protein kinase inhibitor α (NCBI RefSeq accession no. NM_006823.4), respectively.
siRNAs were synthesized by Japan Bio Service, Inc. Target nucleotide sequences of siRNAs are shown in Table S1.
Plasmids and siRNAs were transfected using Lipofectamine 2000 and Lipofectamine RNAiMAX (Thermo Fisher), respectively, according to the manufacturer’s instruction. The cells were analyzed 2 d (cDNA) and 3 d (siRNA) after transfection.
Immunofluorescence microscopy and data analysis
Cells were fixed with 3–4% formaldehyde in 0.1 M phosphate buffer for 15 min and permeabilized with either 0.01% digitonin in PBS for 30 min or 0.5% Triton X-100 in PBS for 10 min. Blocking and antibody dilution were performed with 1–3% BSA in PBS or 4% normal goat serum in PBS. Nuclei were labeled with Hoechst 33342 (Sigma), whereas LDs were labeled with either BODIPY493/503 or LipidTox Red (Thermo Fisher). Samples were mounted in Mowiol 4–88 containing 2.5% 1,4-diazabicyclo-[2,2,2]-octane.
Images were captured using a Zeiss Axio Imager M2 with Apotome2 using Plan-Neofluar 100×/1.30 and Plan-Apochromat 63×/1.40 lenses. For super-resolution microscopy, a Nikon Eclipse Ti with N-SIM module equipped with an Andor DU-897 camera, an Apo TIRF 100×/1.49 Oil DIC N2 lens, and a Nikon NIS-Elements 5.20.02 software was used. The color, brightness, and contrast of presented images were adjusted using Adobe Photoshop CS3 or Zeiss Zen. Quantification was performed as described previously (Ohsaki et al., 2016). The number of LDs was counted manually, and the area was measured using ImageJ. The total nuclear LD area and the total intranuclear lamin B1 receptor–positive area were divided by the total nuclear area to obtain the relative nuclear LD area and the relative NR area, respectively. For NR measurement, signals in the nuclear periphery were excluded.
Live confocal images (except for those shown in Video 2) were obtained using a Cell Voyager CV1000 spinning-disk confocal laser system (Olympus) equipped with 488-nm and 561-nm diode lasers. An electron multiplying charge-coupled device camera (1,000 × 1,000 pixels, ImagEM C9100-14 and 512 × 512 pixels, ImagEM C9100-13; Hamamatsu Photonics) and oil immersion objective lenses (UPLSAPO 100×/1.4 and UPLSAPO 60×/1.35; Olympus) were used. For Video 2, an Olympus SpinSR10 spinning-disk confocal super-resolution microscope equipped with a Hamamatsu ORCA-Flash 4.0 camera, a UPLAPO OHR 60×/1.50 lens, and an Olympus cellSens Dimension 2.3 software was used. Cells were kept at 37°C in a culture chamber filled with 95% air and 5% CO2. Cells were placed in DMEM and 10% FBS for most experiments, except for the experiment shown in Video 2, in which they were placed in FluoroBrite DMEM (Thermo Fisher) supplemented with 10% FBS and 4 mM L-glutamine. For each point, multiple z-stack sections with intervals <1.0 µm were obtained. Images were processed using the Fiji software application.
FRAP and photoconversion
U2OS cells expressing each of the respective fluorescent proteins were cultured with OA for 1 d and observed with a confocal laser-scanning microscope (TiE-A1R; Nikon) equipped with an A1-DUG GaAsP Multi Detector Unit and a Plan Apo ×100/1.45 oil immersion lens. For FRAP, one nuclear LD or up to three cytoplasmic LDs were bleached 10 times at 63-ms intervals using the 561-nm laser at 100% intensity. Images were captured for 20 min at 1-min intervals (nuclear LDs) or for 15 min at 30-s intervals (cytoplasmic LDs). For photoconversion, a cytoplasmic area was illuminated 10 times at 63-ms intervals using the 488-nm laser at 100% intensity. All experiments were performed at 37°C in a humidified atmosphere of 95% air and 5% CO2. Florescence recovery in regions of interest was quantified using ImageJ and normalized to fluorescence in the LDs before bleaching. Results are presented as the mean ± SEM, and curves fitting with nonlinear regression and the exponential one-phase decay model were applied (GraphPad Prism 8).
PC synthesis activity
Cells cultured with 1 µCi/ml 3H-choline (PerkinElmer) for 30 min were incubated with hexane/isopropanol (9/1) for 30 min to selectively extract PC among choline-containing metabolites (Vassar et al., 2007). The 3H-choline radioactivity was measured using a liquid scintillation counter (Aloka) and normalized according to the protein concentration.
Cells were fixed with a mixture of 2% formaldehyde and 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) and postfixed with a mixture of 1% osmium tetroxide and 0.1% potassium ferrocyanide in 0.1 M sodium cacodylate buffer. Cells expressing HRP-KDEL were incubated in an enzyme histochemical reaction solution to form DAB precipitates before osmification (Brown and Farquhar, 1989; Jokitalo et al., 2001). Samples were embedded in epoxy resin, and ultrathin sections were observed using a JEOL JEM1011 electron microscope operated at 100 kV.
Immuno-EM of ultrathin cryosections
Cells were fixed with 4% formaldehyde and 0.25% glutaraldehyde for 60 min, infiltrated with 2.3 M sucrose in 0.1 M phosphate buffer, pH 7.4, and frozen in liquid nitrogen. Ultrathin cryosections were prepared in an ultramicrotome with an attached cryochamber FC7 (Leica), immunolabeled, stained with 2% neutral uranyl acetate, and embedded in 2% methyl cellulose plus 0.4% uranyl acetate (Tokuyasu, 1980).
Cells were either directly dissolved in SDS sample buffer or treated with RIPA buffer (1% Triton X-100, 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 0.1% SDS, and 0.5% sodium deoxycholate) supplemented with protease inhibitor (Nacalai), and the supernatant was used. The reaction obtained with Super Signal West Dura Substrate (Thermo Fisher) was captured using a Fusion Solo S instrument (Vilber Lourmat) and analyzed using Fusion-Capt Advance Software version 16.15.
Total RNA was extracted from U2OS using the ReliaPrep RNA cell miniprep system (Promega). cDNA was obtained using Superscript III reverse transcription (Thermo Fisher). qRT-PCR was performed using Brilliant II SYBR with a Low Rox Master Mix and an Mx3005P RT-PCR system (Agilent Technologies) according to the manufacturer’s instructions. The variances in the cDNA amount were normalized to the level of β-actin cDNA. The primer sequences are shown in Table S2.
PA and PC measurement in isolated nuclei
The nuclear fraction was isolated using a sucrose method (Nabbi and Riabowol, 2015), and lipids were extracted by the Folch method (Folch et al., 1957). The concentrations of PA and PC were measured as previously described (Morita et al., 2009, 2010). Briefly, lipids were incubated with lipoprotein lipase of Pseudomonas sp. (129–04501; FUJIFILM Wako Pure Chemical) and then with glycerol-3-phosphate oxidase (T-60; Asahi Kasei Pharma) for PA, and with glycerophospholipid-specific phospholipase D from Streptomyces sp. (T-39; Asahi Kasei Pharma) and then with choline oxidase from Alcaligenes sp. (037–14401; FUJIFILM Wako Pure Chemical) for PC. Hydrogen peroxide generated by glycerol-3-phosphate oxidase or choline oxidase was reacted with Amplex Red (Invitrogen) in the presence of HRP (Oriental Yeast) to form the fluorescent reporter resorufin, which was measured using a microplate reader (SpectraMax iD3; Molecular Devices).
Statistical analysis and plots
Normality was assessed by the Shapiro–Wilk test. Statistical differences between samples were examined by Kruskal–Wallis ANOVA test followed by Dunn’s test for the data in Fig. 1, A and C; Fig. 2 C; Fig. 3 E; Fig. 4 A; and Fig. S3 D; Mann–Whitney test for the data in Fig. 1 B; Fig. 2, D and E; Fig. 3 C; Fig. 4, B and G; Fig. 5, C, D, and G; Fig. S2 B; and Fig. S3 H; or Welch’s t test for the data in Fig. S3 F. Pooled data obtained in three independent experiments were analyzed to produce the results shown in Fig. 1, A–C; Fig. 2, C–E; Fig. 3, C and E; Fig. 4, A, B, and G; and Fig. 5, C, D, and G. A representative result of two or three independent experiments is shown for all other data. The bars in each scatter plot indicate the average. Box plots are used when the sample number exceeds 40. In box plots, the center line indicates the median, box boundaries indicate the 25th and 75th percentiles, whiskers are Tukey type, and the average is marked as “+.” Statistical analyses were performed with Prism 8 (GraphPad).
Online supplemental material
Fig. S1 shows characterization of nuclear LDs in U2OS. Fig. S2 shows effects of lipin-1 expression and seipin knockdown and overexpression on nuclear LDs. Fig. S3 shows characterization of NLS-Opi1 Q2-GFP and effects of seipin knockdown on PA in the nucleus. Table S1 lists the sequences of siRNAs. Table S2 lists the sequences of RT-PCR primers. Videos 1 and 2 show nuclear LD formation at the nuclear envelope in U2OS treated with OA for 1 d. Video 3 illustrates the appearance of red Dendra2-LPCAT1 signal in nuclear LDs after photoconversion in the cytoplasm. Video 4 demonstrates FRAP of GFP-DGAT2 in nuclear LDs. Video 5 shows the disappearance of GFP-ACSL3 from nuclear LDs. Video 6 displays live imaging of GFP-Lipin1β in U2OS treated with OA and Torin1. Videos 7 and 8 demonstrate FRAP of GFP-ACSL3 in nuclear LDs in U2OS transfected with control siRNA and seipin siRNA, respectively. Video 9 shows live imaging of NLS-Opi1 Q2-GFP in U2OS treated with OA for 1 d.
We thank the staff members of the Division for Medical Research Engineering and the Radioisotope Research Center of Nagoya University Graduate School of Medicine for their skillful support, Dr. Takaki Miyata (Nagoya University, Nagoya, Japan) and Dr. Noboru Mizushima (The University of Tokyo, Tokyo, Japan) for generously permitting the use of their microscopes and reagents, and many other colleagues for the donation of valuable materials.
This study was supported by the Japan Society for the Promotion of Science KAKENHI grants to Y. Ohsaki (18K06829) and T. Fujimoto (15H05902 and 18H04023), and Advanced Bioimaging Support (JP16H06280).
The authors declare no competing financial interests.
Author contributions: K. Sołtysik, Y. Ohsaki, and T. Fujimoto designed the experiments and analyzed the results. K. Sołtysik, Y. Ohsaki, T. Tatematsu, J. Cheng, A. Maeda, and S. Morita performed the experiments. T. Fujimoto wrote the manuscript.
K. Sołtysik’s present address is Department of Biochemistry and Molecular Biology, Graduate School and Faculty of Medicine, The University of Tokyo, Tokyo, Japan.