TMEM147 interacts with lamin B receptor, regulates its localization and levels, and affects cholesterol homeostasis

The endoplasmic reticulum (ER) is the most extensive intracellular membranous compartment, consisting of a network of interconnected flat sheets and tubules, and the contiguous nuclear envelope (NE) (Baumann and Walz, 2001; Lynes and Simmen, 2011). The ER, as the site of protein synthesis, processing and assembly and membrane and lipid supplier of the cell, exhibits not only morphological but also functional compartmentalization (Baumann and Walz, 2001; Chen et al., 2013; Hu et al., 2011; Goyal and Blackstone, 2013; Lin et al., 2012; Shibata et al., 2006). Understanding how structural and functional specificity is formed and maintained is an ongoing challenge.

In our work on ER membrane morphogenesis (Santama et al., 2004; Christodoulou et al., 2016), we recently focused on transmembrane protein TMEM147, a 25 kDa protein that localizes to ER membranes (Dettmer et al., 2010; Rosemond et al., 2011) and possesses seven transmembrane domains with its N-terminus residing in the ER lumen and its C-terminus facing the cytosolic side of the membrane. The amino acid sequence of TMEM147 is highly conserved among mammalian species (human, rat, bovine and mouse sequences are 99% identical) and it is expressed at constant levels across many human tissues (Rosemond et al., 2011). TMEM147 was found to exhibit intricate reciprocal interactions with the ER proteins nicalin and NOMO (NOMO1,-2 and -3) and be a core component of a complex involving all three proteins that is important for early embryonic Nodal signaling in vertebrates (Dettmer et al., 2010). Furthermore, TMEM147 has been shown to influence forward trafficking, via the ER, of the muscarinic acetylcholine receptor M3R (CHRM3) (Rosemond et al., 2011). Both studies indicate important protein interactions of TMEM147 within the ER; its function, however, is still unknown.

At the onset of our experiments, we detected interactions between TMEM147 and lamin B receptor (LBR), which we fully characterize in this work (see Results). LBR is an extensively studied cellular protein that localizes to the inner nuclear membrane (INM). Its nucleoplasmic, hydrophilic N-terminus [amino acids (aa) 1–208 in human, including a chromatin-binding Tudor domain] was found to interact with lamin B (hence its name) and heterochromatin-associated proteins (HP1 and MeCP2) (Makatsori et al., 2004; Pyrpasopoulou et al., 1996; Worman et al., 1988). Through these interactions, LBR is assumed to provide a platform for internal nuclear organization by mediating the deposition of repressed and transcriptionally inactive heterochromatin to the NE periphery. The C-terminus of LBR (aa 208–615) is proposed to contain 8–10 transmembrane domains, whose topological organization within the INM is not settled in literature, and a short C-terminal ‘tail’ whose localization can be either nucleoplasmic or within the lumen between the outer and inner NE sheets (Olins et al., 2010).

A fascinating aspect of LBR structure is that, while the N-terminal part has a distinct nuclear function, unexpectedly, its C-terminus resembles C-14 sterol reductase enzymes (Li et al., 2015; Silve et al., 1998; Worman et al., 1990). In fact, the intron between exons 4 and 5 in the LBR gene is 10 kb long, raising the possibility that the gene may have evolved from a recombination event between two primordial genes, giving rise to a chimeric protein (Schuler et al., 1994; Olins et al., 2010). Specifically, LBR displays significant sequence similarity over its C-terminus with two key sterol reductases involved in late, post-squalene and smooth ER-localized steps of cholesterol biosynthesis; the C-terminus of LBR is 37% identical and 62% conserved with the 7-dehydrocholesterol reductase DHCR7, and 58% identical or 75% conserved with the 3 β-hydroxysterol Δ-14 reductase TM7SF2. Additionally, LBR harbors two ‘sterol reductase family signatures 1 and 2’ within its C-terminus, short motifs with similarities to many sterol reductases, including TM7SF2 and DHCR7 (Olins et al., 2010).

Functionally, human LBR complements the TM7SF2-like C14 sterol reductase ERG24 in Saccharomyces cerevisiae (Silve et al., 1998), and both LBR and TM7SF2 complement ERG3 reductase in Neurospora crassa (Prakash et al., 1999). Furthermore, it has been demonstrated that LBR is required for viability under cholesterol starvation conditions, and is constitutively expressed and essential for cholesterol synthesis, despite the presence of sterol-inducible TM7SF2 in human LBR-knockout cell lines (Tsai et al., 2016). The same study documented that the C-terminal sterol reductase domain of LBR alone is both necessary and sufficient for cholesterol production in HeLa cells and, importantly, nonsense LBR mutations, associated with Greenberg skeletal dysplasia or Pelger–Huët human congenital disorders of cholesterol metabolism, fail to rescue the cholesterol auxotrophy of an LBR-deficient human cell line (Tsai et al., 2016). Although LBR and TM7SF2 catalyze the same biosynthetic step and it appears that they provide some compensatory redundancy, their functional interactions are far from clear. Deletion of mouse Tm7sf2 does not impair cholesterol biosynthesis (Bennati et al., 2008) and, interestingly, a small-molecule inhibitor of TM7SF2 drives oligodendrocyte differentiation from progenitor cells and induces oligodendrocyte remyelination via the concomitant accumulation of the intermediate 8,9-unsaturated sterol (Hubler et al., 2018).

Mutations in DHCR7, characterized by a cholesterol deficit and 7DHC precursor accumulation, are the cause of Lemli–Optiz syndrome (SLOS), the most common cholesterol metabolic disorder (Fitzky et al., 1998; Wassif et al., 1998; Waterham et al., 1998; Tint et al., 1994). Reduced cholesterol levels in SLOS impair hedgehog signaling by inhibiting the appropriate ciliary localization and activation of the smoothened receptor (Blassberg et al., 2016). The understanding of the molecular interactions between DHCR7 and LBR in cholesterol synthesis, pursued in mouse double knockout studies, remains incomplete (Wassif et al., 2007).

Thus, the overall functional interrelationships between LBR, TM7SF2 and DHCR7 in cholesterol biosynthesis require further elucidation. The role of LBR as a multitasking protein at the INM, the possible co-ordination between its dual role and its regulation also remain focal points of interest. In this work, we present new findings showing that the ER transmembrane protein TMEM147 drastically influences LBR levels and LBR targeting to the INM and impacts on cellular cholesterol levels by regulating both LBR and DHCR7.

Extending our interest in the role of ER-transmembrane proteins on ER morphogenesis and function (Christodoulou et al., 2016), we focused in this work on the characterization of TMEM147 (also known as NIFIE14; NM_032635.4) in human cells. Because of lack of an antibody suitable for immunofluorescence, we pursued transient transfections with an N-terminally FLAG-tagged full length TMEM147 cDNA in HeLa cells; this revealed labeling of both the ER (Fig. 1A1,A2), consistent with previous findings (Dettmer et al., 2010; Rosemond et al., 2011), and also of the nuclear rim (Fig. 1A1; this is also visible with COS-7 cells in fig. 5A of Rosemond et al., 2011). We subsequently constructed a stable cell line, HeLa-TMEM147-GFP, constitutively expressing C-terminally GFP-tagged TMEM147. In these cells, TMEM147–GFP displayed the same distribution as the FLAG-tagged version (Fig. 1B1,B2) and confirmed colocalization at the ER upon double labeling with the ER marker protein calnexin (Fig. 1C1–C3) and overlap at the NE upon double labeling with the INM marker Lap2β (Fig. 1D1–D3).

To probe the function of TMEM147, we used RNAi as a tool. We established conditions for reliable siRNA-mediated silencing of TMEM147: typically, TMEM147 mRNA levels were reduced down to 2.10±1.87% (mean±s.e.m.) or by a factor of 47.42 (P<0.0001) compared to what was detected with negative control silencing (mock silencing without siRNAs), as assessed with RT-qPCR (Fig. 2A). In silenced HeLa cells, TMEM147 protein levels, as assessed by quantitative western blot (WB) analysis, represented only 13.36±2.27% (mean±s.d.; P=0.0011) of negative control levels (Fig. 2B,C). When silencing was repeated in the stable HeLa-TMEM147-GFP cell line, it was accompanied by loss of GFP signal in immunofluorescence experiments (as a proxy for TMEM147–GFP expression) in silenced cells only, and not in negative control cells (Fig. 2D, compare D2 and D4), and by a marked reduction of signals for both native TMEM147 and TMEM147–GFP proteins as determined by WB analysis (Fig. 2E).

One consistent observation with TMEM147 silencing was decreased cell viability compared with negative control cells, specifically affecting growth at late stages post-transfection (48–72 h) (Fig. 3A). In addition, the most striking result upon TMEM147 silencing was our observation of a concomitant drastic reduction of levels of INM LBR, which could be detected at earlier time points (already by 48 h). As shown by representative WB experiments in Fig. 3B, LBR is consistently and significantly downregulated. Quantification showed LBR protein levels reaching only 9.93±6.23% (P=0.0065; mean±s.d.) of negative control levels (Fig. 3C), a reduction closely resembling that typically obtained for the TMEM147 protein itself (Fig. 2C). By comparison, the relative levels of emerin, another INM integral protein, and those of the protein-folding chaperone and ER integral protein calnexin, were not detectably altered (Fig. 3B).

The defining reduction of LBR protein was also visualized by confocal microscopy, revealing that specifically in TMEM147-silenced cells, LBR is hardly detectable at the INM (Fig. 3, compare panels D3 and D6), while lamin A, a nuclear lamina constituent demarcating the INM, is essentially unaffected (Fig. 3D2 and D5). The loss of immunofluorescence signal for LBR upon TMEM147 silencing in HeLa cells (also seen by wide-field microscopy; Fig. S1A1–B3) was recapitulated in the stable HeLa-TMEM147-GFP cell line, in which loss of GFP signal (Fig. S1C1,D1) was accompanied by corresponding disappearance of LBR label in the same cells (see Fig. S1C2,D2, and C3 and D3 for overlays).

We sought to further examine the specificity of the observed effect of TMEM147 silencing on LBR protein by utilizing an alternative siRNA (oligo#2), targeting a more downstream exon sequence than that used in the previous experiments (oligo#1) (#oligo1 and oligo#2 are shown in Fig. S2A, displaying part of the gene sequence and a TMEM147 protein membrane topology model). Silencing of TMEM147 with oligo#2 not only effectively reduced TMEM147 protein levels but, again, also decreased LBR levels, as observed by WB analysis (Fig. S2B), and visibly reduced LBR immunofluorescence signal, compared to negative control cells (Fig. S2, compare panels C2 and D2).

These data, therefore, show that reduction of TMEM147 results in a concomitant reduction of LBR protein at the INM.

To analyze the cellular phenotype resulting from TMEM147 silencing and its specific effect on LBR in more depth, we conducted a quantitative morphometric comparison in four independent experiments of TMEM147-silenced cells (n=48 cells) and negative control cells (n=55) labeled with anti-LBR antibody and Hoechst stain and optically sectioned by confocal microscopy (Fig. 4A1 and A2). Quantification of LBR mean fluorescence intensity in all cells was in line with our quantification of LBR protein levels by WB analysis, and again confirmed the significantly reduced signals in TMEM147-silenced vs control cells (P<0.0001) (Fig. 4B1). Interestingly, quantitation of Hoechst fluorescence in the same cells also revealed a clear reduction of Hoechst incorporation, indicative of loss of chromatin compaction specifically in TMEM147-silenced cells (P<0.0001) (Fig. 4A3,A4 and B2) and, consistent with this, a clear reduction of histone H3K9 methylation, serving as a heterochromatin marker and anchor of chromatin to the nuclear lamina (reviewed by Mattout et al., 2015) (Fig. 4A5,A6). Additionally, TMEM147 silencing appeared to introduce changes in nuclear shape (altered sphericity, P<0.0001), presumably due to changes in the shape of the INM, where LBR resides (Fig. 4B3).

Using the stacks of optical sections for all 103 cells analyzed above, we produced 3D-rendered images which revealed a striking mislocalization of LBR, with partitioning of LBR signal to both INM and ER membranes specifically in TMEM147-silenced cells (TMEM147 silencing led to 6.0-fold increased frequency of cells with LBR-positive staining in the ER compared to negative control cells, P<0.0001) (see Fig. 4C,D for representative images; Movie 1 for animation; Table S1 for quantification). Taken together, our analyses suggest that not only does loss of TMEM147 negatively impact LBR production, but it also results in the loss of correct targeting of the residual LBR to the INM and its enhanced mislocalization to the ER. Furthermore, the diminution of LBR levels at the INM is accompanied by chromatin decondensation and a change in nuclear shape.

Prompted by the observed impact of TMEM147 on LBR protein levels and intracellular targeting, we next investigated, by performing co-immunoprecipitation (co-IP) experiments, whether the two proteins also physically interact. First, conducting IP with an anti-GFP antibody in extracts from the HeLa-TMEM147-GFP or a HeLa-GFP-only stable cell line resulted in the co-selection and detection of endogenous LBR specifically from HeLa-TMEM147-GFP cells, excluding unspecific binding of antibody or GFP to LBR (Fig. 5A, compare lanes c and d). Second, additional IP analyses in HeLa-TMEM147-GFP or HeLa-GFP-only cells that were concurrently transiently transfected with FLAG-tagged full-length LBR detected FLAG–LBR specifically in the bound fraction of TMEM147–GFP samples (Fig. 5B, compare lanes c and f). Thus, these experiments confirm a protein–protein interaction between TMEM147 and LBR and, together with the silencing results, imply a functional interaction between the two proteins.

As mentioned in the Introduction, LBR protein displays a modular organization with involvement of its nucleoplasmic N-terminal domain in chromatin binding and organization at the nuclear periphery, and implication of its membrane-spanning C-terminus in cellular cholesterol biosynthesis. This modular structure thus reflects distinct cellular functions of LBR. To probe the functional association between TMEM147 and LBR, we conducted TMEM147-silencing experiments using, in parallel, both wild-type HeLa cells and a stable cell line expressing a GFP-tagged truncated version of LBR, consisting of its full N-terminus plus its first transmembrane domain, HeLa-LBR₂₃₈-GFP (Fig. 6A1; cell line by Ellenberg et al., 1997). In these experiments, we assessed and compared the fate of both the endogenous, full-length LBR and the N-terminal LBR₂₃₈–GFP construct, upon TMEM147 silencing. Owing to their difference in mass, the two forms of LBR can be distinguished in WB analysis with the anti-LBR antibody, while the truncated form is moreover uniquely detectable with anti-GFP antibody (Fig. S3A,B). As expected, TMEM147 silencing (22.60±6.81% of negative control levels, P=0.0025; mean±s.d.) produced a robust reduction of endogenous LBR protein (12.77±11.58% of negative control levels, P=0.045) (Fig. 6A2,A3). In contrast, the signal detected for LBR₂₃₈–GFP either with anti-LBR or anti-GFP antibodies upon TMEM147 silencing exhibited modest reductions relative to control levels (96.39±41.78% and 77.12±18.10%, respectively) that were not statistically significant (Fig. 6A2,A3). This indicated that the observed stability and persistence of the N-terminal truncated LBR, which contrasted with the drastic reduction of the full-length LBR, likely stemmed from a requirement of the LBR C-terminus for the effect of TMEM147 on LBR protein levels to be manifested. We note, however, the lack of LBR promoter elements driving the expression of LBR238–GFP in the stable cell line, which would explain the lack of effect if regulation was only at the transcriptional level.

We, therefore, next directly addressed the interaction between the two proteins. We designed two overlapping truncated, tagged versions of LBR: LBR₃₇₂–GFP (the ‘N-terminal construct’ encompassing the first 372 amino acids of LBR) and LBR_209-615–GFP (the ‘C-terminal construct” containing amino acids 209–615) and also employed full-length LBR–GFP as a positive control for TMEM147–LBR interaction (Fig. 6B1). Upon transient transfection in HeLa cells, all three constructs localized appropriately to the nuclear rim, with some partitioning to the ER, likely due to overexpression (Fig. 6B2–B4). We conducted IP experiments with anti-GFP beads, immobilizing the GFP-tagged N-terminal, C-terminal or full-length LBR as bait and screened for the presence of endogenous TMEM147 in the bound fractions. We found that, compared to binding of TMEM147 to the full-length LBR–GFP, the binding to the C-terminal construct appeared equally strong, whereas binding to the N-terminal construct was very markedly reduced (Fig. 6B5, compare bound TMEM147 signals in lanes c, f and i). In agreement, the LBR₂₃₈–GFP N-terminal construct, used in Fig. 6A1–A3, was also unable to bind the native TMEM147, as was the GFP-only bait, used as a negative control (Fig. S3C). These results seem consistent with a protein interaction of TMEM147 with the C-terminus of LBR, and in agreement with the likely topology and orientation of LBR in the INM. Taken together, the results suggest that this interaction at the C-terminus of LBR is the basis of the functional interaction between the two proteins.

On the basis of our findings that there is a physical and functional interaction between TMEM147 and the C-terminus of LBR, known to exhibit essential C14-sterol reducing activity in cellular cholesterol synthesis, we then investigated whether knockdown of TMEM147 may also impact other key sterol reductases in the same pathway.

Interestingly, quantitative WB analysis of protein levels of 7-dehydrocholesterol reductase DHCR7, the sterol reductase catalyzing the ultimate step in cholesterol biosynthesis and localized, similar to TMEM147, at both ER and the INM (Koczok et al., 2019), revealed that it, too, showed significant reduction to only 32.78±7.98% (mean±s.d.) of control levels specifically upon TMEM147 silencing in HeLa cells (P=0.026) (Fig. 7A,B). Repetition of these experiments with the stable HeLa TMEM147-GFP cells confirmed similar reductions (Fig. 7A and see legend).

This finding was supported by our additional discovery that DHCR7 was one of the proteins identified as interacting with TMEM147–GFP by anti-GFP IP, followed by liquid chromatography coupled with tandem mass spectrometry (Fig. S4). In this proteomic analysis, LBR came up as another enriched hit, consistent with our other results, as did the known TMEM147-interacting proteins nicalin and NOMO (Dettmer et al., 2010), validating the analysis (Fig. S4). Thus, TMEM147 and DHCR7 physically interact, and, similar to what is seen for LBR, DHCR7 protein levels are significantly reduced in the absence of TMEM147. TM7SF2, a second key C14 sterol reductase catalyzing the NADPH-dependent reduction earlier in the cholesterol biosynthetic pathway, was also identified as an interacting protein but, in the absence of a reliable antibody, we were unable to validate this further.

We extended our analysis by assessing whether TMEM147 silencing affected transcription levels of LBR, DHCR7 and TM7SF2. Quantification by RT-qPCR indicated that, upon TMEM147 silencing, there was a clear concurrent reduction of gene expression for both LBR and DHCR7 (Fig. 7C; P<0.001 for both LBR and DCHR7), consistent with the reductions already identified at protein level and the corresponding protein–protein interactions between TMEM147 and these key reductases. In contrast, the level of mRNA for the reductase TM7SF2 remained essentially unchanged (Fig. 7C). We repeated a quantitative analysis of TMEM147, LBR, DHCR7 and TM7SF2 expression levels, this time from cells grown in parallel in either complete or lipid-restrictive medium (no serum) (Fig. S5). We found LBR to be constitutively expressed and unresponsive to lipid starvation (Fig. S5B), in agreement with previous reports (Tsai et al., 2016) and the same applied to the expression of TMEM147 (Fig. S5A). Furthermore, we observed that negative control cells (as previously reported by Bennati et al., 2008 and Tsai et al., 2016) but also TMEM147-silenced (thus LBR-downregulated) cells upregulated TM7SF2 under lipid restriction (Fig. S5C). Interestingly, we detected an even stronger upregulation of DHCR7 under lipid restriction, and this response was also directly correlated with TMEM147/LBR expression (Fig. S5D).

Overall, the interconnection found between TMEM147 and both LBR and DHCR7 was remarkable and raised the possibility of a resulting impact of TMEM147 depletion on cellular cholesterol biosynthesis. We thus next employed a full lipidomic analysis to directly measure lipids, including total cellular cholesterol and cholesteryl esters and also quantify the cholesteryl ester species profile (Table S2). We conducted a multivariate analysis of the lipidomics data; a principal component analysis (PCA) of the cellular lipidome (Fig. S6A) indicated that TMEM147-silenced cells separated from the control-silenced and untreated cells along the first principal component. Using the loadings plot, we identified that cholesterol and cholesteryl esters (CE) were among the most significant features that drive this separation (Fig. S6B). Surprisingly, total free cholesterol cellular levels in TMEM147-silenced cells exhibited a small but significant increase (by 13.4%) when compared with negative control cells (Fig. 7D). Specifically, free cholesterol represented 16.36±0.45 mol% in TMEM147-silenced cells vs 14.43±0.16 in negative control (P<0.05; mean±s.d.) (Fig. 7D). At the same time, total levels of cholesteryl esters (CEs) were greatly reduced, by 40.5% (5.80±1.68 mol% in TMEM147-silenced cells versus 9.74±1.83 compared to negative control cells; P=0.05) (Fig. 7D). In addition, quantitation of the full spectrum of lipid species in CEs displayed TMEM147 silencing-specific alterations (both decreases and increases) in several lipid species (Fig. S6C) with decreases in saturated and short lipid species and increases in several polyunsaturated long fatty-acyl chains of the esterified cholesterol (Fig. S6C).

CEs, as the intracellular storage forms of excess cholesterol, are of central importance to cholesterol homeostasis, and their formation is a measure of the availability of cellular free cholesterol (Luo et al., 2020); reduced CE levels would be consistent with downregulation of sterol reductase activity in TMEM147-silenced cells. At the same time, levels of total free cellular cholesterol in TMEM147-silenced cells, which were found to be modestly increased in our experiments, represent the contribution of both cellular synthesis as well as that of cholesterol uptake from the growth medium. To specifically determine the contribution of cholesterol uptake in the different cell groups, we therefore conducted cholesterol uptake assays using fluorescently labeled cholesterol, comparing negative control and TMEM147-silenced cells under different growth conditions. Cells were grown, in parallel, in complete or lipid-restrictive medium (no serum) or in the presence of cholesterol transport inhibitor U-18666A (as a positive control of increased uptake) (Fig. 7E). In all conditions tested, normalized cholesterol uptake was markedly and significantly increased in TMEM147-silenced cells in comparison with negative controls. In particular, uptake was doubled in restrictive medium conditions relative to negative control, and in complete medium it increased by 2.3-fold (Fig. 7E). Increased uptake, to counteract impaired cellular biosynthesis due to the reduction of both LBR and DHCR7, would explain the maintenance of free cholesterol levels in TMEM147-silenced cells grown in non-restrictive media (Fig. 7D).

Consistent with these findings was the effect of cholesterol rescue in growth assays. Control and TMEM147-silenced cells were grown either in complete medium or in complete medium followed by a 24-h lipid restriction (no serum) either on its own or in combination with exogenous cholesterol, and cell viability was monitored at different time points (Fig. 7F,G). Addition of exogenous cholesterol for the last 8 h during the 24-h lipid restriction rescued growth and restored it by 72 h to the same levels [(14.27±0.95)×10⁴ cells] as those observed for cells at 48 h in complete medium [(14.06±1.13)×10⁴ cells], specifically in TMEM147 silencing (Fig. 7G); in comparison, equivalent TMEM147-silenced cells at 72 h without cholesterol showed the typical drastic reduction in viability [(11.35±0.48)×10⁴ TMEM147-silenced cells with serum starvation at 72 h versus (14.3±0.95)×10⁴ cells with serum starvation plus cholesterol at 72 h (P=0.0046), or (10.10±0.48)×10⁴ TMEM147-silenced cells in complete medium at 72 h compared again with cells in the presence of cholesterol (P=0.0003)] (Fig. 7G). Cholesterol addition had only a very modest positive effect on lipid-restricted negative control cells (Fig. 7F). Interestingly, we observed that in TMEM147-silenced cells that were already growth affected at 48 h, subsequent lipid depletion at 48 h had essentially no additional aggravating impact compared to the rapid growth decline manifested by 72 h in complete medium, in contrast to control cells (Fig. 7F,G), suggesting that the detrimental impact of TMEM147 silencing on growth was severe and already developing in complete medium (see also Fig. 3A). This is different to the observations of Tsai et al. (2016), where decreased viability of LBR-knockout cells occurred under lipid restriction only, and likely underlines the more severe phenotypes resulting from TMEM147-induced depletion of both LBR and DHCR7.

In conclusion, our work provides evidence to indicate a role for TMEM147 as a novel functional modulator of cholesterol homeostasis in cells.

Previously described roles for TMEM147 included membrane complex stabilization (Dettmer et al., 2010), acetylcholine receptor trafficking (Rosemond et al., 2011) and NF-κB activation (Ota et al., 2019). In this work, we present evidence for physical and functional interactions between the ER and INM transmembrane protein TMEM147 and the well-characterized LBR protein, which, through its modular structure, serves both as a heterochromatin organizer at the nuclear lamina and a sterol reductase in the late steps of cholesterol biosynthesis. We discovered that downregulation of TMEM147 in human cells induces a sharp reduction in LBR protein levels and mislocalization of the remaining LBR to the ER at the expense of its typical localization at the INM. Furthermore, we discovered that TMEM147 physically interacts with DHCR7, another key sterol reductase in cholesterol biosynthesis in cells. Interestingly, LBR and DHCR7 exhibit co-ordinated reductions in gene expression upon TMEM147 knockdown, in contrast with what was seen with the sterol reductase TM7SF2, the gene expression of which appeared to be unaffected by downregulation of TMEM147. Consistent with these data on interactions and gene expression, we identified changes in cellular cholesterol and cholesteryl ester levels as well as in cellular cholesterol uptake in TMEM147-silenced cells, supporting the idea that TMEM147 functions as an important regulator of cholesterol homeostasis in cells.

Cholesterol is the major sterol in terrestrial vertebrates and critical for normal growth, development and hedgehog signaling in mammals. It is a regulator of membrane fluidity and organization, a key component of lipid rafts in the plasma membrane, and important in myelin formation in nerves (reviewed by Platt et al., 2014). Most mammalian cell types modulate the cholesterol content of their membranes by finely balancing cellular cholesterol biosynthesis, uptake by receptor-mediated endocytosis and export of esterified cholesterol via ABC transporters. Our findings reveal that in the absence of TMEM147 (and consequent depletion in the levels of both LBR and DHCR7), uptake of cholesterol is increased, presumably to counteract the reduction of endogenously produced cholesterol and its cholesteryl esters. Despite increased uptake, changes in cholesterol metabolism may account for the observed cell death at late stages of TMEM147 silencing, compared to negative control cells, even in the presence of complete medium. Addition of exogenous cholesterol increased cell viability, but we cannot formally exclude additive detrimental effects of TMEM147 silencing contributing to increased cell death rates.

Our finding that TMEM147 downregulation in HeLa cells, resulting in concomitant downregulation of LBR (and DHCR7), does not seem to affect gene expression of the reductase TM7SF2, indicates that there are distinct forms of transcriptional regulation for LBR and TM7SF2, even though both proteins are thought to catalyze the same enzymatic step in cholesterol biosynthesis. These results appear in agreement with both the study by Bennati et al., 2008, in which Tm7sf2^−/− mice had normal LBR protein levels, and the more recent analysis by Tsai et al., 2016, in which primary human cells or cell lines (including HeLa cells) did not show co-ordination of gene expression between LBR (constitutively expressed) and TM7SF2 (sterol-inducible), either under standard conditions or under prolonged lipid starvation. During mouse liver regeneration, TM7SF2 and LBR are again differently modulated and absence of TM7SF2 does not impair cholesterol biosynthesis (Bartoli et al., 2016). TM7SF2 and LBR were also found to undergo differential post-translational regulation, with TM7SF2 manisfesting sterol-dependent turnover while LBR remained stable (Capell-Hattam et al., 2020). The unresponsiveness of TM7SF2 to changes in LBR expression, revealed in all of these studies and by our experiments, is somewhat counterintuitive and would challenge the notion that TM7SF2 can compensate for cholesterol biosynthesis in the absence of functional LBR in order to sustain cholesterol levels (Wassif et al., 2007).

We report here both a novel protein–protein interaction between TMEM147 and DHCR7, a co-ordinate gene expression between LBR and DHCR7 in response to TMEM147 depletion in normal conditions, and an induction of DHCR7 under lipid restriction that is modulated by TMEM147/LBR expression. Despite the wealth of studies on DHCR7 as the cause of SLOS, its functional relationship and molecular interactions with LBR have remained relatively obscure. It would appear that TMEM147 as an upstream positive regulator impacts both LBR and DHCR7, affecting their gene expression and protein levels in a molecular mechanism that remains to be investigated. Interestingly, protein interaction data (https://www.ebi.ac.uk/intact/interactors/id:Q9BVK8), include experimental evidence for physical interaction between TMEM147 and the emopamil-binding protein EBP (or D8D7I), a 3-β-hydroxysteroid-δ(8),δ(7) isomerase catalyzing the conversion of δ(8) sterols to their corresponding δ(7) isomers in cholesterol biosynthesis, upstream of DHCR7 (Silve et al., 1996). EBP (also known as D8D78I), like TMEM147 and LBR, localizes to both the ER and the NE, and its mutations impair cholesterol synthesis and cause X-dominant chondrodysplasia punctata with ichthyosis (Braverman et al., 1999; Liu et al., 1999). EBP displays co-ordinate transcription with DHCR7, with which it also forms a bifunctional complex (ChEH, cholesterol-5-6-epoxide hydrolase) (Kedjouar et al., 2004; de Medina et al., 2010), raising the possibility that the pattern of common gene expression of LBR and DHCR7 influenced by TMEM147 may include further enzymes of late cholesterol synthesis steps, like EBP.

Our observation of LBR mislocalization to the ER in the absence of TMEM147 is intriguing and reminiscent of its similar partitioning when the nucleoporin ELYS (also known as AHCTF1) is downregulated (Clever et al., 2012; Mimura et al., 2016). Following its synthesis, LBR diffuses laterally through the ER and the nuclear pore membrane domains, via lateral peripheral channels of the complexes, before becoming anchored at the INM by binding, via its N-terminus, to the nuclear lamina, heterochromatin or other proteins (Ellenberg et al., 1997; Makatsori et al., 2004; Ungricht et al., 2015; Boni et al., 2015; Giannios et al., 2017). The vast majority of steady-state LBR is at the INM, as opposed to the ER (Nikolakaki et al., 2017). Not only does the number of nuclear pore complexes thus appear important, but also INM localization of LBR was found to be inhibited by SR phosphorylation of the N-terminal domain, involving kinases CDK, SRPKs or ELYS protein acting on protein phosphatases (Nikolakaki et al., 1996; Tseng and Chen, 2011; Fukuhara et al., 2006; Mimura et al., 2016). An interesting point of discussion has been whether the ability of LBR to contribute to cholesterol synthesis in vivo is conditional on its presence at the NE or the ER (Hoffmann et al., 2007). If LBR localization is indeed critical for its enzymatic activity, then it would appear that interaction between LBR and TMEM147 is essential for INM targeting of LBR. We found that absence of TMEM147 not only caused a drastic decline of LBR protein levels but also an altered diffusional mobility of LBR and its relocation to the ER. This mislocalization of LBR may further contribute to the impact of LBR reduction on cholesterol synthesis.

Other phenotypic effects that we report, upon TMEM147 silencing and the associated LBR depletion from the NE, were a change in nuclear shape (but not size or surface area) as well as evidence of loss of chromatin compaction. Both of these features (1) are highly consistent with observed effects upon loss of LBR expression in many studies, and (2) were shown to be tightly entwined in cellular physiology. Specifically, in Pelger–Huët syndrome in humans, mutations resulting in truncated and unstable forms of LBR are accompanied by hypolobulated nuclei in granulocytes (Hoffmann et al., 2002), while in ichtyosis in mice, nuclei are ovoid and with ‘inverted nuclear organization’, that is with heterochromatin masses accumulating internally and non-compacted chromatin at the nuclear periphery (Hoffmann et al., 2007; Shultz et al., 2003). Additionally, it is well documented that in healthy blood granulocyte differentiation in the bone marrow, wild-type LBR is developmentally upregulated as granulopoiesis progresses, relating to both increased nuclear indentation (shaping the characteristic polylobular nuclei) and heterochromatin tethering to the nuclear periphery in mature neutrophils, the main granulocyte cell type (Olins et al., 2008). In myeloid cell line models (human HL-60 and mouse EPRO promyelocytes), correspondingly, the absence of LBR produces profound effects in both nuclear shape and heterochromatin distribution in a dose-dependent manner (Zwerger et al., 2008; Hoffmann et al., 2007; Olins et al., 2010). In the same vein, it was demonstrated that LBR knockdown in HeLa cells resulted in redistribution and unfolding of heterochromatin from the INM towards the nucleoplasm (Lukášová et al., 2017). LBR has additionally been proposed as a stimulator of NE growth on the basis of overexpression experiments resulting in NE overproduction, inducing NE invagination and membrane stack formation (Ma et al., 2007).

The evidence linking the generation of nuclear shape with the function of LBR as organizer of underlying heterochromatin is compelling; however, one thought-provoking novel hypothesis, also taking into consideration the question of why there should be a need for cholesterol production specifically at the INM (in addition to the ER), is the idea that cholesterol, becoming readily accessible to the nucleus, may participate in the assembly of nuclear lipid microdomains (‘nuclear lipid rafts’, in analogy to plasma membrane lipid rafts) that may provide specialized chromatin-anchoring platforms, nucleosome-binding domains, or scaffolds for protein organization or oligomerization (Cascianelli et al., 2008; Codini et al., 2016; Silva et al., 2017; Nikolakaki et al., 2017). This possibility was proposed as an explanation of the unique role of LBR in cholesterol biosynthesis, which cannot be compensated for by the ER sterol reductase TM7SF2, even though both proteins catalyze the same reaction (Nikolakaki et al., 2017). If this were indeed the case, it would provide an intriguing and meaningful framework to reconcile and understand the dual and diverging modular structure and functional role of LBR in both nuclear organization/gene expression and cholesterol biosynthesis. And it would illuminate TMEM147 as a putative upstream functional integrator, via its modulation of LBR, of gene expression and cholesterol homeostasis. Dissecting the molecular mechanism through which the novel physical and functional interaction of TMEM147 with LBR impacts both processes is now essential in order to fully understand their cellular interplay.

HeLa (Kyoto; K; EMBL) cells were cultured in DMEM containing 10% (v/v) fetal bovine serum (FBS), 2 mM glutamine and 50 U/ml of penicillin/streptomycin (complete medium) and maintained at 37°C in 5% CO₂. For some experiments, cells were grown in lipid restrictive medium (complete medium without FBS). Cells were regularly tested by PCR and Hoechst 33342 staining to ascertain lack of mycoplasma infection.

The stable HeLa K TMEM147-GFP cell line was generated by transient transfection of pEGFPN1-TMEM147 into HeLa K cells. Expressing cells were selected in the complete medium supplemented with 0.5 mg/ml G418 (Invitrogen) and further enriched by FACS.

The HeLa LBR₂₃₈–GFP stable cell line, expressing a fusion between the first 238 N-terminal amino acids (aa) of human LBR and a C-terminal in-frame GFP sequence, was a gift from Jan Ellenberg (EMBL Heidelberg, Germany).

For relative quantification of mRNA in silencing experiments, RT-qPCR was conducted on the CFX96 Real-Time PCR system (Bio-Rad) in 96-well plates and primers specific for either human TMEM147 (for, 5′-GTACAACGCCTTCTGGAAATG-3′; rev, 5′-ATGCTGTTCTTGGCCACTTT-3′), LBR (for, 5′-GGTCGACCACCTAAAAGTGC-3′; rev, 5′-CCCCATTATATCTGCTGATGC-3′), DHCR7 (for, 5′-CCCAGCTCTATACCTTGTGG-3′; rev, 5′-CCAGAGCAGGTGCGTGAGGAG-3′) or TM7SF2 (for, 5′-AACTCAGGCAATCCGATTTACG-3′; rev, 5′-GGGTCGCAGTTCACAGAAATA-3′). Melting curve analysis was performed to determine the amplification specificity. Three independent experiments were conducted, and each included two no-template controls; all samples were repeated in triplicate reactions. For data normalization, expression of ribosomal protein L19 (RPL19) and pumilio homolog 1 transcript variant 2 (PUM1) were used as reference genes and mammaglobin B-2 (MGB2) as an unrelated marker (all sequences of PCR primers as per Pantelidou et al., 2007). For statistical analysis of RT-qPCR data, the REST-384^© software was used to calculate the relative expression ratio of the different groups and determine the significance of results (Pfaffl et al., 2002; Pantelidou et al., 2007). For the RT-qPCR analysis in Fig. S5, negative control or TMEM147-silenced cells were grown in parallel either in complete medium (72 h) or lipid restrictive medium (48 h in complete medium+24 h under serum starvation). The Common Base method was used to allow direct quantitative comparisons across samples (Ganger et al., 2017), combined with statistical evaluation by two-way ANOVA followed by Tukey's post hoc test for multiple comparisons (GraphPad Prism 8.0).

Transfections for RNAi experiments were performed using INTERFERin (Polyplus Transfection) and siRNA oligonucleotides specific for TMEM147 (oligo#1, 5′-GGCGGCAUCUAUGACUUCATT-3′; oligo#2, 5′-CGCUAUGAUCUGUACCACATT-3′) (Ambion Inc.), at a final concentration of 40 nM and according to the manufacturer's specifications. Negative control was a mock, no-oligo, transfection reaction. Cells were harvested 72 h after addition of siRNAs (96 h for oligo#2) for immunofluorescence (IF), RT-qPCR and western blot (WB) analysis.

Protein samples were heated only to 60°C for 10 min to avoid aggregation of TMEM147 (Dettmer et al., 2010). SDS-PAGE was performed on a Mini-Protean II Electrophoresis Cell, WB on a Mini Trans-Blot Transfer Cell (Bio-Rad), using 48 mM Tris-HCl pH 9.2, 39 mM glycine and 20% (v/v) methanol for transfer, and the ECL System (GE Healthcare) with G:BOX (SYNGENE) for visualization. For quantification of protein levels, intensity volumes (area×height) of signals were extracted with ImageJ 1.49n and normalized using same-sample and same-membrane band intensities for the housekeeping protein GAPDH (in triplicate reactions for each independent experiment). GAPDH-normalized protein expression values were subjected to normality tests (GraphPad Prism 8.0) and significance of knockdown for individual protein bands was assessed by corresponding parametric or non-parametric statistical comparisons (Microsoft Excel and GraphPad Prism 8.0) (see figure legends for further details).

A mouse monoclonal antibody against human TMEM147 (NM_001242598.1) had previously been generated and characterized (Dettmer et al., 2010) and was used here at 1:8 dilution for WB. Additional commercial primary antibodies used were as follows: rabbit anti-LBR (E398L) (Abcam ab32535; 1:300 for IF and WB), mouse anti-LAP2β (BD Transduction Laboratories 611000; 1:500 for IF), mouse anti-GAPDH (Santa Cruz Biotechnology sc-32233; 1:500 for WB), mouse anti-FLAG (Sigma-Aldrich F4042; 1:500 for IF and WB), mouse anti-EGFP (Roche 11814460001; 1:1000 for IF and WB), rabbit anti-calnexin (Santa Cruz Biotechnology sc-11397; 1:100 for IF and 1:200 for WB), rabbit anti-emerin (Abcam ab14208; 1:400 for WB), rabbit anti-LEM4 (Asencio et al., 2012; 1:100 for WB), mouse anti-lamin A (Abcam ab8980; 1:500 for IF), rabbit anti-DHCR7 (Origene TA349889; 1:300 for IF), anti-histone H3K9-me3 (Millipore 17-625; 1:500 for IF). Primary antibodies were used in conjunction with appropriate fluorescently labeled Alexa-Fluor secondary antibodies (Thermo Fisher Scientific) for immunofluorescence, or HRP-labeled secondary antibodies (Santa Cruz Biotechnology) for WB. Nuclei were counterstained with Hoechst 33342 (0.5 µg/ml).

For expression in HeLa cells, the full length ORF of TMEM147 was amplified from HeLa K with primer set TMEM147F, 5′-CGCTCGAGATGACCCTGTTTCACTTCGGGAACTG-3′ and TMEM147R, 5′-GCGAATTCCGGAGTGCACATTGACAACGGCGACAT-3′, transferred as an XhoI/EcoRI fragment into pEGFP-N1 (Clontech) and transferred as a BglII/KpnI fragment into pFLAG-CMV-2 (Sigma-Aldrich).

An equivalent cloning strategy was followed for generating full-length GFP- and FLAG-tagged LBR constructs for expression in HeLa cells with primer set LBR1-615F, 5′-CGCGCTCGAGATGCCAAGTAGGAAATTTG-3′ and LBR1-615R: 5′-CGCGGAATTCGGTAGATGTATGGAAATA-3′. To clone truncated version LBR₃₇₂–GFP into pEGFP-N1, oligo LBR1-615F was used in conjunction with LBR1-372R, 5′-CGCGGAATTCGACGGCCAATGAAGAAATCATA-3′. Finally, to generate truncated version LBR_209-615–GFP, oligo LBR209-615F, 5′-CGCGCTCGAGTTTGGAGGAGTACCTGGTG-3′ was used in conjunction with LBR1-615R.

HeLa K TMEM147-GFP cell line and HeLa K GFP cell line (negative control), either alone or transiently transfected with FLAG–LBR constructs, were lysed in lysis buffer [20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% v/v NP40 and 1× Complete™ (a protease inhibitor cocktail by Roche)]. Each of the extracts was incubated with a 10 μl slurry of GFP-Trap_A beads (Chromotek) for 2 h at room temperature. After binding, beads were extensively washed in lysis buffer and bound proteins were eluted from beads using 30 μl SDS-PAGE sample buffer. Samples were heated to 60°C for 10 min and then analyzed by SDS-PAGE, followed by WB.

For the co-IP experiment shown in Fig. S4, proteomic analysis was carried out at the EMBL Heidelberg Proteomics Facility. IP samples (TMEM147-GFP IP as test sample, TMEM129-GFP IP for comparison, and GFP-only IP as negative control) from two independent experiments in parallel (six samples in total) were prepared for MS analysis using the SP3 protocol (Hughes et al., 2019). Peptides were labeled with TMT6plex Isobaric Label Reagent (ThermoFisher), according the manufacturer's instructions, pooled and cleaned up further with an OASIS^® HLB µElution Plate (Waters). Offline high pH reversed phase fractionation was carried out on an Agilent 1200 Infinity high-performance liquid chromatography system, equipped with a Gemini C18 column (Phenomenex). MS data acquisition was performed with an UltiMate 3000 RSLC nano LC system (Dionex), fitted with a trapping cartridge (µ-Precolumn C18 PepMap 100) and an analytical column (nanoEase™ M/Z HSS T3 column Waters), directly coupled to an Orbitrap Fusion Lumos (ThermoFisher) mass spectrometer, using the proxeon nanoflow source in positive ion mode. Peptides were introduced via a Pico-Tip Emitter 360 µm OD×20 µm ID; 10 µm tip (New Objective) and an applied spray voltage of 2.4 kV at capillary temperature at 275°C. Full mass scan was acquired with mass range 375–1500 m/z in profile mode with Orbitrap resolution of 60,000 and fill time at maximum of 50 ms. Data-dependent acquisition was performed with Orbitrap resolution at 15,000, fill time of 54 ms and a limitation of 10⁵ ions. A normalized collision energy of 36 was applied and MS² data acquired in profile mode. For MS data analysis, IsobarQuant (Franken et al., 2015) and Mascot (v2.2.07) were used to process the acquired data, which were searched against Uniprot H. sapiens UP000005640 proteome database, containing common contaminants and reversed sequences. Raw output files (protein.txt) of IsobarQuant were processed using the R programming language (ISBN 3-900051-07-0). Only proteins quantified with at least two unique peptides were used for data analysis. Raw signal-sums (signal_sum columns) were cleaned with the ‘removeBatchEffect’ function from the limma package (Ritchie et al., 2015) and further normalized using the variance stabilization normalization package (Huber et al., 2002). All experimental conditions were normalized separately to keep differences in protein abundance. Proteins were tested for differential expression using limma, based on false discovery rate (FDR) correction (Q=5%) and scoring a fold-change of at least 100% as a hit. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (EMBL-EBI) with the dataset identifier PXD019598.

For immunofluorescence, cells were grown on coverslips, fixed with 3.7% (w/v) paraformaldehyde in PHEM (30 mM HEPES, 65 mM PIPES, pH 6.9, 10 mM EGTA and 2 mM MgCl₂) for 10 min, permeabilized for 15 min with 0.5% v/v Triton X-100 in PHEM and immunolabeled with appropriate primary and secondary antibodies. Samples were analyzed with a Zeiss Apochromat 63×1.4 NA oil lens on a Zeiss Axiovert 200M inverted microscope with an AxioCam MRm camera.

Confocal imaging in Fig. 3D was carried out with a Zeiss LSM710 Αxiovert confocal microscope using a 63× Plan-Neofluar 1.4 NA oil immersion objective lens, and in Fig. 4 with a Leica TCS SP2 DMIRE2 confocal microscope using 63×1.4 NA oil immersion objective lens.

Images were acquired with Zeiss Axiovision 4.2 software or LSM Zen or Leica LAS (v.4.1.0), processed using Adobe Photoshop CS6 and assembled into figures with Adobe Illustrator CS6.

Fluorescence and morphology parameters [area, volume, oblate ellipticity, prolate ellipticity, sphericity and LBR (green)/Hoechst (blue) mean fluorescence intensity], as determined by Imaris (Bitplane AG, v 9.2.1), were compared in four independent experiments for a total of 103 cells (n=55 negative control treatment, n=48 TMEM147 siRNA treatment) in stacks acquired on a Leica TCS SP2 DMIRE2 confocal microscope, using identical acquisition settings across samples. MANOVA using Pillai's test and α=0.05 (Addinsoft XLSTAT) was used to calculate the statistical difference between treatments (negative control and TMEM147-silencing), across all parameters. To assess statistical significance, individual parameters were analyzed by Mann–Whitney U-test (GraphPad Prism 8.0) with P-value adjustment using the FDR approach by the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli (Q=5%).

3D rendering of serial confocal z-stacks of HeLa nuclei (from the same control and TMEM147-silenced cells as above) was also performed with Imaris. 3D reconstruction of each nucleus was rendered using the ‘new surface’ algorithm with surface detail magnitude set at 0.250 μm and manual thresholding until voxels covered the entire surface of the nucleus according to LBR staining. Snapshots of 3D images were acquired to compare the localization of LBR in control and TMEM147-silenced cells. Elevated occurrence of ER localization for LBR labeling was tested by two-sided Fisher's exact test in Prism 8.0 (GraphPad) for the total of four independent experiments.

For lipid analysis, 3×10⁴–5×10⁴ cells were used per condition. Acidic extractions were performed as previously described (Özbalci et al., 2013), in the presence of an internal lipid standard mix containing 50 pmol phosphatidylcholine (13:0/13:0, 14:0/14:0, 20:0/20:0; 21:0/21:0; Avanti Polar Lipids), 50 pmol sphingomyelin (d18:1 with N-acylated 13:0, 17:0, 25:0), 100 pmol D6-cholesterol (Cambridge Isotope Laboratory), 25 pmol phosphatidylinositol (16:0/16:0; Avanti Polar Lipids), 25 pmol phosphatidylethanolamine and 25 pmol phosphatidylserine (both 14:1/14:1, 20:1/20:1, 22:1/22:1), 25 pmol diacylglycerol (17:0/17:0; Larodan), 25 pmol cholesteryl ester (9:0, 19:0; Sigma), 24 pmol triacylglycerol (D5-Mix, LM-6000/D5-17:0/17:1/17:1; Avanti Polar Lipids), 5 pmol ceramide and 5 pmol glucosylceramide (both d18:1 with N-acylated 15:0, 17:0, 25:0), 5 pmol lactosylceramide (d18:1 with N-acylated C12 fatty acid), 10 pmol phosphatidic acid (21:0/22:6; Avanti Polar Lipids), 10 pmol phosphatidylglycerol (14:1/14:1, 20:1/20:1, 22:1/22:1) and 5 pmol lyso-phosphatidylcholine (17:1; Avanti Polar Lipids). Neutral extractions were performed in the presence of a phosphatidylethanolamine plasmalogen (PE P-)-containing standard mix, supplemented with 22 pmol PE P-Mix 1 (16:0p/15:0, 16:0p/19:0, 16:0p/25:0), 31 pmol PE P-Mix 2 (18:0p/15:0, 18:0p/19:0, 18:0p/25:0) and 43 pmol PE P-Mix 3 (18:1p/15:0, 18:1p/19:0, 18:1p/25:0). Lipid extracts were resuspended in 60 µl methanol, 2 µl-aliquots were diluted 1:10 in 96-well plates in methanol, and ammonium acetate was added to a final concentration of 10 mM. Samples were analyzed on a SCIEX QTRAP 6500+ mass spectrometer (Sciex, Canada) with chip-based (HD-D ESI Chip, Advion Biosciences, USA) nano-electrospray infusion and ionization via a Triversa Nanomate (Advion Biosciences, Ithaca, USA), employing precursor ion or neutral loss scanning as described (Özbalci et al., 2013). The remaining sample was evaporated and free cholesterol was subjected to acetylation and MS analysis as described previously (Özbalci et al., 2013). Data evaluation was undertaken using LipidView (ABSciex) and an in-house-developed software (ShinyLipids). Statistical evaluation of comparisons was assessed by multiple t-test with multiple testing correction, based on an FDR of <0.05 (Fig. 7).

For lipidomics, multivariate statistics analyses were performed in MetaboAnalyst version 4.0. All variables were log transformed and scaled by performing mean-centering and division by the standard deviation of each sample. All variables were then subjected to an unsupervised principal component analysis (PCA), generating a scores plot showing the variance in the dataset. The loadings plot for the generated model was used to determine which variables drive the separation between classes. The most discriminant variables were visualized using univariate statistics (Fig. S6).

For univariate statistics, GraphPad Prism 8.0 software was used. All data were expressed as mean±s.e.m. In GraphPad, one- or two-way ANOVA was performed where appropriate. For one-way ANOVA, Dunnett's post hoc multiple-comparisons test was performed, while for two-way ANOVA, Šidák's post hoc multiple-comparisons test was used (Fig. S6).

Cellular uptake of fluorescently tagged cholestrol was measured in three independent experiments using the Cholesterol Uptake Assay Kit (ab236212; Abcam) in a 96-well format, following the manufacturer's specifications. Briefly, cells were silenced for 48 h, subsequently incubated for a further 24 h with fluorescently tagged NBD-cholesterol at 20 μg/ml in either complete or serum-free medium or in the presence of cholesterol transport inhibitor U-18666A at 1 μM (as a positive control of increased retained cholesterol). NBD-cholesterol is an established probe for examining lipoprotein-mediated cholesterol uptake in vivo and in cultured cells (Frolov et al., 2000; Huang et al., 2015). Fluorescence emission readings with an FITC/GFP filter set were normalized to protein concentrations, determined for the same wells with bicinchonic acid (BCA) assays, followed by subtraction of background fluorescence. To assess statistical significance in uptake assays, multiple t-tests with multiple testing correction were performed (GraphPad Prism 8.0), using one unpaired t-test per row without assuming a consistent s.d. and applying the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli (Q=5%) to determine discovery (Fig. 7).

For the experiments shown in Fig. 7F,G, cells were seeded in multiple 24-well plates (15×10³/well), grown in parallel either with control or TMEM147 silencing and sampled by trypsination and counting in triplicate at time points 24 h, 48 h and 72 h post silencing. For each silencing regime, cells were either grown in complete medium for 72 h, or in complete medium for 48 h and under serum starvation until 72 h, or in complete medium for 48 h and under serum starvation until 72 h with the addition of exogenous cholesterol (20 μg/ml) at 64 h (i.e. for the last 8 h before sampling). Three independent experiments were performed and statistical evaluation was assessed for parametric (medium) comparisons after confirming normality with the Shapiro–Wilk test, with one-way ANOVA based on the two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli (Q=5%) to determine discovery (GraphPad Prism 8.0).

We are indebted to Prof. Iain W. Mattaj, in whose lab at the EMBL (Heidelberg) this project was initiated, for his kind and sustained support. We are grateful to Dr Chrysoula Pitsouli (University of Cyprus) for granting us access to a Leica confocal microscope SP2 and to Dr Jan Ellenberg (EMBL, Heidelberg) for providing the HeLa-LBR238-GFP cell line. We are thankful to Drs Mandy Rettel and Frank Stein for their expert assistance at the EMBL Proteomic Facility and Ms Anastasia Raoukka (University of Cyprus) for background work on TMEM147.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.245357.reviewer-comments.pdf