The ins and outs of endoplasmic reticulum‐controlled lipid biosynthesis

Membrane composition and PtdCho synthesis

PtdCho is the most abundant lipid of mammalian membranes, and its synthesis is central to membrane biogenesis. The CCTα enzyme (human gene: PCYT1A) is typically rate‐limiting for PtdCho production in animal cells (Fig 1A). CCTα cycles between a soluble and membrane‐bound state (or onto the single lipid leaflet surrounding lipid droplets 44) and is subject to auto‐inhibition mediated by a carboxy‐terminal region termed the M‐domain 44, 45. When CCTα is soluble, the M‐domain inhibits the catalytic domain and prevents CDP–choline synthesis. However, this inhibition is relieved when the M‐domain folds into an amphipathic helix and absorbs onto a lipid surface (Fig 2C). Thus, CCTα membrane binding represents an on–off switch regulating catalytic activity and therefore PtdCho production.

Whether domains such as the M‐domain interact with membranes to fold into a helix depends on the biophysical properties of available membrane surfaces. One critical factor is that membrane binding is typically promoted by looser lipid packing and voids between lipid headgroups (Fig 2B). Such packing voids are more frequent when membranes contain a higher percentage of conical lipids compared with cylindrical lipids (Fig 2B) 38, 44, 46. While this can be a relatively non‐specific activation mechanism, in the case of the ER it tunes CCTα activity to PtdCho levels since these are the main cylindrical, bilayer‐stabilizing lipids of mammalian membranes. Thus, CCTα membrane binding and activity increase when PtdCho levels drop relative to the conically shaped PtdEtn. Conversely, a high concentration of PtdCho reduces CCTα membrane binding and activity.

Amphipathic helix integration into a membrane is also influenced by the surface charge of a membrane, via interactions between the protein and charged lipid headgroups. Typically, this involves negatively charged lipids that further stabilize the membrane–protein binding interaction. While phosphorylated PtdIns species are centrally important for how such proteins associate with most membranes, these lipids are rare in the ER where instead PtdA often promotes protein–membrane binding, including that of CCTα 47 (Fig 2C). This inherently matches CDP–choline production to the amount of PtdA being produced at the beginning of the GPL/GL network (Fig 1A). Interestingly, there are also several other mechanisms by which PtdA sensitivity further tunes CCTα activity to detect changes in ER membrane composition. One is that the small charged headgroup of PtdA is typically more accessible for protein binding interactions when membranes contain more conical lipids (e.g., PtdEtn) than cylindrical PtdCho (Fig 2B). This means again that the membrane binding and activity of CCTα increase when PtdCho levels drop. In addition, PtdA exists in mono‐ and divalent forms, and proteins typically interact more strongly with the greater charge of divalent PtdA. In turn, several factors regulate the relative abundance of mono‐ and divalent PtdA that intriguingly includes that the primary amine headgroup of PtdEtn promotes divalent PtdA through hydrogen bonding interactions 48. Thus, although unproven in vivo, a membrane lipid imbalance with high PtdEtn levels relative to PtdCho can potentially increase CCTα activity via elevating divalently charged PtdA, and thus counteract the imbalance.

Through these mechanisms, the CCTα M‐domain represents an elegant means by which a water‐soluble step of lipid synthesis is regulated by the end product lipid, even though this is packed within a lipid bilayer. The membrane binding of CCTα also shows how a lipid‐synthesizing activity is “tuned” to the composition of the ER membranes, and thus inherently corrects against membrane disequilibrium. Furthermore, once bound into the membrane, the amphipathic helix of CCTα is exposed to the surrounding lipid environment and may undergo further conformational adaptation—such as shown for the lipid packaging sensor Mga2 and inositol‐requiring enzyme (IRE1) 49, 50—to further influence CCTα activity and PtdCho production.

Sterol regulatory element‐binding proteins (SREBPs)

The SREBP system also allows lipid species in the ER membrane to feedback on their own biosynthesis. SREBPs are transmembrane ER proteins that were first identified as players in mammalian cholesterol synthesis. Mammals have three SREBP gene products: SREBP‐1A, SREBP‐1C, and SREBP‐2 (Fig 2D). All three have a cytosolic N‐terminal domain that is a transcription factor of the basic helix‐loop‐helix leucine zipper family. This domain activates gene transcription when it binds to sterol response elements (SREs) in promoters. However, the domain is held inactive and unable to bind DNA while attached to the transmembrane C‐terminus of SREBP. The proteases responsible for SREBP cleavage are localized in the Golgi, while under basal conditions SREBP is retained in the ER via interactions with SCAP (SREBP cleavage‐activating protein) and the ER‐resident protein, Insig (insulin‐induced gene) (Fig 2D) 13. The system is sensitive to the membrane lipid environment as it influences SCAP conformation 51, 52. In the classic cholesterol feedback model, low cholesterol in the ER membrane causes SCAP to alter its conformation and release from Insig. SREBP/SCAP then escape the ER in secretory pathway vesicles, arrive in the Golgi, and undergo proteolytic processing that releases the active N‐terminus of SREBP into the cytosol from where it can traffic to the nucleus and access SREs (Fig 2D) 53. The importance of this mechanism for cholesterol synthesis is supported by experiments showing that forced expression of the mature isolated N‐terminal domain of mammalian SREBP‐2 causes a 75‐fold increase in the levels of HMGCR mRNA that encodes the rate‐limiting enzyme of cholesterol synthesis. This led to a ~30‐fold increase in cholesterol synthesis 54. Additionally, ER membranes contain particularly low amounts of cholesterol (Fig 2A). This means that even small molar changes in cholesterol levels represent a relatively large percentage change, and thus can trigger SCAP/Insig dissociation 55. This mechanism of SREBP activation is therefore another example where ER membrane composition is a central regulator of cellular lipid synthesis.

Although it was first described in cholesterol regulation, the SREBP system is not specific to cholesterol synthesis; 1,003 different genes show statistically significant expression changes in the liver of mice with forced expression of the N‐terminus of SREBP‐1A, while 505 are affected by SREBP‐2 56. Not all genes are directly affected, however, and the overall picture from many studies of SREBP‐1 and SREBP‐2 signaling is that SREBP‐2 has the classic, originally described role in cholesterol synthesis, while SREBP‐1 isoforms predominantly upregulate genes of fatty acid and GPL/GL metabolism, particularly TAG synthesis. This includes that SREBP‐1 isoforms elevate the expression of fatty acid synthase, fatty acid elongase, and GPAT enzymes to upregulate flux into the GPL/GL network. The general importance of SREBP signaling is also clear when considering non‐mammalian organisms. Caenorhabditis elegans and Drosophila do not synthesize cholesterol, but still have SREBP signaling. In these organisms, SREBP activation specifically affects GPL/GL synthesis. Indeed, C. elegans SBP‐1 is activated by low PtdCho, as well as by abnormal ratios of PtdA and DAG 57, 58, while Drosophila SREBP is activated by low PtdEtn 59. In both organisms, SREBP is an essential positive regulator of membrane lipid biogenesis 59, 60, 61. Thus, although SREBP was named for its role in mammalian cholesterol homeostasis, this may be a relatively recent evolutionary adaptation of a general membrane regulator. The next steps will be to better understand the disturbances in membrane composition that activate the different SREBPs. In addition, as we explain below, it is important to fully catalog the molecular mechanisms of activation given that SREBP‐1, in particular, contributes to cellular programs such as cell growth and TAG lipogenesis, and has been implicated in the proliferative capacity of cancer cells 62.

The unfolded protein response (UPR); lipids as well as unfolded proteins

The UPR maintains ER protein homeostasis in response to the accumulation of misfolded proteins in the ER lumen 63, 64, 65, 66, 67, 68. Higher eukaryotes have three UPR branches mediated by different transmembrane sensors: (i) IRE1 which signals through XBP1 (Fig 2E), (ii) ATF6, and (iii) PERK that acts via elF‐2α. All three branches are activated by misfolded proteins in the ER lumen and classically signal to slow translation, upregulate chaperone expression, and/or induce apoptosis 69. Additionally, as we describe here, they also strongly affect lipid metabolism.

A major link between the UPR, proteopathic defects, and lipid synthesis is the importance of ER volume for resolving ER dysfunction. In fact, increasing ER volume helps rescue luminal protein misfolding independent of factors such as chaperone levels 70. In turn, increased ER volume is synonymous with increased membrane, and thus increased membrane lipid production. The IRE1/XBP1(S) branch, in particular, is a well‐characterized positive regulator of membrane lipid synthesis. Forced XBP1(S) expression is sufficient to increase the size of the ER 71, 72, while genetic deletion of the Xbp1 gene reduces ER size 73. The ER membrane expansion induced by XBP1(S) is concomitant with elevated PtdCho and PtdEtn biosynthesis 71, and increased expression of lipid synthesis genes including several of fatty acid synthesis, LIPIN and AGPAT isoforms 71, 74. The specific mediator between IRE1/XBP1(S) and PtdCho and PtdEtn synthesis has been difficult to pin down. It appears to involve increased CCTα activity but, surprisingly, given that IRE1/XBP1(S) is primarily a transcriptional mechanism, PCYT1A/CCTα mRNA and protein are relatively unaltered 71, 74. There is however evidence that CCTα might be activated via changes to the ER membrane environment 74, such as increased PtdA levels that elevate CCTα membrane binding.

ATF6 and PERK also affect lipid biogenesis. Some of their reported effects may be indirect given that, at least in some conditions, ATF6 promotes IRE1 and XBP1 expression alongside increasing membrane production 75, 76. On the other hand, forced ATF6 signaling can drive de novo membrane lipid synthesis to the same extent as XBP1(S), and independent of XBP1(S) 77. In addition, ATF6 signaling (rather than XBP1(S)) underlies the lipid synthesis and ER proliferation that occurs when the ER is overloaded with membrane proteins 78. The lipid enzymes targeted by ATF6 also differ from those of XBP1(S), further supporting that this UPR branch has its own (currently less defined) role in promoting membrane lipid production.

The UPR has been primarily considered as a protein homeostatic mechanism. However, there are now substantial data challenging the concept that UPR control of lipid biosynthesis is a corollary to its role in protein homeostasis. Indeed, not only does the UPR control cellular lipid levels, but UPR sensors are activated by abnormal ER membrane composition including abnormal lipid saturation and altered sterols 79, 80, 81, 82. Furthermore, it is clear that this indeed occurs because UPR sensors are directly influenced by their surrounding membrane, and not just by the presence of luminal unfolded proteins. Excess saturated lipids activate IRE1 or PERK even when these sensors lack the luminal regions that detect misfolded proteins 82. Similarly, mutant yeast IRE1 that cannot detect misfolded proteins nevertheless mediates UPR activation in cells with abnormal lipid metabolism 83. Further, IRE1 in C. elegans is activated by disturbed lipid metabolism even when this fails to induce protein misfolding, and in a manner that is insensitive to chemical chaperones that rescue protein misfolding 79.

Yeast studies also provided further insight into the molecular basis of how UPR proteins sense membrane composition 49. Just as for unfolded proteins, IRE1 activation in response to abnormal ER membrane composition occurs via the formation of higher order IRE1 oligomers that splice an mRNA (XBP1 or Hac1 in yeast) to allow translation of a full‐length soluble transcription factor (XBP1(S)/HAC1) that mediates many downstream events of the UPR (Fig 2E). Uniquely, however, lipid‐driven UPR activation requires an amphipathic helix in IRE1 that is embedded in the luminal face of the ER membrane and lies adjacent to the transmembrane section of the sensor. This juxtamembrane amphipathic helix is also evolutionarily conserved in animal IRE1 proteins, as well as in the PERK sensor. Molecular dynamic simulations point to this helix driving IRE1 clustering if ER membrane thickness and/or lipid packing density increase; notably, the ER membrane is particularly thin and loosely packed under normal conditions. This means that IRE1 activation is sensitized to imbalances in the levels of different GPL classes, as well as GPL saturation, given how these both impact membrane packing (Fig 2B). Additionally, it likely means that the UPR is sensitized to changes in the normally low levels of ER membrane sphingolipids and cholesterol given that these also promote thicker and more densely packed membranes.

It is therefore now clear that the UPR detects and solves defects in the lipid composition of ER membranes, just as it does detect proteomic defects within the ER lumen. Additionally, the two different insults can act additively since they drive IRE1 dimerization through distinct molecular mechanisms—thus establishing a new model where abnormal ER membrane composition increases the likelihood that unfolded proteins in the ER lumen activate the UPR, and vice versa 49.

Lipin PtdA‐phosphatase activity at the center of GPL/GL biosynthesis

PtdA is the basic lipid structure from which other GPL/GL derive. It lies at the branch point between the pathways that produce PtdIns versus PtdCho/PtdEtn. PtdA conversion to DAG further leads to TAG production (Fig 1A). PtdA is also a signaling lipid that is detected by many peripheral membrane proteins, and, for example, helps to activate CCTα to drive membrane production, as described above. Consequently, enzymes that control PtdA levels exert multiple layers of control over the GPL/GL synthesis network.

In eukaryotes from yeast to plants to animals, the lipin PtdA‐phosphatases influence ER‐localized PtdA levels 84 (note that a distinct set of lipid phosphatases (LPP enzymes) hydrolyze PtdA in other organelles 85). Lipins hydrolyze the phosphate headgroup of PtdA to produce DAG (Box 1 Fig, panel A) with high specificity for PtdA above other phosphorylated lipids 86, 87. As mentioned above, animal lipins also directly influence lipogenic transcriptional programs through their C‐terminal domain that interacts with transcriptional regulators and in a PtdA‐phosphatase‐independent manner 88. However, how this role integrates with lipin's role in de novo lipid synthesis remains poorly understood and is not further considered here.

Despite having a substrate and product in the lipid bilayer, lipins are not integral membrane proteins and must associate with membranes to access their substrate (Fig 2F). Just as for CCTα, the regulation of lipin membrane binding has evolved as a central regulator of its catalytic activity. Lipin associates with the ER membrane through an amino‐terminal amphipathic helix, and this event is especially sensitive to PtdA levels 89. The addition of PtdA increases yeast lipin (pah‐1) binding to micelles by five‐ to sixfold 90. Inclusion of PtdA also strongly promotes the otherwise weak binding of lipin1 to synthetic liposomes even when these have a mix of PtdCho/PtdEtn that mimics the ER membrane composition 91. PtdA sensing is mediated through a series of basic amino acids that are downstream of the membrane binding domain of lipin 26, 91, and it will be interesting to know how these two domains interact, as well as their relationship to the C‐terminal catalytic region—information that will be gained when the structure of the lipin enzyme is solved.

Lipin is heavily phosphorylated under many conditions, and this regulates its PtdA‐phosphatase activity (Fig 2F). There are more than 17 distinct serine/threonine phosphorylations recorded for mammalian lipin1, 15 in lipin2, and a similar number for yeast pah‐1/lipin 27, 92, 93, 94. The roles of individual phosphorylation sites have been difficult to pin down, even in yeast where these are fewer and there is more advanced understanding of lipin biochemistry. However, phosphorylation is uniformly associated with reduced PtdA‐phosphatase activity, leading to an experimentally well‐supported model that dephosphorylated lipin proteins are active PtdA‐phosphatases, while highly phosphorylated lipin is inactive. It is also clear that phosphorylation inhibits PtdA‐phosphatase activity (at least in part) because it reduces membrane binding and/or PtdA detection (Fig 2F) 91, 93.

Lipin phosphorylation and dephosphorylation also control whether lipin localizes in the nucleus or cytosol (Fig 1B (III)). Indeed, all lipins carry a NLS, and there is a consistent relationship between dephosphorylated lipin, nuclear localization, increased membrane binding, and elevated PtdA‐phosphatase activity. Highly phosphorylated lipin is cytosolic, not membrane‐bound, and has less activity. It is difficult to understand why nuclear localization correlates with membrane binding and more PtdA‐phosphatase activity, although we note that nuclear‐localized lipins can access PtdA within the nucleoplasmic face of the inner nuclear membrane. One interesting possibility is that the inner nuclear membrane localization relates to PtdA‐dependent crosstalk between lipin and CCTα. This negative feedback between lipin metabolism of PtdA and reduced CCTα activity has been demonstrated in yeast, plant, and fly. Both lipin and CCTα carry an NLS and both often concentrate in the nucleus, and thus will cycle to and from the inner nuclear membrane (Fig 1B (III)). Perhaps concentrating both enzymes to the inner nuclear membrane amplifies how efficiently lipin regulates CCTα, and data from fly indeed show lipin regulating whether or not CCT1 interacts with the inner nuclear membrane 28. Furthermore, segregating the lipin/CCTα crosstalk may prevent altered PtdA levels from impacting the many other proteins that bind and/or are regulated by PtdA 95, 96.

There are still many unknowns about how lipin is regulated. The interplay between multiple phosphorylations, sensitivity to membrane composition, and partitioning between the nucleus and cytosol have made it difficult to dissect the roles of each event. The conserved CTDNEP1/NEP1R1 phosphatase complex (Nem1 and Spo7 in yeast) is a centrally important lipin activator in yeast and animals that often specifically localizes in the inner nuclear membrane 97, 98, 99. As we discuss below, lipin phosphorylation is also a target of insulin and mTOR signaling. In addition, torsin ATPases, which reside inside the ER lumen, were recently identified as lipin inhibitors, and are also implicated as sensors of the ER redox state 28, 100, 101. Thus, lipin is positioned to integrate multiple types of information and couple these to lipid flux through the GPL/GL metabolic network.

Cell growth, growth factors, and mTORC1

Cell growth is controlled by external information such as hormones and nutrients, integrated with internal states relating to cellular energy and stress. Cell growth is an essential element for many aspects of animal development and physiology, while the dysregulation of cell growth is pathogenic in the context of cancer. Cell growth inherently requires increased cellular biomass, which includes membrane and thus increased membrane lipid production. Consistently, lipid production is a target of cell growth signaling pathways under normal as well as pathological conditions.

The evolutionarily conserved mTOR serine/threonine kinase is a master integrator of growth information and coordinates the downstream processes that increase cellular biomass. This includes signaling for the increased production of lipids 102. There are two distinct mTOR kinase‐active complexes, mTORC1 and mTORC2. mTORC1 is inhibited by rapamycin, activated by pro‐growth PI3K and Akt pathways, and is the complex that most often couples pro‐growth signaling to lipid biogenesis 102. Indeed, the majority of events where nutrients alter lipid production are either blocked by rapamycin and/or stimulated by genetic knockout of the TSC1/2 upstream inhibitors of mTORC1. This means that mTORC1 is central to how pro‐growth signals increase the synthesis of PtdCho and PtdEtn membrane lipids 103, 104.

Pro‐growth signaling/mTORC1 regulates lipid production through several mechanisms, of which SREBP‐1 activation is the most consistently important (Fig 3A). Indeed, this fundamental conserved mechanism has been described driving cell growth in invertebrate and vertebrate species 103. The mechanism of how mTORC1/pro‐growth affects SREBP‐1 has been most studied in liver cells, a cell type under particular scrutiny due to its central role in nutrient responses 102 and which has recently been discussed in detail 105. In brief, coupling to SREBP‐1 typically requires that mTORC1 activates its classic ribosomal S‐6‐kinase substrate, and impacts SREBP‐1 at multiple levels including elevating SREBP‐1 cleavage, increasing the half‐life of cleaved mature SREBP‐1, and upregulating SREBP‐1 mRNA expression at least in part because the SREBP‐1 promoter contains a SRE and thus responds to mature SREBP‐1 protein 106. How pro‐growth signals promote SREBP cleavage from the membrane remains relatively poorly understood, including whether/how ER membrane composition plays a role. One interesting newer discovery is that increased SCAP glycosylation decreases its binding to Insig‐1, and increases SREBP traffic to the Golgi for cleavage (Fig 2D). This SCAP glycosylation is sensitive to cellular glucose levels, which, in turn, are regulated by pro‐growth signals 62. This provides a mechanism whereby SREBP cleavage is upregulated by a signaling cascade, although this particular mechanism appears to be mTORC1‐independent. Overall, there is much to uncover about the coupling of pro‐growth pathways to SREBP‐1, including how this is adapted in different cell types, how abnormal activation of SREBP‐1 helps fuel oncogenic cell growth, and whether SREBP‐1 may represent a useful chemotherapy target 62, 107.

Pro‐growth signaling/mTORC1 also controls lipid production via lipin phosphorylation that reduces its PtdA‐phosphatase activity 91, 108. Again, this is a conserved mechanism with similar events described in yeast, fly, plant, and mammalian cells. A highly conserved serine residue (position 106 in mammalian lipin1) is targeted by insulin and mTOR 93, although mTORC1 activation goes beyond this and 17 distinct mTOR‐sensitive serine/threonine phosphorylations have been described in mammalian lipin1 27. How lipin inhibition contributes to mTOR‐driven cell growth is not yet fully explained. It may relate to the feedback between PtdA and CCTα 28, 47, whereby lower lipin PtdA‐phosphatase activity allows PtdA levels to increase, thus activating CCTα and driving membrane GPL production (Fig 3B). In addition, PtdA is itself an mTORC1 activator via a PtdA‐binding motif in the same mTORC1 domain that rapamycin binds 95, 109, 110, possibly meaning that lipin inhibition amplifies mTORC1 activation.

There is an additional interaction between lipin and the SREBP branch of the cell growth response. mTOR‐driven lipin phosphorylation causes lipin to localize in the cytosol. Conversely, mTOR inhibition promotes a nuclear localization and thus lipin access to the inner nuclear membrane (Figs 1B (III) and 3B). When mTOR is off, the PtdA‐phosphatase activity of lipin modulates nuclear morphology and sequesters cleaved SREBP at the nuclear periphery to prevent SREBP from activating gene transcription. Conversely, mTOR‐dependent lipin phosphorylation overcomes this inhibitory control over SREBP (Fig 3B) 27. The in vivo relevance of this inhibitory interaction between lipin and SREBP is shown by genetic knockout of an ubiquitin ligase that targets phosphorylated lipin for degradation. Its loss elevates lipin levels in parallel with reducing SREBP target gene expression and lipid synthesis in liver cells 111. Interestingly, lipin phosphorylation driven by pro‐growth signaling is reversed by the CTDNEP1 phosphatase that resides at the nuclear periphery, and pro‐growth signals in yeast also control the lipin phosphorylation state via this phosphatase (Fig 3B) 112.

Tailoring lipid synthesis to cell‐type‐specific functions

Lipid synthesis is regulated by differentiation programs concomitant with how they alter the size and morphology of cells, and/or the volume of internal organelles. In addition, organelle scaling occurs in response to environmental cues, for example, when hepatocytes expand the size of the ER to house detoxifying cytochrome P450 enzymes 113, or in professional energy‐processing cells that alter TAG production according to nutritional cues (Fig 3A).

The UPR machinery, especially the IRE1/XBP1(S) branch, is the main driver of lipid synthesis coupled to changes in ER and secretory pathway volume (Fig 3A). XBP1 is required in secretory cells, such as pancreatic acinar cells and salivary gland cells, and XBP1 deletion causes lethal phenotypes in mice because these cells fail to achieve fully differentiated states that support high‐level protein secretion 73. XBP1 deletion also inhibits the ER membrane expansion that occurs when B cells differentiate into antibody‐secreting plasma cells, while forced XBP1(S) expression is sufficient to induce ER expansion and differentiation of this cell type 114. Thus, although IRE1/XBP1(S) were first described in the context of a pathological response to protein misfolding, this pathway also promotes membrane lipid synthesis during normal differentiation.

Another well‐characterized scaling of lipid synthesis occurs with the synthetic pathway leading to TAG (fatty acid synthesis, GPAT, AGPAT, lipin, and DGAT enzymes; Fig 1A). It is particularly active in cell types that convert nutrients into energy stores, such as mammalian adipocytes and hepatocytes. In addition, environmental signals further refine TAG pathway activity, and there are important roles for the UPR and SREBP regulatory mechanisms in TAG production (Fig 3A). Here, again, IRE1/XBP1(S) play roles. Conditional deletion of XBP1 from mature mouse hepatocytes decreases flux into the GPL/GL network (detected by [14C]‐acetate incorporation into lipids) concomitant with reduced expression of fatty acid synthesis and Dgat2 genes. Hepatocytes that lack XBP1 are also less able to upregulate lipid synthesis in response to a high‐carbohydrate diet 115. This XBP1‐mediated control over hepatic lipogenesis appears to occur via IRE1‐dependent and IRE1‐independent mediated mechanisms that overall remain poorly described 115. In addition, TAG synthesis in hepatocytes is further promoted by insulin signaling acting via SREBP‐1. Mouse studies have shown that hepatocytes lacking SREBP‐1 release less TAG into the circulation, while forced SREBP‐1 (or SREBP‐2) expression leads to excess TAG deposition 116.

The mechanism(s) that allow SREBP‐1 activation to elevate TAG synthesis in this context, while the same pathway promotes cell growth in other cells/situations, remain poorly defined. In addition, there are also similar questions about how the XBP1(S) pathway has differing effects in different contexts. Since both systems are transcriptional, they may target different gene sets in different cell types 117, although this is an incomplete explanation since gene expression changes usually correlate poorly with altered lipid production. It may relate to differences in whether SREBPs and XBP1(S) are activated by factors such as insulin signaling, versus activated by ER membrane composition. For example, ER membrane composition will also modify the final outcome of XBP1(S) and SREBP signaling by repressing or augmenting the activity of newly expressed lipid synthetic enzymes via mechanisms described above (Fig 2C–F). Additionally, ER membrane composition may impact on the gene targets of XBP1(S) or SREBP‐1. A conceptual example of this comes from lipin. As explained above, lipin PtdA‐phosphatase activity [which, in turn, is affected by ER membrane composition (Fig 2F)] controls whether cleaved SREBP is sequestered to the nuclear periphery thus preventing its binding to SREs (Fig 3B) 27.

ER lipid synthesis in pathology

In addition to lipid synthesis, the ER stores calcium and processes newly translated secretory pathway proteins. These core ER functions are so tightly linked that defects in one typically cascade to impact the others; for example, reduced calcium often causes protein misfolding due to impaired function of calcium‐dependent chaperones, in turn activating the UPR and thus altering lipid synthesis. This means that ER dysfunction can drive altered lipid production downstream of different primary insults. In addition, there is strong evidence that abnormal lipid synthesis is itself the primary driver of ER dysfunction in some human metabolic and obesity‐related pathologies 118, 119, 120, 121, 122, 123, 124. These focus on hepatocytes that are functionally specialized to process nutrients into storage lipids, and also have large amounts of ER dedicated to secretory pathway functions.

Hepatocytes are a cell type that upregulates TAG synthesis in response to nutrients. The TAG synthetic pathway overlaps with that of membrane GPL/GL, diverging only at the final step catalyzed by DGAT (Fig 1A). Thus, when excess nutrients elevate flux into and through hepatocyte GPL/GL synthesis pathways, the network must buffer against spillover into non‐TAG branches, in order to maintain membrane composition. Indeed, under normal conditions, the fatty acyl chain profile of liver ER membrane lipids is unrelated to that of dietary lipids, showing segregation of membrane GPL and TAG production. However, the saturation profile of membrane GPL is altered in obese mice toward mono‐unsaturated acyl chains, suggesting that pathway buffering breaks down under conditions of excess TAG production 120. The same study also identified that hepatic ER membranes from obese animals contain excess PtdCho compared to PtdEtn. This occurred alongside elevated expression of CCTα, together with the PEMT enzyme (Fig 1A). The study then elegantly examined whether the abnormal ratio between PtdCho and PtdEtn had a role in the pathology of obese mice. This was done via Pemt knockdown, which halts PtdEtn to PtdCho interconversion and thus specifically reduces PtdCho relative to PtdEtn. This genetic manipulation not only corrected the PtdCho/PtdEtn imbalance, but also reduced ER dysfunction in the mice including restoring the function of the SERCA calcium pump that has a well‐characterized sensitivity to ER membrane composition 125 (Fig 3C).

This study and others 119, 120, 121, 122, 123, 124 have demonstrated that increased hepatic lipid synthesis is a primary driver of liver pathology in metabolic diseases related to excess food intake. Along these lines, genetic knockout of certain elongase and desaturase enzymes reduces obesity‐related phenotypes in mice, implicating abnormal acyl chain incorporation into lipids as harmful 126. Indeed, the interconnected nature of ER functions gives rise to a model where a vicious cycle of abnormal or excess lipid synthesis impacts the calcium and protein folding functions of the ER that, in turn, further amplify or alter lipid synthesis through membrane composition and/or UPR activation. There is also a strong correlation between hepatic ER pathology and the development of systemic metabolic defects such as hyperglycemia and insulin resistance, because of the central role of the liver in controlling circulating lipids. For example, the Pemt knockdown described above rescues systemic abnormalities in parallel with restoring normal hepatic ER membrane composition 120, clinical data correlate specific acyl chain TAG profiles with type 2 diabetes 127, and genetically reducing lipin degradation in mice (thereby allowing SREBP suppression) reduces high‐fat‐diet liver pathology and circulating insulin levels 111. Thus, abnormal ER lipid synthesis contributes to the pathology of obesity, which explains the clinical benefit of drugs that regulate lipid enzymes such as statins (HMGCR inhibitors) as well as inhibitors of TAG production 128, 129, 130, 131.