Structural Diversity within the Endoplasmic Reticulum—From the Microscale to the Nanoscale

Abstract

The endoplasmic reticulum (ER) is a continuous, highly dynamic membrane compartment that is crucial for numerous basic cellular functions. The ER stretches from the nuclear envelope to the outer periphery of all living eukaryotic cells. This ubiquitous organelle shows remarkable structural complexity, adopting a range of shapes, curvatures, and length scales. Canonically, the ER is thought to be composed of two simple membrane elements: sheets and tubules. However, recent advances in superresolution light microscopy and three-dimensional electron microscopy have revealed an astounding diversity of nanoscale ER structures, greatly expanding our view of ER organization. In this review, we describe these diverse ER structures, focusing on what is known of their regulation and associated functions in mammalian cells.

The endoplasmic reticulum (ER) is the largest membrane-bound compartment in eukaryotes, containing more than half of the total cellular membrane, but its structure is complex and highly dynamic (Phillips and Voeltz 2016), exploring more than 95% of the cytoplasm every 15 min (Valm et al. 2017). The ER's crucial functions include serving to translate and fold proteins, wrapping the nucleus, synthesizing lipids, and storing/releasing calcium (Staehelin 1997; Goyal and Blackstone 2013). In addition, the ER forms extensive contacts with neighboring organelles, coordinating and synchronizing biochemical processes at disparate sites throughout the cell (Scorrano et al. 2019).

Early electron microscopy (EM) studies examining specialized cell types identified two major ER structures, tubules and sheets, and proposed that these structures engaged in different biochemical functions (Porter et al. 1945; Porter and Kallman 1952; Porter 1953; Palade and Porter 1954; Palade 1955; Porter and Palade 1957). The ER sheets were identified as large, two-dimensional regions of nearly flat membrane with constant luminal spacing. Studded with ribosomes, the sheets, coined “rough ER,” were presumed to be involved in protein translation (Palade 1955). ER tubules were thought to be elongated cylindrical structures stretching across the cytoplasm. Because tubules were depleted of ribosomes in EM cross-sections, they were dubbed “smooth ER” and presumed to lack protein translation functions (Palay and Palade 1955; Porter and Palade 1957). Supporting this structure/function differentiation, subcellular fractionation studies revealed heavy and light fractions of ER, with the heavy fraction containing ribosomes and secretory folding intermediates and the light fraction replete with enzymes involved in lipid synthesis and detoxification (Claude 1946; Siekevitz and Palade 1958, 1959; Ernster et al. 1962). The combined ultrastructural and biochemical results suggested the ER is structurally and functionally organized into at least two subdomains—one devoted to protein biosynthesis (rough ER) and the other involved in alternate ER activities (smooth ER).

More recent studies employing genetic and advanced imaging approaches in cell cultures and tissues, as well as in vitro reconstitution of model membranes, have significantly expanded our understanding of structure and function in the ER. These studies have shed light on machinery involved in forming different ER domains and have shown the ER has many more unique structural morphologies than previously appreciated. Advanced imaging technologies, for example, revealed previously unseen, highly intricate ER networks of interlinking tubules and stacked sheets, and three-dimensional EM has identified that most ER tubules also have ribosomes docked on them (Terasaki et al. 2013; Nixon-Abell et al. 2016; Heinrich et al. 2021). Use of genetic mutations and physiological disruptions to perturb ER structure showed these treatments could have dramatic effects on ER functions ranging from protein folding and secretion to calcium regulation and signaling (Goyal and Blackstone 2013). In fact, ER structural defects often give rise to specific human pathologies (Blackstone 2018). Together, the wide-ranging results have suggested the ER consists of interconnected macroscopic and nanoscopic levels of structural organization, collectively impacting the ER's diverse functions.

Previous reviews on ER structure have generally focused on the role of ER membrane-localized shaping proteins (Shibata et al. 2009; English and Voeltz 2013a; Goyal and Blackstone 2013; Westrate et al. 2015; Phillips and Voeltz 2016; Schwarz and Blower 2016). In vitro assays using these proteins, many of which are evolutionarily conserved, can generate membrane shapes that are highly reminiscent of ER structures in cells (Wang and Rapoport 2019). However, it is becoming increasingly clear from whole-cell imaging strategies that numerous ER structures exist with more ambiguous or complex morphologies than are able to be easily reconstructed using in vitro assays. In this review, we describe the diversity of common ER nanostructures known to date, discussing their possible mechanisms of formation and potential tasks in the broader ER system.

ER DISTRIBUTION AND STRUCTURE SEEN AT THE GLOBAL CELLULAR LEVEL

Early light microscopy studies using fluorescently labeled lipids revealed that the ER is a single continuous membrane-bound compartment that stretches throughout the cell (Pagano et al. 1981, 1983; Terasaki et al. 1984). When viewed at the whole cell level with confocal or widefield microscopy, several distinct ER subregions are visible, correlating with the ER's distance from the nucleus (Fig. 1A–D). The first of these subregions is the nuclear envelope (NE), which wraps around the nucleoplasm to create a distinct environment from the rest of the cytosolic space. The NE is connected to the rest of the ER through narrow tubules that extend off the NE to merge with a surrounding dense network of membranes that comprises the ER's perinuclear region. Here, the ER is extensively associated with other perinuclear structures, including the centrosome, the Golgi apparatus, perinuclear mitochondria, and endolysosomal compartments. Historically, this perinuclear subregion has been associated with stacked ER sheets, but more recent high-resolution experiments suggest additional structural variations (see below). Surrounding the perinuclear ER is a transitional ER subregion, where the ER becomes sparser, and clustered structures of ER become apparent in light microscopy images (Fig. 1B). Still further out is the peripheral ER subregion, which exists in the thin perimeter areas of the cell. Here, the ER appears as an expansive network reaching to the farthest edge of the cell and is dominated by thin tubular structures connected at three-way junctions. Many cells have distinct variations in these ER subregions, with some cell types forming little transitional or peripheral ER (e.g., Fawcett 1966; van Anken et al. 2021). Other cells, like neurons, have peripheral ER that extend long distances into dendrites and axons, where they differentiate into specialized shapes to stabilize and support ER function (Gray 1959; Kuwajima et al. 2013; Yalçın et al. 2017; Terasaki 2018).

Figure 1.

Endoplasmic reticulum (ER) distribution at the cellular level. (A) Airyscan micrograph of an ER membrane marker (mEmerald-Sec61β) in an interphase COS-7 cell. Dashed lines delineate phenotypically distinct subregions of the ER network. (B) Radial crop, indicating fluorescence intensity of ER membrane across each subregion. (C) Tiled airyscan micrograph of ER distribution across several neighboring U2-OS cells. (D) 3D volume rendering of ER (Halo-Sec61β) in a COS-7 cell. (E) Airyscan micrograph indicating substantial overlap of ER (Halo-Sec61, orange) with microtubules (EMTB-3 × GFP, blue). (F) Airyscan micrograph of ER (Sec61β) in a central slice through a metaphase HeLa cell. Scale bars, 10 µm (A,B); 20 µm, (C); 5 µm (E,F).

Direct and indirect interactions of ER with the cytoskeleton represent one mechanism for controlling the overall distribution of ER in cells (Fig. 1E). For example, microtubules facilitate anterograde extension of ER to maintain peripheral ER (Waterman-Storer et al. 1995; Waterman-Storer and Salmon 1998). Simultaneously, inward moving actin flows likely draw ER back toward the nucleus, stabilizing the perinuclear ER (Terasaki and Reese 1994; Waterman-Storer and Salmon 1998; Lynch et al. 2011). Consequently, cytoskeletal perturbing agents like latrunculin A and nocodazole, which respectively disrupt actin filaments and microtubules, result in dramatic shifts in the distribution of ER throughout the cell (Terasaki and Reese 1994; Lu et al. 2009; Joensuu et al. 2014).

Several ER-cytoskeleton linkers have been identified, but the mechanism and relative importance of each component is only beginning to be understood. Many ER-cytoskeleton linking proteins play additional, cytoskeleton-independent roles in ER shaping, so uncoupling the two effects has been challenging. One recently demonstrated example is the case of three well-characterized sheet-stabilizing proteins: CLIMP63, p180 (RRBP1), and KTN (Savitz and Meyer 1990; Toyoshima et al. 1992; Klopfenstein et al. 1998). These proteins both bind to microtubules (Toyoshima et al. 1992; Klopfenstein et al. 1998; Ogawa-Goto et al. 2007) and stabilize the structure of ER sheets (Shibata et al. 2010), but the two processes are largely dependent on different domains of the proteins. Remarkably, these proteins bind microtubules in a manner dependent on tubulin's state of posttranslational modification (Zheng et al. 2022). Consequently, they redistribute ER and its associated organelles to perinuclear or peripheral regions based on how tubulin is modified within cells, independently of their sheet-stabilizing and/or membrane-cross-linking functions (Zheng et al. 2022). Thus, CLIMP63, p180, and KTN likely impact ER architecture through both their effects at the molecular scale on local ER shape and their effects on global ER distribution through direct cytoskeletal interactions.

Global ER architecture changes in response to shifting cell states or nutritional needs. In mitosis, for example, the ER undergoes dramatic rearrangements, with NE and peripheral ER subdomains disappearing and either ER sheets or tubular networks appearing that surround the mitotic spindle (Fig. 1F; Puhka et al. 2007, 2012; Lu et al. 2009; Moore et al. 2021). Conditions of nutrient deprivation also impact ER structure, with the perinuclear ER domain expanding and making extensive contacts with lysosomes, mitochondria, and lipid droplets (Valm et al. 2017). The mechanisms underlying these changes in ER structure and distribution are only beginning to be clarified.

DIVERSITY OF LOCAL ER STRUCTURE AT THE NANOSCALE

The global distribution of ER visualized in confocal and widefield microscopy is composed of numerous local ER structures at the nanoscale connected to one another, but the identification of these structures directly from light microscopy data is challenging. In the peripheral ER of cells cultured on glass, ER tubules and three-way junctions can usually be unequivocally identified due to the sparsity of other ER around them. However, this becomes much more challenging as cell shapes become more three dimensional in tissues or mixed cell preparations. Even under ideal conditions, ER tubules appear at least three times larger due to the blurring effect of diffraction-limited microscopes (Reilly and Obara 2021). This problem is exacerbated as ER structures become more concentrated, complex, or three dimensional—all of which happen as the ER transitions to its perinuclear form close to the center of the cell.

An explosion of new technologies in optical superresolution microscopy have empowered a change in our understanding of local ER membrane structures (Fig. 2A–E), far beyond the simple tubes and sheets described in early EM studies and more recent confocal/widefield microscopy work. Numerous studies have now identified structures in the cellular periphery that appear continuous by conventional light microscopy but have more complex structure when visualized with superresolution microscopy (e.g., Fig. 2A,B; Nixon-Abell et al. 2016; Gao et al. 2019; Schroeder et al. 2019; Sun et al. 2020). These structures can be primarily tubular in nature (Fig. 2D), primarily sheet-like in nature (Fig. 2B), or, most often, are indeterminate even with the increased resolving power of a superresolution technique (Fig. 2A,C). The higher the spatial resolution of the technique, the more complex the structures often appear (Fig. 2C,D; Nixon-Abell et al. 2016), suggesting that even the highest resolution superresolution techniques may still under report the complexity of local ER structure. Thus, even with the powerful advantages of some optical superresolution techniques, unequivocal identification of nanoscale ER structures currently requires accompanying EM (Fig. 2E).

Figure 2.

Superresolution microscopy of endoplasmic reticulum (ER) structure. (A) Widefield image of Halo-Sec61β in the periphery of a COS-7 cell (left). Inset region is shown by both widefield (center) and structured illumination microscopy (right). The enhanced resolution of SIM reveals previously unappreciated structural complexity. (B) Laser scanning confocal image of an ER luminal marker (Halo-KDEL) in a metaphase HeLa cell. Inset region is shown by both confocal and stimulated emission depletion (STED) microscopy. Improved resolution in the STED image reveals previously unresolvable stacked ER sheets. (C) 3D SIM of complex peripheral ER structures pseudo-colored by axial position. (D) Lattice light sheet (LLS) paint microscopy of the lipid-binding dye Bodipy-TR. Inset indicates a complex tubular ER structure. (E) FIB-SEM reconstruction of a 3D-ER matrix (white) with simulated confocal image below (orange). Scale bars, 10 µm, 1 µm, 1 µm (A); 5 µm, 500 nm, 500 nm (B); 10 µm (C); 5 µm, 1 µm (D).

The high spatial resolution afforded by EM and the high labeling density of traditional EM stains has made EM the workhorse for mapping ER nanostructure for many years. However, unambiguous identification of ER morphologies from traditional two-dimensional EM sections is also challenging. Imaged in two dimensions, three-dimensional ER structural complexity is often obscured or completely unappreciated. Additionally, the low specificity of standard EM stains can make some ER structures difficult to distinguish from other membrane-bound organelles in the cytoplasm, especially from a single section. Thus, three-dimensional EM has become the standard for nanoscale ER structural determination. Groundbreaking studies using laborious serial section or serial block face EM techniques have identified diverse ER structures, including helicoidal and stacked ER sheets (Terasaki et al. 2013), complex and variable structural rearrangements in mitosis (Puhka et al. 2007, 2012), and the remarkably thin, specialized ER tubules in axons of the descending corticospinal tract (Terasaki 2018). These studies have provided a framework for local ER structure determination, but they are limited by the extensive labor required and the inherent limitations in resolution in the third dimension mandated by taking physical sections.

More recently, newer 3D EM techniques have emerged that facilitate understanding of ER structural organization. Platinum replica EM has been effective in resolving the structure of the ER at its sites of interaction with the basal plasma membrane (Aggeler et al. 1983), although the mechanical unroofing required to carry out this technique limits its utility in resolving ER structures deeper in the cell. Small sections of ER can be resolved in cryo-electron tomography without the need for potentially perturbative staining steps, but the contrast of ER membrane is low due to the absence of counterstains. The small imaging volumes that are possible in this technique also make it challenging to infer cellular context (Hoffmann et al. 2021; Xu et al. 2021).

An approach that surmounts many of these limitations is focused ion beam-scanning EM (FIB-SEM). Recent increases in stability of the technique (Xu et al. 2017, 2021) and implementations using freeze substitution for labeling the sample have made it a viable tool for solving local ER structure over large cellular regions (Hoffman et al. 2020; Heinrich et al. 2021; Weigel et al. 2021). Although freeze substitution can lead to occasional warping artifacts, it greatly improves membrane structural integrity (Hicks et al. 1976; Keene and McDonald 1993; Sosinsky et al. 2008; Hoffman et al. 2020) and is compatible with full reconstruction of entire cells. Recent work has used this tool to image ER-bound ribosomes (Fig. 3A) and to describe the fine shape of some specific ER-associated substructures, like ER exit sites (Weigel et al. 2021) and organelle contact sites (Fig. 3B; Heinrich et al. 2021).

Figure 3.

Nanostructure of the endoplasmic reticulum (ER). (A) Location of ribosomes on planar and tubular ER (planar, green; tubular, blue). (Panel A reproduced from Heinrich et al. 2021 with permission from the authors and Springer Nature 2021.) (B) FIB-SEM reconstruction indicating specialized ER domains, including mitochondria-ER contact sites (red) and ER exit sites (ERES, green). (C–J) 3D EM reconstructions of tube-based (C–F) and sheet based (G–J) ER nanostructures. (Panel H is adapted from Terasaki et al. 2013 with permission from Elsevier 2013.) Structures are not shown to scale.

Collectively, the use of three-dimensional EM is greatly expanding the repertoire of known local ER structures. There are at least four subclasses of ER sheets now described, and ER tubules and junctions can cluster at varying densities with highly variable membrane shape and properties (Fig. 3C–J). We review below these diverse local ER structures, discussing their characteristics and their potential functions within the broader ER system.

ER Tubules

ER tubules are the most prevalent component of the peripheral ER (Fig. 4A,B). These long, thin ER structures are pervasive throughout cells and can vary significantly in diameter between organisms (Porter 1953; Porter and Palade 1957; Stephenson and Hawes 1986; Terasaki 2018; Pain et al. 2019), between cell types (Terasaki 2018), and even within single cells (Wang et al. 2022). Tubules have been described to have diameters as large as 125 nm in some cultured animal cells (Hoffman et al. 2020; Heinrich et al. 2021) and as small as 10 nm or less within the plasmodesmata of plants (Fig. 4C; Wright and Oparka 2006). Biophysical studies of model membrane tubules in vitro have established that thin membrane tubules are relatively fragile structures due to their high positive curvature in one dimension (Roux et al. 2005). Unsurprisingly, thin ER tubules are susceptible to fixation-induced fragmentation (Fig. 4D) and in live cells have been shown to easily break and reform from the action of passing organelles (e.g., Guo et al. 2018). Given elongated ER tubules are nonetheless highly prevalent within cells, robust mechanisms for fusing together separated tubules and/or for stabilizing tubules to prevent fragmentation must exist.

Figure 4.

Endoplasmic reticulum (ER) tubules. (A) Laser scanning confocal image of peripheral ER tubules (Halo-Sec61β) in a live COS-7 cell. (B) FIB-SEM reconstruction of an isolated ER tubule. (C) Cartoon schematic indicating the variability of ER tubule width across different cell types. (D) Structured illumination micrographs of ER tubules in cells fixed with either 0.25% glutaraldehyde at 37°C (left) or methanol at −20°C. Methanol fixed tubules appear vesiculated. (E) Cartoon of schematized reticulon (blue) and REEP (green) family hairpin proteins inserted into a curved ER membrane. (F) Cartoon representing ER tubule pearling observed in 3D reconstructions from high pressure frozen or high-speed imaged cells. (G) Time-lapse montage of ER tubule extension in a live cell. (H) Cartoon indicating distinct mechanisms of ER tubulation/extension. (I) Grazing incidence-structured illumination microscopy (GI-SIM) of peripheral ER tubules, with kymographs at demarcated positions indicating tubule movement over time. Scale bars, 2 µm, 500 nm (A); 2 µm, 2 µm (D); 2 µm (G); 2.5 µm, 500 nm, time bar, 750 msec (I).

Several classes of evolutionarily conserved ER-localized proteins have been demonstrated to play a role in the stabilization of ER tubules. The most prevalent of these proteins are the canonical reticulon and DP1/REEP/Yop1p families of proteins, which have been demonstrated to promote curvature in vitro, in Xenopus egg extracts, and in cells (Fig. 4E; Craene et al. 2006; Voeltz et al. 2006; Hu et al. 2008, 2009). There is little sequence homology between these two families, but they have a shared structural signature comprised of an elongated hydrophobic region predicted to be primarily helical in nature, referred to as a hairpin. Because both termini of the hairpin are known to face the cytoplasm, the structure has been proposed to wedge into the outer leaflet, supporting positive curvature to help form and stabilize thin tubules (Voeltz et al. 2006; Tolley et al. 2008, 2010; Sparkes et al. 2010). Supporting this idea, overexpression experiments in plants and in vitro reconstitution studies have shown narrowing of tubule diameters when there are more hairpin-containing proteins in the membrane (Hu et al. 2008; Brady et al. 2015; Breeze et al. 2016; Wang et al. 2016, 2021). Additionally, regions of specialized ER highly enriched in hairpin-containing proteins, such as axonal ER, have been found to be remarkably thin in serial section EM reconstructions (Terasaki 2018).

ER tubules can often show long regions of essentially uniform diameter (Fig. 4A,B). However, several studies in high pressure frozen and freeze-substituted cells have described pearling ER tubule shapes (i.e., tubules appearing as beads on a string), especially in peripheral ER (Fig. 4F; Hoffman et al. 2020; Weigel et al. 2021). While it is unclear how this structure is formed, one possibility is that it arises from dynamic rearrangements of luminal content due to changes in osmotic balance or Ca²⁺ content in ER (Subramanian and Meyer 1997; Kucharz et al. 2011; King et al. 2020). Alternatively, the pearling shapes could arise from pulling forces exerted on ER tubules by microtubule motors (Bar-Ziv and Moses 1994; Weigel et al. 2021; Zucker and Kozlov 2022), or from ER shaping protein distribution (i.e., reticulons or DP1/REEP/Yop1p families of proteins), which may cause differential tubule narrowing (Holcman et al. 2018; Espadas et al. 2019). Use of superresolution microscopy has revealed that the traditionally sheet-associated protein CLIMP63 can also associate with regions of larger tubular diameter, while reticulons associate with narrower constrictions in the tubule, which could contribute to this phenomenon (Gao et al. 2019). Rapid rearrangements of pearling ER have been observed by high-speed imaging (York et al. 2013; Li et al. 2015; Nixon-Abell et al. 2016; Guo et al. 2018), so distinguishing among these possibilities may be tractable in the future, for example, by testing whether ER shaping proteins rearrange dynamically on tubules to stabilize inter-pearled regions.

ER tubules can be formed de novo from the existing ER network by the application of a pulling force to the ER membrane (Fig. 4G; Terasaki et al. 1984). Most of these tubulation events are associated with the microtubule cytoskeleton (Terasaki et al. 1986), and they have been described to occur by any of four major mechanisms (Fig. 4H). (1) Sliding tubule extensions occur along preexisting microtubules through the direct association of ER membrane proteins with a molecular motor. (2) Hitchhiking tubule extensions are caused by ER association with moving organelles, often lysosomes. (3) Nascent ER tubules can also form from the pulling force of microtubule polymerization. Proteins within the ER membrane associate with the polymerizing tip attachment complex (pTAC) so the ER tubule and the microtubule extend together. Many ER proteins have been implicated in this process, including STIM1 (Grigoriev et al. 2008; Smyth et al. 2012; Rodríguez-García et al. 2020), p22 (Andrade et al. 2004), ARL6IP1 (Dong et al. 2018), and TAOK2 (Nourbakhsh et al. 2021). (4) TAC-associated ER tubule growth can also occur at the depolymerizing end of microtubules undergoing catastrophe (dTAC), although the linkage machinery is not yet identified.

A recent study in COS7 cells quantified these four forms of tubule extension, finding that ∼20%–40% of tubule extensions occur via TAC or dTAC-mediated motion, and the remainder are associated with sliding extensions or hitchhiking, consistent with previous work in lung epithelium (Waterman-Storer and Salmon 1998; Guo et al. 2018). The mechanism of extension can sometimes be quantified just from the speed of the event, since sliding and hitchhiking events occur at the speed of molecular motors (∼40 µm/min). By contrast, pTAC- and dTAC-mediated tubule growth occurs at the speed of the respective process (pTAC, 8 µm/min; dTAC, 16 µm/min). A much rarer form of ER tubule extension known as “budding” extension has also been observed in some systems, where ER tubules are formed in the absence of any guiding microtubule. The mechanism for these events is not well understood, but it may depend on forces from other cytoskeletal elements such as actin or intermediate filaments.

Independently of microtubules, ER tubules have been observed to undergo significant oscillatory motion in live cells on millisecond timescales (Fig. 4I; York et al. 2013; Nixon-Abell et al. 2016; Guo et al. 2018). The mechanisms governing this motion remain enigmatic, but the oscillation is largely dependent on cellular energy consumption and is not solely the result of thermal motion (Nixon-Abell et al. 2016). Several studies have attempted to model this motion as elastic behavior of a range of physical structures, including semiflexible polymers and viscoelastic filaments (Georgiades et al. 2017; Pain et al. 2019; Perkins et al. 2021). Other studies point to the motion reflecting the natural fluctuations in local pressure that have been described within the cytosol of living cells (Brangwynne et al. 2009; Guo et al. 2014). Regardless of the exact mechanism, fluctuating ER motion likely plays an important role in ensuring the ER's widespread distribution throughout the cell, a feature that is critical for the ER to interact with other organelles and to sense and respond to the overall cellular environment (Valm et al. 2017).

Three-Way Junctions

To form the network that makes up the peripheral ER, individual tubes are linked together at three-way junctions. While the appearance of a junction in light microscopy can be achieved by simply tethering the tip of one tubule to the side of another, the continuity of the structure seen in photobleaching studies suggests most of the visible junctions represent continuous structures where the membrane and lumen can freely exchange between all branches (Pagano et al. 1983; Nehls et al. 2000; Snapp et al. 2006). Modeling work suggests that the presence of continuous three-way junctions can also release curvature tension within the ER system, since such junctions contain regions of negative curvature and would likely contain a small region of flat membrane on the top and bottom of the junction (Fig. 5A,C; Shemesh et al. 2014). Supporting these models, recent volumetric FIB-SEM has revealed that many three-way ER junctions have regions of flat membrane on the top and bottom of the junction (Nixon-Abell et al. 2016; Heinrich et al. 2021).

Figure 5.

Tubule-based endoplasmic reticulum (ER) structures. (A) Three-dimensional FIB-SEM reconstruction of a three-way junction from a COS7 cell. (B) Cartoon schematic indicating atlastin-mediated tubule fusion to generate lunapark-stabilized three-way junctions. Atlastin is pseudocolored in pink and purple, lunapark in cyan. (C) Three-dimensional schematic of a three-way junction showing flattening of membrane on top and bottom. (D) Structured illumination micrograph showing peripheral ER network (white) and distribution of three-way junctions (red). (E) ER overlaid with tubule skeleton (white) and three-way junction position over time (green). (Panel E reprinted from Nixon-Abell et al. 2016 with permission from the authors who hold the copyright.) (F) Airyscan image of a 3D matrix (left) with inset (center) paired with equivalently sized FIB-SEM reconstruction of a 3D matrix. (G) SIM image of a 2D matrix (left) with inset (center) and equivalently sized FIB-SEM reconstruction of a 2D matrix. (H) GI-SIM of ER matrices, with kymographs at indicated positions indicating tubule rearrangements within the matrix over time. (Panel H reprinted from Nixon-Abell et al. 2016 with permission from the authors who hold the copyright.) Scale bars, 2.5 µm (D); 1 µm (E); 5 µm, 1 µm (F); 2 µm, 500 nm (G); colored bars are 1 µm. The height of each kymograph is 2.5 sec. (H).

The specific pathways involved in tethering and fusing three-way junctions are still incompletely understood. This is in part because ER tubules that are tethered to one another cannot be easily distinguished from those that are fused by light microscopy. One clear contributor is the evolutionarily conserved atlastin family of ER-localized GTPases (Hu et al. 2009; Orso et al. 2009). Atlastin-1 alone is capable of tethering and fusing ER membranes within in vitro systems and in Xenopus egg extracts (Fig. 5B; Liu et al. 2015; Wang et al. 2016), and overexpression of Atlastin-1 appears to increase the abundance of ER three-way junctions in cells (Hu et al. 2009; Nixon-Abell et al. 2016). Furthermore, the yeast atlastin homolog Sey1p and a curvature stabilizer (e.g., Yop1p) are sufficient to reconstitute a tubular network from liposomes in vitro (Powers et al. 2017). Atlastin proteins are not the only player in the process of ER tethering and fusion, however, since knockdown or genetic deletion of all three isoforms of atlastin is insufficient to remove all junctions (Zhao et al. 2016). Additional proteins have been implicated in this process in mammalian cells including the AAA ATPase spastin (Evans et al. 2006; Sanderson et al. 2006) and several Rab GTPases (English and Voeltz 2013b; Gerondopoulos et al. 2014), but the mechanisms and how they interface with atlastin are unclear. Additionally, not all these proteins are conserved across eukaryotes, so they may represent more recent adaptations to a conserved atlastin-based system for generating three-way junctions.

The highly conserved protein lunapark has also been proposed to play a role in three-way junction formation and/or maintenance, but exactly how is unknown (Chen et al. 2012, 2015; Wang et al. 2016). In some cells, lunapark specifically localizes to three-way junctions in a manner dependent on its N-myristoylation motif (Fig. 5B; Moriya et al. 2013; Wang et al. 2016). Lunapark and atlastin clearly interact in vitro and in cells (Chen et al. 2012, 2015; Zhou et al. 2019), but even though both promote three-way junction formation, their interactions appear to be antagonistic (Chen et al. 2012; Zhou et al. 2019). Supporting the complexity of this relationship, both overexpression and knockdown of lunapark are associated with the appearance of sheet-like structures in the peripheral ER (Shemesh et al. 2014; Chen et al. 2015; Wang et al. 2016). One potential explanation for these conflicting observations is that tightly clustered three-way junctions can appear as sheets in confocal images. Thus, experiments examining lunapark overexpression are likely to be unclear unless superresolution imaging approaches are used. Indeed, superresolution imaging of plant cells overexpressing lunapark showed induction of tight assemblies of three-way junctions, with these junctions appearing as sheet-like structures when imaged by conventional diffraction-limited light microscopy (Sun et al. 2020). Work in Xenopus egg extracts suggests that phosphorylation of lunapark can decrease the formation of three-way junctions (Wang et al. 2016). As lunapark phosphorylation also increases during mitosis (Wang et al. 2016), this may provide a mechanism by which ER morphology is regulated during the cell cycle.

Three-way junctions are highly dynamic structures in living cells (Fig. 5D,E). Besides forming or breaking, the junctions themselves can be moved by direct or indirect action of molecular motors, through association with organelles, or by cytoplasmic fluctuations (Nixon-Abell et al. 2016; Guo et al. 2018). The latter process is dependent on broad energy consumption by the cell and could arise from changes in local cytoplasmic pressure from the collective action of cytoskeletal assembly and disassembly events (Guo et al. 2014). Not surprisingly, studies directly tracking the motion of three-way junctions in live cells have revealed highly heterogeneous behavior (Nixon-Abell et al. 2016).

ER Matrices

When three-way junctions cluster together tightly, either in a single plane or as a three-dimensional structure, they form what are termed ER matrices (Fig. 5F,G; Nixon-Abell et al. 2016). Conventional widefield or confocal imaging approaches are unable to distinguish ER matrices from continuous sheet-like ER because the fine nanostructures of ER matrices are blurred without superresolution imaging (see Fig. 2A). High-speed superresolution imaging experiments have shown ER matrices can arise through coalescence of existing three-way junctions (Nixon-Abell et al. 2016; Zucker and Kozlov 2022). Once formed, they are more stable in location over time than isolated tubules or three-way junctions, but the spaces within them appear to open and close very rapidly (Fig. 5H; Nixon-Abell et al. 2016; Schroeder et al. 2019). The molecular mechanisms driving this coalescence and its subsequent dynamic stabilization are unclear. Initial observations using light microscopy found significant heterogeneity in conventional ER shaping protein localization to ER matrices (Nixon-Abell et al. 2016). While this may represent the inability to distinguish subclasses of ER matrix with many light-based imaging tools, overexpression of atlastin isoforms, which mediate homotypic ER membrane fusion, were shown to increase ER matrix abundance in cells (Nixon-Abell et al. 2016).

Recent biophysical modeling work has suggested the membranes at ER matrices may have low tension relative to ER tubules or sheets due to the high positive curvature of the tubules being balanced by the negative curvature of junctions (Zucker and Kozlov 2022). Thus, formation of matrices could represent a mechanism for decreasing elastic tension in the ER membrane. This might make the membrane in ER matrices more flexible than on isolated tubules or single isolated sheets. ER processes needing flexible membrane surfaces (e.g., budding sites involved in lipid droplet biogenesis, ER export, or peroxisomal biogenesis) might thus find ER matrices to be optimal sites for their activities. Because ER matrices have significantly higher surface area-to-volume ratios compared to similarly sized ER sheets, they may also be ideal sites for accumulation of excess membrane proteins.

Stacked Helicoidal and Twisted Sheets

ER sheets are flattened ER surfaces that maintain a constant luminal spacing over a long distance. In many cases, sheet edges have a high curvature that resembles that seen for tubules in a cross section (Porter and Palade 1957; Terasaki et al. 2013). ER sheets are often tightly stacked on top of one another with uniform spacing between the sheets (Fig. 6A; Porter et al. 1945; Porter and Palade 1957; Fawcett 1966). Reconstructions from serial EM sectioning revealed that these structures are helicoidal in nature, resembling the layers of a parking garage (Fig. 6A; Terasaki et al. 2013). The helicoidal ramps connecting the layers were shown to be both right- and left-handed in orientation and were proposed to reduce elastic stress on the structure induced by sheet edges (see below) (Terasaki et al. 2013; Shemesh et al. 2014). Stacked helicoidal sheets have been shown to be highly prevalent in professional secretory cells such as mature B cells or mouse salivary gland, but they are also present in the perinuclear region of many other cells including neurons (Palay and Palade 1955; Terasaki et al. 2013) and tissue culture cells (Nixon-Abell et al. 2016; Heinrich et al. 2021). Stacked helicoidal sheets are generally decorated with ribosomes and often appear as a fused, blurry structure in fluorescence microscopy images (e.g., Fig. 2B).

Figure 6.

Sheets. (A–D) Distinct sheet-based structures observable in 3D EM data, including (A) stacked helicoidal sheets, (B) twisted sheets (arrow indicates pitch), (C) cisterna with irregular luminal spacing, and (D) flat sheets with fenestration. (E) XZ FIB-SEM slice of a U2-OS cell with zoomed inset of the basal nuclear envelope. Yellow = nucleus, blue = cytoplasm. (F) XY slices of the nuclear face of the inner membrane (INM) showing chromatin (left), the perinuclear space showing nuclear pores (center), and the cytoplasmic face of the outer nuclear membrane (ONM) showing polyribosomes. Scale bars, 2.5 µm, 500 nm (E); 1 µm, 1 µm, 1 µm (F).

Although it seems that most cells have at least some stacked sheets in them, not all sheets are stacked. Isolated sheets have been described in the perinuclear and transitional region of several cells using three-dimensional EM (Puhka et al. 2007, 2012; Nixon-Abell et al. 2016; Heinrich et al. 2021). Like the stacked sheets, these structures are not actually flat. The studied examples seem to show some degree of slope or pitch, which often manifests as a twisted structure reminiscent of one layer from a helicoidal stack (Fig. 6B). Isolated sheets have not been extensively studied, but they are highly prevalent in cultured cells and seem likely to play similar biological roles to the stacked sheets described above.

Several protein players have been implicated in stabilizing ER sheets. The proteins CLIMP63, p180, and KTN were identified early as stabilizers of ER sheets in mammalian cells, because they are enriched in ER sheets and overexpression causes an expansion of the perinuclear ER, where such structures are more prevalent (Shibata et al. 2010). All three proteins contain dimerizing coiled-coil domains, which are thought to be important for their sheet-stabilizing function. While it remains unclear how the sheet-stabilizing function of p180 and KTN works, the coiled-coil domains of CLIMP63 are in the lumen and are thought to act as a luminal spacer by dimerizing in trans across the structure (Shibata et al. 2010; Shen et al. 2019). Indeed, overexpression of fluorescently tagged CLIMP63 stabilizes the width of ER sheets, but also a variety of tubular or fenestrated ER structures (Shibata et al. 2010; Gao et al. 2019; Schroeder et al. 2019). This may reflect the fact that luminal spacing could also be accomplished in round structures if the CLIMP63 dimers are not arranged in parallel, so the microtubule-binding function may also be required for sheet-stabilizing capacity (Schweitzer et al. 2015). Consistent with this possibility, simultaneous knockdown of all three proteins destabilizes the luminal spacing of sheets while not eliminating the presence of these structures (Shibata et al. 2010). Reticulons and DP1/REEP/Yop1p families of proteins have been suggested to localize preferentially at the edges of sheets, where they could stabilize the high curvature (Shibata et al. 2010; Shemesh et al. 2014). A combination of sheet edge stabilization by the curvature stabilizers and of luminal cross-linking by proteins like CLIMP63, therefore, could give rise to the constant luminal width of sheets. Other cellular components, such as ER-localized polyribosomes have also been implicated in stabilizing the flat regions of ER sheets, presumably by tethering sheet-associated translocon complexes to one another with the shared mRNA molecule they are translating (Shibata et al. 2010; Lin et al. 2012). However, ribosomes are frequently observed on locally flattened regions of ER tubules as well (Fig. 3A; Heinrich et al. 2021), so it remains to be seen whether this sheet-stabilization is a cause or an effect of polyribosome docking. Cytoskeletal components are not thought to be required to form ER sheets.

ER Cisternae and Fenestrated Sheets

It is increasingly clear from EM studies of cells from diverse tissues that many regions of ER are large, continuous structures that do not have the constant luminal spacing characteristic of twisted or classically stacked ER sheets (Fig. 6C). We refer here to these structures as ER cisternae. ER cisternae are highly prevalent in the perinuclear region of most cells and may swell to having a diameter of many microns under defined conditions of ER stress or protein misfolding (Porter and Palade 1957; Palade 1975; van Anken et al. 2021). Preliminary analysis of ER cisternae in the few annotated three-dimensional EM data sets where ribosomes are also visible suggest that these structures are often coated with polyribosomes, suggesting involvement in protein translation (Heinrich et al. 2021).

Fenestrated ER sheets are sheets heavily riddled by pores, known as fenestrations, of varying densities. They are essentially planar structures (not stacked or twisted) and can be seen in both the perinuclear and peripheral ER (Fig. 6D; Porter and Palade 1957; Puhka et al. 2007, 2012; West et al. 2011; Nixon-Abell et al. 2016; Schroeder et al. 2019; Hoffman et al. 2020). The fenestrations can be just tens of nanometers in diameter, making it extremely challenging to unambiguously distinguish these structures from sheets, cisternae, and especially planar ER matrices. However, a recent study has clearly identified fenestrated sheets as an abundant additional component of the ER structural repertoire (Schroeder et al. 2019).

Having only been described recently, the mechanisms of fenestrated sheet formation are not yet clear. One possibility is the formation of fenestrations de novo on a stacked or twisted sheet. Another possibility is that fenestrated sheets arise from closures of the spaces in two-dimensional ER matrices through curvature-favoring fusogens like atlastins, or through ring closure events on the lateral edges of existing sheets (Zucker and Kozlov 2022). Regardless of the mechanism of their formation, fenestrated sheets appear to have less three-dimensional pitch associated with them than the continuous twisted or stacked sheets. This suggests the possibility that negative curvature in fenestrations could release some of the membrane tension inherent in continuous structures with highly curved edges (Fig. 6D; Shemesh et al. 2014; Zucker and Kozlov 2022).

Initial studies have not found heavy enrichment of the sheet-associated protein CLIMP63 within fenestrated sheets in normal cells (Schroeder et al. 2019). This might help explain why the lumen of the continuous portion of fenestrated sheets is highly variable in width as seen by high-resolution imaging, as overexpression of CLIMP63 was shown to stabilize the width (Schroeder et al. 2019). High-resolution tracking of sheet fenestrations revealed two distinct subpopulations of fenestrations. Some fenestrations were stable over seconds to minutes, with the curved inner edge of the membrane in these structures likely stabilized by reticulons. However, other fenestrations appeared to rapidly open and close on a millisecond timescale. Either way, the dynamism of these fenestrated surfaces was unprecedented and seems likely to be a site for unique biological phenomena (see Fig. 5H).

NE Sheets and ER-NE Junctions

The NE of interphase cells is essentially a single continuous ER sheet that wraps the entire nucleoplasm to create a distinct environment from the cytosol. Like other ER sheets that are not stacked or twisted, the NE's sheet-like structure is riddled with fenestrations (Fig. 6E,F). In the case of the NE, however, the fenestrations, called nuclear pores, are specialized structures. Highly stable, nuclear pores span the NE's outer and inner membranes, creating a selective passageway for cytosolic and nucleoplasmic proteins. The nuclear envelope is a complex structure with many proteins dedicated to maintaining its unique properties and shape. Recent excellent reviews on this topic include Rothballer and Kutay (2013), Bahmanyar et al. (2014), Burke and Stewart (2014), Janota et al. (2017, 2020), Buchwalter et al. (2019), Bahmanyar and Schlieker (2020), and Pawar and Kutay (2021).

ER-NE junctions have been difficult to capture in 2D EM, but these structures are proposed to be unique tubule-to-sheet junctions angled at ∼90°. Given their unusual morphology, ER-NE tubular junctions operate as diffusion barriers for specific ER proteins and may also play this role for membrane lipids (Bahmanyar and Schlieker 2020). This could occur either as a direct physical consequence of the proximal regions of highly positive and highly negative curvature, or it could be mediated through the function of specific proteins at the junction. Supporting this latter possibility, lunapark has been shown to localize to these junctions, where it has been proposed to regulate the bulk flow of lipids into the NE from the ER (Kume et al. 2019; Bahmanyar and Schlieker 2020; Hirano et al. 2020).

CONCLUSIONS

In this review, we have detailed the various morphological structures of the ER, which are remarkably different—ranging from dense arrays of intersecting tubules to expansive sheets, with all manner of structures between. These structures show significant local variation in membrane curvature, which may have dramatic effects on the biophysical environment experienced by associated macromolecules. As biological processes occurring in the ER become more effectively mapped to specific ER structures, it will be fascinating to see how ER shape governs local ER functions and vice versa. This is of particular interest in the context of many of the newer structures described in this review, which do not yet have clearly defined or proposed functions.

One understudied component of the ER is its local lipid content, which may vary across the structure. Numerous lipid species have structures that cause them to prefer specific membrane curvatures (Cooke and Deserno 2006; McMahon and Boucrot 2015). It is likely that mechanisms that change local ER structure (whether through shaping proteins or applications of force from cytoskeleton or other components) are likely to have effects on local lipid content. Additionally, at sites of altered lipid composition or lipid synthesis, the lipids themselves may directly contribute to the ER curvature or structure in ways that we do not yet understand. The development of tools for directly observing or manipulating ER lipid content will be crucial for understanding this interplay.

Because the ER's structural diversity is likely to emerge from many factors (including the presence of specific shaping proteins and lipids), disrupting one or more of its morphologies is likely to feed back on the rest of the system, making it challenging to understand the specific roles of individual structures and the proteins that stabilize them. Because of this, it will be important to now think of the ER holistically as an entire system, one that is working to maintain itself to be responsive to cellular changes. Only with such a conceptual framework can we hope to understand how the different mutations in ER proteins and regulation pathways give rise to so many human pathologies.