Three-dimensional reconstruction of cytoplasmic membrane networks in parietal cells

Striking morphological transformations characterize the stimulation of the gastric parietal cell; the canalicular apical membrane greatly expands at the expense of an extensive intracellular membrane network. This membrane network,rich in the gastric proton pump (H,K-ATPase), has conventionally been called tubulovesicles. On these facts, there is a broad consensus. A subject of ongoing controversy has been the precise morphological form of the tubulovesicles within the parietal cell. From this controversy, another question naturally arises: what is the mechanism of the stimulatory morphological transformation?

At the heart of the tubulovesicle controversy lies the question of whether tubulovesicles are a distinct membrane compartment or whether they are continuous with the apical plasma membrane. These two views lead to very different models of morphological transformation. The view that tubulovesicles are a distinct membrane compartment leads to the conclusion that parietal cell stimulation and consequent relaxation are exocytic and endocytic events,respectively. This so-called membrane recruitment/recycling hypothesis holds that proton pump-rich tubulovesicles are recruited to the plasma membrane as the cells stimulate and then recycle it back to the cytoplasm as food leaves the stomach and the cells return to a resting state (Forte and Yao, 1996;Forte et al., 1977). Proponents of the view that the tubulovesicle system is one membrane continuous with the plasma membrane posit that the apparent expansion of the apical canaliculus is actually the result of ions and water moving into the highly collapsed`tubulovesicular' continuum as a result of osmotic forces generated by the activated pump (Berglindh et al.,1980). This view is called the osmotic expansion hypothesis.

Although there is a wide assortment of structural and functional evidence to support the membrane recycling hypothesis (Forte and Yao, 1996;Agnew et al., 1999;Peng et al., 1997;Duman et al., 1999), arguments favoring the osmotic expansion view persist. Most of these latter arguments are based on morphological evidence. For example, Pettitt et al.(Pettitt et al., 1995;Pettitt et al., 1996) have contested earlier electron microscopic evidence, noting that glutaraldehyde fixatives can fragment intracellular membrane compartments and asserting that the model of resting parietal cell morphology embraced by the recruitment/recycling hypothesis was based on artefact. These authors employed rapid-freeze fixation and suggested that the tubulovesicular compartment might be arranged as a series of tightly coiled membranes, which they alleged were continuous with each other and with the apical canaliculus. To deal with functional studies, such as those involving the penetration of electron dense tracers and measurements of electrical capacitance, they proposed small sized openings for the `connections' and tight packing of the membrane coils. Using rapid freeze fixation and high-resolution scanning electron microscopy, Ogata and his colleagues have offered an alternative interpretation of tubulovesicular morphology (Ogata 1997; Ogata and Yamasaki,2000a; Ogata and Yamasaki,2000b). They proposed that the compartment is actually composed of multiple flattened membrane sacs, or cisternae. They also proposed that the cisternae intercommunicate with each other by means of very narrow tubular connections that are present even in the resting cell, but the communication with the surface only occurs with stimulation. Thus, it is clear that important structural details must be elucidated for function of gastric-acid-secreting cells to be fully appreciated.

In the present work, we used high pressure rapid freezing techniques to fix samples for electron microscopy and tomography. We sought to investigate whether the tubulovesicular membranes are, in fact, distinct from the apical canaliculus, determine the actual morphology of the tubulovesicular membranes and appreciate the tubulovesicular compartment in the context of other cellular membranes within the parietal cell. Our data provide an exciting view of the parietal cell interior. They fully support the claim that tubulovesicles are distinct from the canalicular membrane and provide evidence consistent with the tubulocisternal morphology promoted by Ogata(Ogata, 1997). However, there is no evidence of permanent continuities between the individual elements of the tubulocisternal system of the resting parietal cell.

Gastric glands were isolated from New Zealand White rabbits after collagenase digest of gastric mucosa as previously described(Reenstra and Forte, 1990). Glands were shaken at 37°C with 100 μM cimetidine in oxygenated minimal essential medium. After 20 minutes, the glands were gently sedimented,transferred to Type A high-pressure freezing planchettes and frozen in a Bal Tec HPM 010 High Pressure Freezer. Samples were removed within 1-2 seconds and transferred to liquid nitrogen.

Glands were next freeze-substituted on dry ice at -78°C with 2% osmium tetroxide and 0.1% uranyl acetate in acetone for 3 days and then warmed to room temperature over 12 hours (McDonald,1999). After three 10-minute rinses with pure acetone, glands were infiltrated and embedded in Epon-Araldite resin. Resting gastric gland preparations were made from three rabbits. Each preparation was embedded in multiple blocks, and sections were examined in three to five blocks from each gland preparation.

Ultra-thin (60-70 nm thick) serial sections were cut from the embedded glands with a Reichert Ultracut E microtome and poststained with lead citrate and uranyl acetate. We visualized the sections in a JEOL 100CX (JEOL USA)transmission electron microscope operating at 80 kV. Electron micrographs were collected for the same regions of parietal cell interiors, usually including at least one canaliculus, in each of the series of sections. Negatives of all images were subsequently digitized by scanning.

The tomogram was collected as previously described(Ladinsky et al., 1999;Marsh et al., 2001;McIntosh, 2001). The tomogram was generated at the Laboratory for Three-Dimensional Fine Structure at the University of Colorado, Boulder. Briefly, 400 nm thick sections of the embedded material were cut on a Leica Ultracut UCT microtome (Leica, Inc.) and placed on formvar-coated, copper-rhodium slot grids. Following staining as per above, 15 nm colloidal gold particles were placed on both surfaces of the grid to serve as fiducial markers for subsequent image alignment. The grid was stabilized with carbon and placed on a stage capable of high tilt and rotation. It was visualized with a JEM-1000 (JEOL USA) operating at 750 KeV. The section was rotated from +60° to -60° with images being captured at 1.5° intervals; the section was then rotated 90° and a similar series of images taken. Tomograms calculated from each tilt-series were combined into a single high-resolution tomogram using a warping procedure(Mastronarde, 1997).

Models were constructed on Silicon Graphics computers running MIDAS and IMOD software (Kremer et al.,1996). Image stacks of particular areas were aligned using MIDAS. Because the electron beam mildly distorts ultra-thin sections, images had to be aligned with respect to specific structures of interest and re-aligned to model structures that were distant from these. This step was not necessary for the tomograms. We used IMOD to stack the aligned images and to draw contours(outlines) on specific cellular structures. Contours of the same structure from different serial sections were arranged into objects using IMOD. Using the IMODmesh feature of IMOD, we joined the contours of each object to form a 3D model.

Movies of these models rotating in space were made using Mediarecorder and are available at jcs.biologists.org.

Fig. 1 is a micrograph of an ultra-thin section through a well-preserved parietal cell at low magnification. This cell exhibits typical parietal cell features:cross-sectional profiles of canalicular intrusion into the cytoplasm,plentiful microvilli decorating the canaliculi and an abundance of mitochondria distributed throughout the cytoplasm. At higher magnificationFig. 2 shows the extensive network of cytoplasmic membranes in the juxta-canalicular region. These tubulovesicular membrane profiles crowd the preponderance of cytoplasm not occupied by canaliculi and mitochondria. Previous studies have shown that these tubulovesicles are the H,K-ATPase-rich membrane pool(Ogata and Yamasaki, 2000b;Okamoto et al., 2000). In thin section, many of these membranes appear as flattened sacks, some reaching 0.5μm in length along their major axes and apparently arranged in cisternal stacks (arrowheads). Vesicular profiles are also visible, although they comprise a small fraction of the total pool. It is also noteworthy that the canalicular membrane shows a proliferation of invaginations consistent with the clathrin coated endocytic structures described by Okamoto et al.(Okamoto et al., 2000) in the resting parietal cell. As frequently observed, parietal cells are rich in mitochondria. For the rabbit gastric glands in the present study we found that mitochondria occupied 40.3±1.8% of the cross-sectional area of five separate parietal cells (after excluding nuclei and canalicular compartments). This agrees with previous estimates of mitochondria representing 20% and 42%of parietal cell volume in dog, mouse and pig(Schofield et al., 1979;Zalewsky and Moody, 1977;Black et al., 1981).

In order to appreciate the three-dimensional organization and arrangement of parietal cell organelles, three-dimensional models were created from serial ultra-thin sections of gastric glands. Fig. 3 shows the steps in the modeling process applied to a region of the canaliculi extending through 29 sections accounting for a depth of ∼2μm (approximately 120 μm³ volume). Morphological complexity made it difficult to follow the exact outline of the entire canalicular surface including each microvillar extension, so we first included only the surface that outlined the canaliculus at the base of the microvilli.Fig. 3A represents one section in the stack and shows contours outlining two canaliculi (in blue); a series of 29 unconnected contours are assembled into an object inFig. 3B; andFig. 3C represents the final meshed model in which three canaliculi wending through the cytoplasm can be seen. The canaliculi clearly form a branching network, demonstrated by the joining of two of these structures; we have seen many such interconnections. In order to appreciate the microvillar relationships at the apical surface, we first modeled a single segment of a canaliculus as one object, then modeled the inherent microvilli as a separate series of objects and finally put them together in a reconstruction. Fig. 4 highlights the interior of an individual canaliculus through cutaway views of two orientations: a longitudinal(Fig. 4A) and a cross-sectional view (Fig. 4B). When one looks at an individual ultrathin section, a large number of microvilli are typically seen in cross section, but there is little information about their orientation within the canaliculus. This reconstructed model, in which only ∼50% of the microvilli have been modeled, shows a random distribution of microvilli projecting up, down or in line with the normal axis of the canaliculus. This random arrangement is clearly different from what is typically observed in other microvilli-rich epithelial cells.

Fig. 5 is a model reconstructed from a series of mitochondrial profiles from a region of ∼70μm³ within the parietal cell cytoplasm. The model clearly demonstrates that the mitochondria do not resolve into discrete units. In fact, they form an extensive reticular network coursing throughout the cytoplasm. The tight radii of some of the interconnections suggest the possibility of transient connections that may open and close in time, although this cannot be confirmed within these data.

A three-dimensional model of many tubulovesicular structures from a region near a canaliculus is shown in Fig. 6A. This reconstruction from serial sections reveals that the tubulovesicles exist primarily in the form of small cisternal stacks: each stack including from three to six cisternae. Often, small tubular or vesicular structures are apposed to these stacks. Careful review of many sections did not reveal any connections between membranes within the stack; each tubulovesicle (or tubulocisterna) appears to be a discreet membranous unit. Furthermore, orientation of the axis for a given tubulocisternal stack appears to be independent of nearby stacks. Fig. 6B shows a model of several stacks of tubulocisternae juxtaposed with the apical canalicular membrane. Although the membrane stacks are packed quite close to one another and frequently are subadjacent to the canaliculi,we did not observe any visible connections among tubulocisternae or between tubulocisternae and the canaliculus. Thus, while our data clearly show membrane interconnections between canalicular segments, and among mitochondria, they argue against intercompartmental connections in the resting parietal cell.

A possible artefact in these models is the distortion of thin section geometry that might be caused by the electron beam, as mentioned above. Also,the modeling data are resolved only between ultrathin sections that are about 70 nm in thickness. To examine the system at a higher point to point resolution, we employed tomography, which involves recording and deconvoluting a series of images at different tilt angles through thick sections of tissue. The tomographic procedure yields a stack of images similar to those obtained through serial sections but without the motional distortions between different sections. Thus the image stack thus possess extraordinary continuity with a distance of only 2.5 nm between successive sections. Individual image quality is also quite good, as seen in Fig. 7.

Tomograms taken from 0.4 μm sections were used to model the tubulovesicles/tubulocisternae, employing the same methods with which we modeled the ultra-thin sections. Fig. 7 shows the result of this process. The insert toFig. 7A represents one section from the stack in which we have outlined the contour of many tubulocisternal profiles. Fig. 7A depicts contoured stacks of tubulovesicles and tubulocisternae within a section about 0.3 μm thick. A stack of contours for a nearby mitochondrion is shown for comparison of size and orientation. In Fig. 7B the profiles have been meshed and the entire reconstruction is rotated. Individual colors were assigned to each membrane structure and maintained through each part of the figure; the red contour lines outline a vicinal mitochondrion. This model is in excellent agreement withFigs 6A and 6B, which were reconstructed from serial sections. Close apposition favors some sort of stacking of the tubulocisternae, and small vesicles and tubules cluster around the stacks. As for the analysis of serial section data, interconnections between individual cisternae, vesicles and tubules are not observed.

We undertook this study to bring new advances in fixation and modeling technology to bear on the controversial nature of the tubulovesicular compartment in gastric parietal cells. The technique of fixation by high pressure freezing offers significant advantages over conventional methods. The contents of the sample are immobilized in milliseconds(McDonald, 1999), leading to a much higher level of tissue integrity than standard fixation, which takes considerably longer. Also, many cell structures do not react completely with the standard gultaraldehyde fixative, giving them freedom of movement at room temperature until post-fixation with osmium tetroxide. High pressure freezing allows the post-fixation to occur at extremely cold temperatures so that cell components are not lost or rearranged during the course of sample preparation(McDonald, 1999).

Ever since the ultrastructure of the parietal cell was first examined more than 40 years ago, the system of extensive intracellular membranes has been of great interest, especially with respect to its possible role in the HCl secretory process. The morphological form of these membranes has variously been called vesicles, tubules, bulbotubules, tubulovesicles, coiled tubules and, most recently, tubulocisternae. Although not always conceding to the specific morphological form, most authors have referred to the system generically as `tubulovesicles'. These descriptive differences have been due in part to the animal species examined and in part to the methods of tissue preservation (Forte and Forte,1971; Helander,1981; Ito, 1987). The discovery that H,K-ATPase, the gastric proton pump, was the predominant protein in tubulovesicles (Forte et al.,1975), coupled with secretion-related interconversion between tubulovesicles and apical canalicular membrane, led to the membrane recycling hypothesis of acid secretion (Forte et al.,1977). However, as noted in the introduction, other authors have offered alternative views for secretion-related changes in membrane morphology that very much depend upon the specific form of the tubulovesicular system in the resting cell (Pettitt et al.,1995). Thus the need to examine parietal cell ultrastructure in three dimensions and with superior fixation techniques is readily apparent.

From the structural analyses presented here, we conclude that the H,K-ATPase-rich intracellular membranes in rabbit parietal cells are predominantly comprised of flattened cisternae, although tubules and vesicles can also be found. Earlier work using the same high pressure freezing method,but with freeze-substitution fixation more suitable to immunocytochemistry, we showed that these membranes are in fact the loci of H,K-ATPase(Okamoto et al., 2000). Moreover, there are no apparent permanent connections among the tubulocisternal components or between the tubulocisternae and the apical plasma membrane.

We are not the first group to apply rapid freezing methods to the problem of parietal cell ultrastructure; however, our conclusions are at variance with several other authors. From their analysis of micrographs and 3D modeling,Pettitt et al. (Pettitt et al.,1995) concluded that the H,K-ATPase-rich membrane system was composed of densely packed helical coils of tubule, each having an axial core of cytoplasm and joined together by connecting straight tubules. Furthermore they predicted that the coiled tubules were interconnected with the apical canalicular membrane and that the secretagoguemediated resting/stimulated transition was the result of unwinding of the coils, thus exposing the cytoplasmic cores as shaped elements resembling extended microvilli of the stimulated apical surface. Although their model is interesting, it is difficult to reconcile with the raw data. Micrographs of unstimulated parietal cells in Pettitt et al. (Pettitt et al.,1995) reveal a preponderance of elongated membrane profiles,sometimes straight but often curved. In nine of their published micrographs of unstimulated parietal cells we counted 75.6% (SD=10.8) of the membrane profiles as `elongate' (for simplicity we define `elongate' as any profile whose length is >2× the width; `circular' as a profile where length is <2× width). For an ideal thin section through a system of straight tubules we would predict that at least 60% of the image profiles should appear as `circular'; the percentage would be even higher for curved tubules. Thus a model of coiled tubules cannot be the predominant form for the H,K-ATPase-rich intracellular membrane system of mammalian parietal cells. That is not to say that coiled tubules do not exist or that tubules and vesicles are not among the H,K-ATPase-rich membranes, but that there would probably be a significant number of cisternal elements in the pool, as presented by Pettitt et al.(Pettitt et al., 1995). We would also like to point out that while tubulocisternae form the preponderance of the H,K-ATPase-rich membranes in these mammalian cells, the same is not true for amphibian cells where the principal morphological form is that of relatively straight tubular structures(Forte and Forte, 1971).

We are also not the first group to suggest that the H,K-ATPase-rich membranes occur in the form of tubulocisternae. Ogata et al. came to this conclusion independently using alternate techniques, that is, scanning electron microscopy of macerated gastric glands(Ogata and Yamasaki, 2000a,Ogata and Yamasaki, 2000b;Ogata 1997). However, our picture of the resting cell cytoplasm is not in complete agreement with theirs. Ogata's group observed a number of connections, in the form of slender tubules, between the various tubulocisternae. They also reported that tubules could also occasionally be seen connecting the apical canaliculus and tubulovesicles. Although Ogata generally does not favor the osmotic expansion hypothesis, he does propose a schematic model that includes extensive interconnections among the elements of the H,K-ATPase-rich membranes. Thus,despite our similar conclusions about a tubulocisternal structure, there remain substantive differences in observation and interpretation. The models we developed from the present data are based on structures that have received a maximum care in structural preservation, and we find no evidence for connections within and among the tubulocisternae. Even if connectivities were difficult to see in ultra-thin sections, they should have been readily apparent in the high-resolution tomograms.

A functional implication of our model is that it clearly adds support to the membrane recruitment/recycling hypothesis. If the membranes containing the pump are not continuous with the plasma membrane, they surely must be at the time of acid secretion. Aside from this obvious consequence, a number of questions remain. How does the cisternal morphology of H,K-ATPase-rich membranes play a role in the process of acid secretion? Are the tubules and vesicles that we do observe intermediates in a sorting or general housekeeping pathway? How are the cisternal stacks regenerated upon cell relaxation? All of these questions present challenges for the future.

The structure of the apical canaliculus brought few surprises. In agreement with a host of previous work from light microscopy, the canalicular membranes were shown to form a network wending through the cytoplasm. Microvilli were plentiful, even in the resting cell. Although there were no fixed continuities between the canalicular membrane and tubulocisternae in the resting cell,there was abundant evidence for endocytic activity of clathrin-rich membranes as documented in an earlier study (Okamoto et al., 2000).

Another interesting observation regards the nature of parietal cell mitochondria. The modeling data clearly show branching and interconnections among many of the mitochondria, supporting the evolving idea of a large dynamic mitochondrial network within cells. Using 3D reconstruction of thin EM sections, Hoffmann and Avers (Hoffmann and Avers, 1973) reported that mitochondria in Sacchromyces cerevisiae were arranged in a single large reticular network and predicted that this might be a common arrangement in eukaryotic cells. Many subsequent studies suggested that cellular mitochondrial networks are frequently dynamic. Experimental evidence, both from mitochondrial and membrane-potential tracers, provides support for the model of a large dynamic mitochondrial organelle undergoing constant remodeling(De Giorgi et al., 2000;Amchenkova et al., 1998). Moreover, there is evidence for a variety of specialized mitochondrial fusion/fission proteins(Otsuga et al., 1998;Smirnova et al., 1998). These observations raise some fascinating questions about how its structure and dynamism contribute to the function of mitochondria. In a review of these data Skulachev has developed an hypothesis whereby mitochondrial connectivities contribute to power transmission in the form of a transmembrane electrical potential difference over long distances within (or even between) cells(Skulachev, 1990). He pointed out that mitochondrial reticula are often associated, or appear to be, under conditions of hard work or energy deficiency, for example, in skeletal muscle mitochondrial these structures are associated with red rather than white muscle (Ogata and Yamasaki,1997). This may have special relevance to the parietal cell where there is a prominent mitochondrial presence and energy requirements owing to the proton pump are enormous. The huge demand of a functioning H,K-ATPase drives ATP conversion to ADP + Pi at the apical plasma membrane. Oxygen and substrates to regenerate ATP come into the cell at the basal surface. Even with extensive canalicular invaginations there are substantial distances, up to 20 μm, between apical and basal surfaces. The opportunity for power transmission along mitochondrial networks could reduce the time constraints of 3D diffusion for nucleotides and substrates, providing an efficient and well coupled use of the entire mitochondrial compartment in the face of a localized energy sink. Dynamic mitochondrial networks, and their apposition to Ca²⁺ storage compartments, have also been implicated in the regulation of cell signaling (Rizzuto et al., 1998). Because of its abundance of mitochondria, marked difference in energy usage between resting and stimulated states and modulation of secretory activity by calcium, the parietal cell may present itself as an ideal native model to address issues of mitochondrial networks.

The current work offers a new 3D view of parietal cell structures. It also prescribes a number of future directions. We will seek to model the stimulated cells to better appreciate the nature of the stimulated canaliculus. Hopefully we will answer the question as to whether there is an increase in number, as well as length, of apical microvilli in the rest/stimulated transition. Even more exciting, we will model cells transiting both to the stimulated and to the resting state. These observations should greatly assist in addressing the questions that we have posed in this paper regarding the particular advantages and consequences of the model that we have proposed for the resting parietal cell.