[Methods in Cell Biology] Cellular Electron Microscopy Volume 79 || Reconstructing Mammalian...
Transcript of [Methods in Cell Biology] Cellular Electron Microscopy Volume 79 || Reconstructing Mammalian...
CHAPTER 8
METHODS IN CELL BIOLCopyright 2007, Elsevier Inc.
Reconstructing Mammalian MembraneArchitecture by Large AreaCellular Tomography
Brad J. MarshInstitute for Molecular Bioscience, Queensland Bioscience PrecinctThe University of Queensland, BrisbaneQueensland 4072, Australia
Centre for Microscopy and MicroanalysisThe University of Queensland, BrisbaneQueensland 4072, Australia
School of Molecular and Microbial SciencesThe University of Queensland, BrisbaneQueensland 4072, Australia
I. In
OGY,All rig
troduction and Rationale
VOL. 79 0091hts reserved. 193 DOI: 10.1016/S0091
-679X-679X
II. M
ethods and Materials A. M ammalian Cell and Tissue Culture B. F ast-Freezing and Freeze-Substitution C. M icrotomy and Grid Preparation D. B asic Instrumentation Requirements for ET of Thick Plastic Sections E. A cquisition of Digital Tilt Series from Large Cellular Areas byAutomated IVEM/HVEM Montaging Tomography
(
F.
3 D Reconstruction of Large Cellular Areas G. 3 D Segmentation and Quantitative AnalysisIII. D
iscussion A. F uture DirectionsIV. S
ummary R eferencesMicroscopy has provided crucial insights into the fundamental features and
architecture of mammalian cells and organelles for now over a century. These
glimpses of cellular fine structure have thus guided our impressions—as molecular
/07 $35.0006)79008-2
194 Brad J. Marsh
cell biologists—of how mammalian cells and their numerous membrane-bound
internal compartments are organized within three dimensions (3D), although for
the most part these extrapolations have come from static two-dimensional (2D)
images taken from the light or electron microscope. Recently, however, we have
finally been aVorded the chance to dissect subcellular membrane architecture and
dynamics at the nanoanatomical level in 3D through development of the tech-
nique referred to as electron microscope (EM) tomography, also frequently
termed ‘‘cellular tomography’’ (ET). With ET now in hand as a tool that eVec-tively allows one to study molecular membrane traYc ‘‘in context’’ (in situ), it has
become increasingly important to push the continued advancement of this method
so that large cellular volumes can be reconstructed without sacrificing the ability to
clearly resolve structures of interest. Likewise, it has become critical to do this as
quickly and eYciently as possible in order to generate statistically significant
sample sizes that oVer reliable insights into cell organization under diVerentphysiological or experimental conditions. In this chapter, some of the technical
developments, as well as key biological questions that have driven the development
of large area ET, will be presented and discussed.
I. Introduction and Rationale
The mammalian Golgi complex was first observed by light microscopy over
100 years ago by Camillo Golgi (Golgi, 1898) and visualized by electron micros-
copy by Ernest Fullam, Albert Claude, and Keith Porter in 1944 (Porter et al.,
1945). Despite this, the limitations of current and/or conventional cell biological
and biochemical techniques for elucidating the structure–function relationships of
such a complicated organelle can be evidenced by the controversies which remain
today with respect to even the most fundamental mechanisms of transport to,
through and from this organelle (Elsner et al., 2003; Gu et al., 2001; Lippincott-
Schwartz et al., 2000; Marsh and Howell, 2002; Storrie and Nilsson, 2002).
A myriad of approaches ranging from conventional two-dimensional (2D) EM
of thin-sectioned material, in vitro functional (cell free) assays of reconstituted
Golgi membranes, live-cell imaging studies to organellar proteomics analyses have
so far failed to satisfactorily answer basic questions related to molecular mem-
brane traYc through the Golgi (Breuza et al., 2004; Farquhar and Palade, 1998;
Mellman and Simons, 1992; Orci et al., 1989a; Palade, 1988; Rothman and
Orci, 1992; Wu et al., 2004; Yates et al., 2005).
Part of the problem lies in the fact that during the 1980s, as progress was
accelerating in molecular biological and live-cell imaging studies of the Golgi
and intracellular membrane traYc, conventional fine structure studies of intracel-
lular membranes in mammalian cells were for the most part considered passe due
both to their static nature and the limitations of 2D analysis, and resources for
EM in general were significantly downgraded (GriYths, 2004; McIntosh, 2001).
At the same time, high-resolution EM studies that focused on structure were for
8. Reconstructing Mammalian Cytomembranes by 3D EM 195
the most part dismissed in favor of alternative EM approaches that for the first
time allowed the immunolocalization of an antigen or antigens of interest on
standard thin sections (typically 60–100 nm), and thus provided crucial informa-
tion regarding the subcellular distributions of proteins (GriYths et al., 1984; Orci
et al., 1984; Slot and Geuze, 1983). However, as with any other method, techniques
for immunolabeling suVer from their own limitations. As noted above, Golgi
membranes in particular are often highly convoluted, and so it frequently becomes
diYcult to unambiguously follow connectivity from one section to the next. Other
limitations stem from the fact that only antigens exposed at the section surface can
be labeled, and that steric hindrance due to the immunogold itself can limit spatial
resolution to between 10 and 20 nm in some cases; labeling eYciency is also
compromised. In addition, immunolabeling studies of mammalian membrane
traYc have relied primarily on classical chemical fixation techniques to stabilize
membranes for visualization in the EM. As this process takes anywhere from
seconds to minutes, labile structures and highly dynamic events, which we now
know from live-cell imaging studies to be typical for many of the steps in intracel-
lular membrane traYc, are unlikely to be reliably or reproducibly captured (Gilkey
and Staehelin, 1986; Lippincott-Schwartz et al., 2000, 2001). It is now more
apparent than ever that conventional chemical fixation often captures the cell’s
(and/or organelle’s) response to the fixative as opposed to immobilizing cellular
structures in a ‘‘close-to-native’’ state (Biel et al., 2003; Dubochet, 1995). Thus,
it follows that reliable descriptions of morphology for the membranes of a
rapidly changing organelle such as the Golgi can only be achieved through the
use of methods for preservation that immobilize all cellular activity within milli-
seconds, such as plunge freezing or high-pressure freezing. To be supremely useful,
however, such methods for improved ultrastructural preservation of biological
specimens will eventually have to incorporate a capacity for the precise localiza-
tion and identification of proteins, lipids, and inorganic ions of interest within
fast/high-pressure frozen cells or tissues (see Chapters in Part IV, this volume).
Three-dimensional (3D) studies of Golgi organization by stereo-scanning EM
(Ho et al., 1999; Rambourg et al., 1974), together with ‘‘pseudo’’ 3D studies by
analysis of stereo pairs of 2D images of thick specimens tilted in the EM—carried
out from the early 1970s through the late 1990s—contributed enormously to the
body of knowledge about the 3D organization of Golgi and endoplasmic reticulum
(ER) membranes (Hama et al., 1994; Lindsey and Ellisman, 1985; Rambourg and
Clermont, 1997). However, all of these studies were ultimately frustrated by the
fact that the superimposition of 3D information in the 2D images restricted
resolution along the z-axis to the thickness of the section at best (Marsh, 2005).
The application of high-resolution ET to the study of Golgi membranes and
associated tubules and vesicles in mammalian cells, first in conventionally
prepared (Ladinsky et al., 1994), then in fast-frozen/freeze-substituted normal rat
kidney (NRK) cells (Ladinsky et al., 1999), convincingly demonstrated the utility
of this approach for providing novel insight into dynamic traYcking events at the
Golgi complex and, in particular, at the exit face of the Golgi—the trans-Golgi
196 Brad J. Marsh
network (TGN) (GriYths and Simons, 1986). This ability to truly dissect the
resultant 3D information voxel-by-voxel and/or in any desired orientation (see
Fig. 2 in Ladinsky et al., 1994) revealed the underlying power of this technique for
starting to address long-standing questions related to Golgi structure–function
relationships.
At the inception of our own work, and as noted above, 3D EM studies of the
Golgi and associated subcellular compartments had been performed in a number
of cell types (Katsumoto et al., 1991; Ladinsky et al., 1994; Rambourg and
Clermont, 1997; Tanaka et al., 1986), but not for the insulin-secreting beta cells
of the pancreas. Most of what is currently accepted as ‘‘conventional wisdom’’
regarding the mechanisms for insulin granule formation/maturation, traYcking
and release emerged from conventional 2D EM surveys of thin sections cut from
chemically fixed beta cells in the 1970s and 1980s (Howell and Bird, 1989; Howell
and Tyhurst, 1984; Orci, 1986; Orci et al., 1988). However, disparities had started
to appear between some of the key concepts regarding structure–function relation-
ships in the beta cell that came out of those early structural studies, and more
recent advances in understanding the molecular mechanisms underlying insulin
granule biogenesis and exocytosis (Arvan and Castle, 1998; Goodge and Hutton,
2000; Guest et al., 1997; Kowluru and Morgan, 2002; Kuliawat and Arvan, 1994;
Molinete et al., 2000; Rutter, 1999; Tsuboi and Rutter, 2003).
Thus, in light of the dramatic improvements that had taken place since the late
1980s in terms of methods for high-quality/reliable sample preparation (Gilkey
and Staehelin, 1986; McDonald and Morphew, 1993), instrumentation for EM in
general and ET in particular, and hardware/software for complex 3D analysis of
biological data (Kremer et al., 1996; Mastronarde, 1997), we felt the time was right
to revisit questions regarding precisely where and how are insulin secretory gran-
ules formed, and how changes in the organization of the Golgi and other subcel-
lular membranes under key physiological conditions relate to structure–function
relationships in the beta cells of the endocrine pancreas.
II. Methods and Materials
For our own studies of the insulin biosynthetic pathway, it was clear from the
outset that to attain the most reliable insights into the membrane traYc itinerary
of insulin in the pancreatic beta cell, our best approach would be to closely follow
the example of Ladinsky et al. (1999). Ladinsky and colleagues employed a
combination of the then disparate techniques of fast-freezing/freeze-substitution
(followed by plastic embedment) with dual-axis ET of four serial semithick
(250 nm) sections. This modification of the standard single-axis tomographic
approach improves the resolution and symmetry of cellular structures in all 3D
(Mastronarde, 1997; Penczek et al., 1995; Taylor et al., 1984). The end result was
a comparatively large (�1 � 1 � 4 mm3) volume of the Golgi ribbon in an NRK
cell reconstructed at high (�7 nm) resolution, which demonstrated near optimal
8. Reconstructing Mammalian Cytomembranes by 3D EM 197
structural preservation and clarity throughout. From this, Ladinsky et al. were
able to computationally dissect Golgi membranes and associated vesicle and
tubule budding profiles with maximal confidence, and were aVorded a reasonably
large portion of the NRK Golgi ribbon within a single cell for analysis.
A. Mammalian Cell and Tissue Culture
Since ET of large cellular regions using the techniques that will be described below
still remains a relatively painstaking and labor-intensive process in comparison to
conventional ET, it has been crucial for us to ensure that cells and/or tissue to be
studied in detail are as healthy and as physiologically viable as possible. We initiated
our studies with the Syrian hamster-derived beta cell line, HIT-T15, because at early
passages it retains the capacity to synthesize and secrete insulin in response to
extracellular glucose (Fig. 1) (Marsh et al., 2001a; Santerre et al., 1981).
The fact that these cells are modestly granulated compared with beta cells in vivo
provided an advantage; it allowed us to conduct a detailed analysis of interac-
tions between the microtubule cytoskeleton and the key organelles involved in
membrane traYc/secretion of insulin (i.e., the ER, Golgi, and insulin granules)
(NovikoV et al., 1975; Rios and Bornens, 2003; Yorde and KalkhoV, 1987; seeFig. 2). We initially tried various techniques for rapid freezing, since immortalized
beta cell lines are usually cultured as a crude 2D monolayer in the absence of
supplementation with growth factors and/or extracellular matrix (Ohgawara et al.,
1995), and thus are theoretically thin enough to be amenable to plunge freezing.
However, we consistently found that a higher fraction of beta cells were well
preserved by high-pressure freezing. This technique is one of a number of diVerentfast-freezing methods that can be used to vitrify cells/tissue, so as to capture
cellular events in a nativ e, frozen- hyd rated stat e (see Chapt er 1 by Dubochet,
Chapter 2 by McDon ald, an d Chapt er 3 by Hess , this volume ).
Although immortalized beta cell lines are easily obtained and cultured, and
continue to serve as useful models for the study of the insulin secretory pathway
and diabetes (Gleason et al., 2000; Poitout et al., 1995), there are obvious and
inherent limitations to their usefulness. In general, they contain relatively low
levels of insulin (and low numbers of insulin granules) compared to normal adult
pancreas (Breant et al., 1992), and they respond poorly (in terms of granule release)
to physiologically relevant levels of glucose (Breant et al., 1992; Poitout et al.,
1996), suggesting that their mechanisms for sorting/processing insulin might diVersubstantially from beta cells in situ in intact islets. Additionally, islet beta cells are
polarized in vivo (Bonner-Weir, 1988; Orci et al., 1989b). Although transformed
beta cell lines’ cells can be induced to polarize experimentally (Cortizo et al., 1990),
they are typically not polarized when grown in culture. Such a loss of cell polarity
undoubtedly aVects cellular traYcking pathways (Lombardi et al., 1986).
Consequently, we modified our cryopreparative techniques for use with intact
islets of Langerhans obtained from the pancreata of adult, female Balb/c mice.
Care was taken to ensure that islets were as healthy as possible at the point of
Fig. 1 Organization of the Golgi region at high (�6 nm) resolution and in 3D in a mammalian cell
that makes and secretes insulin in a regulated manner (Marsh et al., 2001a) (Copyright 2001, National
Academy of Sciences, USA). (A) A single 2D image of a 400-nm plastic section cut from an immorta-
lized rodent beta cell line (HIT-T15) prepared by high-pressure freezing and freeze-substitution,
followed by plastic infiltration/embedding. This image is one of 80 such images that together comprise
a ‘‘tilt series,’’ collected for each axis by imaging the specimen every 1.5�as it was tilted over a range of
�60�. These images were aligned using 10-nm gold fiducials, as described in the chapter by O’Toole, this
volume. (B) The tomograms calculated from each set of aligned tilts are then brought into register and
combined in 3D to produce a single, dual-axis reconstruction. To enable the study of a larger portion of
the Golgi ribbon this process was repeated for multiple 400-nm sections, and the serial
reconstructed volumes were aligned with each other to create a single, large (�3.1 � 3.2 � 1.2 mm3)
volume reconstruction. Once all of the visible structures in the Golgi region had been modeled using the
IMOD software package (Kremer et al., 1996), any given (modeled) object could be studied closely
in 3D either alone or in context with any other object(s). (C) The Golgi complex with seven cisternae
198 Brad J. Marsh
8. Reconstructing Mammalian Cytomembranes by 3D EM 199
cryopreservation by first culturing them under normal conditions either for several
hours or overnight after isolation. This step allows the isolated tissue to recover
following the isolation procedure itself (Sandler and Andersson, 1984), and
promotes both the viability (e.g., glucose responsiveness, protein synthesis, and
insulin secretory capacity) and integrity of isolated islets by allowing repair/
reformation of the connective tissue/collagen capsule at the periphery of rodent
islets (Bonner-Weir, 1989; Wang and Rosenberg, 1999). Islets were not cultured or
used for experiments more than 72 h following isolation.
Fig. 2 3D display of part of the Golgi ribbon shown in Fig. 1 revealing the in situ physical relation-
ships between the cis-most cisterna and the microtubule cytoskeleton. (A) Microtubules closely follow
and occasionally form contacts with the membranes of the cis-most cisterna. Note that in the modeled
region, microtubules do not exhibit a radial organization. Scale bar ¼ 500 nm. (B) A higher-
magnification view oriented to show that the paths of some microtubules closely follow the membranes
over considerable distances. Microtubules traversing the Golgi stack can also be observed. Original
images are from Marsh et al. (2001a) and are reproduced with permission from Proceedings of the
National Academy of Sciences USA.
(cis–trans: C1–C7) is at the center. The color coding is as follows: C1, light blue; C2, pink; C3, cherry
red; C4, green; C5, dark blue; C6, gold; C7, bright red. The Golgi is displayed in the context of all
surrounding organelles, vesicles, ribosomes, and microtubules: ER, yellow; membrane-bound ribo-
somes, blue; free ribosomes, orange; microtubules, bright green; dense core vesicles, bright blue;
clathrin-negative vesicles, white; clathrin-positive compartments and vesicles, bright red; clathrin-
negative compartments and vesicles, purple; mitochondria, dark green. Two views of the modeled
(segmented) Golgi region are provided, rotated 180�with respect to each other around the vertical
axis. This figure has been reproduced with permission from: ‘‘Lessons from tomographic studies of the
mammalian Golgi, Biochim. Biophys. Acta (2005), 1744, 273–292.’’ Scale bars ¼ 500 nm.
200 Brad J. Marsh
B. Fast-Freezing and Freeze-Substitution
As noted earlier, to preserve membrane structure and organization in beta cells
with high fidelity, we initially employed two independent techniques for cryoim-
mobilization: plunge freezing (Ladinsky et al., 1999) and high-pressure freezing
(McDonald and Morphew, 1993). While plunge freezing yielded some improve-
ment in the preservation of cellular ultrastructure over conventional chemical
fixation methods, more reproducible results were generally obtained by high-
pressure freezing. This is likely due to the fact that this latter technique inhibits
ice crystal growth in thicker samples, even though the rate of cooling is lower than
with plunge freezing (Gilkey and Staehelin, 1986;McDonald andMorphew, 1993).
Importantly, no major diVerences could be discerned in secretory granule, vesicle
or Golgi morphologies between plunge and high-pressure frozen cells, assuring
us that high pressure per se did not appreciably perturb cellular membranes or
architecture.
For immortalized beta cell lines such as HIT-T15 that essentially grow as mono-
layers, cells were seeded onto small plastic chips cut from ThermanoxÔ coverslips
(Nalge Nunc International, Rochester, NY) and cultured for 2–3 days prior to
experiments (Marsh et al., 2001a). Immediately prior to freezing, each chip was
transferred into a small brass, sandwich device comprised of two, interlocking
halves referred to as a ‘‘planchette’’ (Swiss Precision, Inc., Millbrae, CA), also
prewarmed to 37 �C. The half of the planchette holding the plastic chip on which
the cells had grown was filled with regular medium warmed to 37 �C and buVeredwith HEPES (10 mM). The second half of the planchette was filled with medium
containing 10% dialyzed Ficoll (Sigma, St. Louis, MO) as an extracellular cryoprotec-
tant, and was placed on top of the half of the planchette containing the cells. After
brief (<10 sec) manipulation into the tip of the planchette holder, cells were frozen
within 10–20 msec under high pressure (�2100 atm) using a HPM 010 high-pressure
freezer (BAL-TEC AG, Balzers, Liechtenstein). Frozen specimens were stored under
liquid nitrogen. Water was removed from these specimens by freeze-substitution with
anhydrous acetone (Ernest F. Fullam, Inc., NY) containing 0.5% glutaraldehyde/0.1%
uranyl acetate (UA) at �90�C for 1–2 days, followed by substitution with acetone
containing 1% OsO4/0.1% UA (Electron Microscopy Sciences, PA) at �70 �C.Specimens were allowed to warm to 0 �C over 2–3 days, whereupon they were
rinsed five times with anhydrous acetone. Samples were then warmed to room
temperature, infiltrated with increasing concentrations of Embed812-Araldite
resin (Electron Microscopy Sciences, PA), and flat-embedded between two Teflon-
coated glass microscope slides (Miller-Stephenson Chemical Co., Inc., Sylmar,
CA; VWR Scientific, Inc., West Chester, PA). Resin was polymerized at 60 �C in
a dry oven over 1–2 days.
Isolated islets cultured overnight as a suspension of free-floating cell clusters were
similarly maintained at 37�C on a humidified heating block in culture medium
buVered by the addition of 10-mMHEPES immediately prior to freezing. Typically,
10–30 islets (depending on size) were gently manipulated into one-half of the
8. Reconstructing Mammalian Cytomembranes by 3D EM 201
prewarmed holder prefilled with HEPES-buVered medium containing fetal bovine
serum (FBS) 10% (v/v) (Sigma, St. Louis, MO). All manipulations were carried out
on ParafilmÒ placed on top of an inverted heating block warmed to 37�C under a
dissecting microscope. The second half of the planchette, filled with RPMI con-
taining 10% dialyzed Ficoll (MW ¼ 70 kDa) and 0.5% Type IX ultra-low temper-
ature gelling agarose (Sigma, St. Louis, MO) as extracellular cryoprotectants, was
then placed on top of the half of the sample holder containing the islets. The pair of
interlocking hats was then secured into place at the tip of the specimen holder and
the islets frozen under high pressure, freeze-substituted, and plastic embedded
essentially as described previously and above (Marsh et al., 2001a, 2004).
C. Microtomy and Grid Preparation
Well-preserved beta cells (in the case of beta cell lines) or islets were individually
identified by phase-contrast light microscopy, excised from the resin, and remounted
onto plastic stubs.HIT-T15 cells on plastic chips weremounted in an orientation that
permitted en face sectioning; orientation is irrelevant for intact islets. Ribbons of
thin (40–60 nm) or thick (300–400 nm) sections were cut on a microtome (Leica
Microsystems, Vienna, Austria) for conventional 2D survey at 80–120 kilo electron
volt (keV) to assess the quality of cell/islet preservation and to select regions for
subsequent study or for high-tilt ET on instruments operating at higher voltages
(�300 keV) (see below), respectively (Marsh et al., 2001a, 2004). Ribbons of serial
thick sections were collected onto Formvar-coated copper (2 � 1 mm) slot grids
(Electron Microscopy Sciences, Hatfield, PA) and poststained with 2% aqueous
UA or 3% UA in 70% methanol (15 min) and Reynold’s lead citrate (3 min). These
samples often require an additional carbon-coating step to minimize charging/
movement in the electron beam, particularly when tilted for tomography (Marsh,
2005). Colloidal gold particles (10 nm) were then deposited on both surfaces of
these sections for use as fiducial markers during subsequent image alignment.
D. Basic Instrumentation Requirements for ET of Thick Plastic Sections
Atvarious points in our studies,wehave successfully viewed and collected tilt series
images from ribbons of equivalently stained 300- to 400-nm-thick sections using (in
decreasing order of operating voltage) a JEM-1000 high-voltage EM (HVEM)
operated at either 500 keV, 750 keV, or 1 million electron volt (MeV) (JEOL USA,
Inc., Peabody,MA), a JEM-4000FX intermediate-voltageEM(IVEM) (JEOLUSA,
Inc.) operated at 400 keV, and with a Tecnai F30 field emission gun (FEG) IVEM
(FEI) operated at 300 keV. Regardless of operating voltage, the grid holding the
specimen was tilted in a eucentric goniometer typically at either 1� or 1.5� intervalsover a range of 120�–140� about two orthogonal axes (Mastronarde, 1997), as
summarized in the legend that accompanies Fig. 1. In earlier tomographic studies
of the Golgi region in mammalian cells (Ladinsky et al., 1994, 1999; Marsh et al.,
2001a), single images taken at each tilt were collected on film (SO163, Kodak
Eastman Company, or 23D56, Agfa-Gevaert NV), with the negatives subsequently
202 Brad J. Marsh
digitized on a motorized light table using a high-resolution STAR-1 cooled charge-
coupled device (CCD) camera (Photometrics, Ltd., Tucson, AZ), yielding a final
pixel size of �2.3 nm. Although the processing of film and digitizing of EM
negatives is labor intensive and time consuming, this procedure inherently aVordedthose investigating extended membrane structures, like the Golgi, a much larger
field of view for a ‘‘single snapshot’’ at each tilted view than could be accomplished
otherwise without sacrificing resolution by imaging at a lower magnification to
overcome the smaller field of view of a typical CCD (Ladinsky et al., 1999; Marsh
et al., 2001a). Even with film, however, to take advantage of as much of the image
area on the EM negative as possible required the user to digitally tile a larger image
from ‘‘montaged’’ image arrays in x and y with defined areas of overlap (typically
10% of the pixel dimensions of each piece of the montage), using software to
precisely control the digitizing camera as well as movement of the motorized
light table (Fig. 1).
E. Acquisition of Digital Tilt Series from Large Cellular Areas by Automated IVEM/HVEMMontaging Tomography
Thick sections of 300–400 nm were typically imaged at 12,000� using a JEM-
1000 HVEM operating at 750 keV (JEOL USA) or at 15,500�, 20,000�, or
23,000� using a Tecnai F30 IVEM (FEI). Motorized, tilt-rotate specimen holders
(Models 650 and CT3500TR; Gatan, Pleasanton, CA) were available on both
kinds of microscopes. Tilt series data were digitally recorded using semiautomated
methods for CCD data acquisition, image focus, and alignment as the sections
were serially tilted through 1� or 1.5� increments over a range of 120�–140� abouttwo orthogonal axes. The camera and microscope were controlled by the program,
SerialEM (Mastronarde, 2005).
To best understand complex spatial and structural relationships for extended
cytomembranes, such as the Golgi or ER, it is important to examine a large
cellular area. We overcame the limited field of view typically aVorded by CCD
imaging on the EM by using a variation of the methods for digital montaging of
tiled image arrays, discussed above. This required the automation of image
montaging using image shift directly in the EM, which is now implemented in
the SerialEM data collection package (Mastronarde, 2005). Briefly, a set of images
was collected at diVerent coordinates in x and y at each tilt, using the software to
automatically reposition and capture each individual image piece, yielding a
montage of overlapping frames at each tilt (Fig. 3A). For our studies of the
Golgi region in insulin-secreting cells, we have typically used montages of either
2 � 2 or 3 � 3 panels in x and y (depending on the magnification and thus the final
CCD pixel size) to accommodate a total cellular area �4 mm2 per reconstruction
(Fig. 3).
Although there is a concomitant increase in the time required to digitally collect
a montaged tilt series, overall the yield of data for the time invested is higher with
this procedure (Mastronarde, 2005). For example, a 3 � 3 montage with images
8. Reconstructing Mammalian Cytomembranes by 3D EM 203
collected every 1.5� over an angular tilt range of 120� (�60�) means an increase in
the number of individual image frames from 81 to 729, but the steps of tilting,
focusing, and tracking are performed more or less as they would be for a single
panel tilt series rather than for every frame. The largest additional temporal cost
related to data collection/image acquisition—aside from the CCD acquisition
time—is usually due to the additional brief (anywhere from 0.5 to 4 sec) preirra-
diation required just prior to image capture for each piece of the montage, to
combat specimen charging that is often inherent to tomography of thick, stained,
plastic sections.
F. 3D Reconstruction of Large Cellular Areas
The computer software package IMOD comprises a set of image processing,
modeling and display programs used for tomographic reconstruction from EM tilt
series, as well as for 3D reconstruction of EM serial sections and optical sections
(Kremer et al., 1996). Briefly, the package contains tools for assembling and
aligning data within multiple types and sizes of image stacks, viewing 3D data from
any orientation, modeling and display of the image files, and to obtain accurate
quantitative information in either 2D or 3D. IMOD was developed primarily by
David Mastronarde, Rick Gaudette, Sue Held, and Jim Kremer at the Boulder
Laboratory for 3D Electron Microscopy of Cells.
Using the IMOD package that incorporates the eTomo graphical user interface
for the alignment and tomographic reconstruction of digitally tiled images, indi-
vidual pieces of the montaged tilt series images acquired for each of the two ortho-
gonal axes were first brought into register with one another by cross-correlation of
overlapping image data to determine the relative displacements in x and y between
adjacent frames for a given tilted view (Fig. 3A, B, and B0). Occasionally it was
necessary to manually correct errors in correlation-based frame registration using
the computer tool MIDAS to ensure precise positioning of individual pieces of the
montage, and/or to account for any shifts in x and/or y that were too large to be
handled automatically. Likewise, cross-correlation was used to calculate the set of
translations in x and y that needed to be applied from one tilted view to the next
through the stack of tilt series images to generate a crudely prealigned stack.
Subsequently, individual frames or pieces of the montage were ‘‘blended’’ across
the zone of overlap, using redundant image data in the adjacent image pieces and
fused to yield a single image for each tilted view (Fig. 3C and C0).The images of the now-blended tilt series were then treated essentially like a
typical image stack and more accurately aligned by tracking the positions of �200
of the 10-nm gold fiducial markers on the top and bottom surface(s) of the
sections, preferentially with an equal number of fiducials on both surfaces. Ideally,
most of the fiducial points selected and tracked for the first axis are also tracked
for the second axis (Fig. 3C, C0, D, and D0). This step is accomplished using the
program transferfid, which automatically determines the most reliable estimate
of how fiducial markers correspond between the tilt series collected around the
Fig. 3 Digital montaging and image alignment of large areas for ET begins with data collected on the
IVEM as set of overlapping pieces or frames in x and y, collected over a large number of tilts over a
range of 120�–140
�, and around two orthogonal axes (Mastronarde, 2005). (A) This diagram illustrates
how image shift is used to move from one position to the next in a regular array in x and y (e.g., x1,y1,z1;
204 Brad J. Marsh
8. Reconstructing Mammalian Cytomembranes by 3D EM 205
two axes. transferfid rotates the ‘‘seed’’ fiducial model generated for a single tilt
view (typically close to, or at, 0� tilt) from the first axis by 90� (trying first clockwisethen counterclockwise to find the best fit) and searches for the pair of views from
each tilt series that has the best fiducial correspondence. The program then
transfers the fiducial model from the first series to generate a seed model for the
second axis. The program also indicates which fiducials failed to transfer and
which corresponding fiducials should be specified when setting up to combine the
tomograms from each axis (Fig. 4), a particularly crucial step for dual-axis recon-
struction, since the coordinates of corresponding fiducials are used for the initial
calculation of 3D rotation, distortion, and shift between the volumes.
A large number of fiducials is required for accurately tracking and aligning large
area tilt series data because of the heterogeneous distortions that result from
specimen exposure to the electron beam. In our experience, the selection of
�200 fiducials for a cellular area �4 mm2 provides a suYciently high ratio of
known versus unknown parameters to accurately calculate image alignment while
accounting for diVerential section thinning, specimen shrinkage, and minor bend-
ing. While such distortions may be negligible when handling tilt series data collect-
ed from a relatively small area, they are critically important to solve for when
attempting to accurately reconstruct large cellular regions (i.e., >3 mm in x and y)
at high (5–7 nm) resolution (Marsh et al., 2001a). To deal with these issues, we
modified the tilt alignment procedure so that subsets of fiducial markers were used
to solve for local alignments in a limited area. tiltalign first calculates a single global
distortion solution by employing all of the fiducial markers to solve for tilt angle,
rotation, overall magnification, and linear distortion within the plane of the
section, including distortion due to specimen thinning in the electron beam. The
program next finds a series of solutions in a regular array of overlapping local areas
x1,y2,z1; x1,y3,z1; x2,y1,z1; x2,y2,z1; and so on), acquiring an image at each position, for a given tilt.
For a montage with an odd number of pieces such as a 3� 3 array, an initial image of the central piece in
the array (i.e., at position x2,y2) is taken and used as a reference for the accurate alignment/positioning
of each piece of the montage. Subsequent precise registration of individual pieces in IMOD is achieved
by the correlation of image data in zones of overlap (shaded, typically set at around 10% of the total
pixel area of each frame) between adjacent frames (see below). (B) Raw montaged images shown for
0�tilt. (B0) A comparable image from the second tilt series which is taken after the specimen has been
rotated 90�counterclockwise. When viewed in 3dmod, the crude registration of elements in the montage
can be seen because of the minor diVerences in contrast between adjacent pieces and/or minor displace-
ment by up to several pixels (compare B and B0). Once individual frames have been blended across the
regions of overlap and fused to generate a single image for each tilted view, such borders can be no
longer distinguished. A large number (�200) of colloidal gold markers are selected from images in the
first tilt series (C and D), then transferred to the corresponding view from the second (C0 and D0). Theseprovide the information necessary to solve for distortion (including rotation, translation, and magnifi-
cation) and accurately transform and align individual tilted views in the series. They also help in
matching and combining the volumes from each axis to make a dual-axis tomogram (see also Fig. 4).
Scale bars ¼ 1 mm.
Fig. 4 Matching and combining the tomograms generated from data collected around two orthogo-
nal tilt axes depends on knowing the displacement between corresponding subvolumes in the two
tomograms. The coordinates of the fiducials that correspond between the two tomograms allow an
initial determination of rotation, distortion, and translation between the two tomograms. However,
to combine two volumes precisely one must generate a set of nonlinear transformations by cross-
correlating ‘‘patches’’ of a defined size between the first volume and the transformed second volume
at an array of positions in 3D. The displacements between the two volumes at each position, which are
the output from this process, can be visually examined in the form of a ‘‘patch vector model.’’(A) Such
models are often highly informative, since each displacement vector is visualized as an exaggerated line
whose length is 10 times the actual length of the original displacement vector. Careful examination of
the patch vector model usually provides the user with visual cues as to the nature of the transformations
necessary to bring the volumes into register. When combining large cellular areas from montaged tilt
series, the large displacements seen in this figure typically result from the use of patches that are too
small to provide suYcient information for a good correlation between the corresponding subvolumes.
Although standard patches/subvolumes of 80 � 80 � 40 pixels would normally provide a good measure
of agreement between tomograms, for the large area generated from a 3 � 3 montage used in this
example, an increase in patch size in z (i.e., to 80 � 80 � 70 pixels) dramatically improved the resulting
displacement model (B). However, as noted in Section II, it is essential to exclude from the combine
206 Brad J. Marsh
8. Reconstructing Mammalian Cytomembranes by 3D EM 207
or ‘‘patches’’ for each 2D projection. These local solutions are incremental to the
global solution, so all of the alignment parameters are constrained to vary slowly
through the tilt series. The positions of the fiducial markers in 3D are fixed at
the values found in the global solution, as the dual-axis tomographic approach
requires that the two reconstructed volumes must be accurately aligned with one
another (Mastronarde, 1997) (Fig. 4). These constraints dramatically reduce the
number of parameters that have to be solved for and increase the ratio of measure-
ments to unknowns, thus allowing accurate local solutions to be obtained from as
few as six gold fiducial particles. tiltalign then produces a set of linear transforma-
tions based on the global solution; these are used to produce globally aligned
images. For each local area, it also produces a set of refining transformations
based on accurately aligning the images for a reconstruction of that area alone.
An R-weighted back-projection program then uses the refining transformations
to determine which pixel in a projected image back projects to a given voxel in
the reconstructed volume (Wilson et al., 1992). Tomographic volumes calculated
by R-weighted back projection from each set of aligned tilts are then matched to
each other in 3D by nonlinear transformation using the program patchcrawl3d.
Basically, patchcrawl3d determines the 3D displacement between the two volumes
at a regular array of positions using sets of patches of defined sizes in x, y, and z,
and then runs the program corrsearch3d to determine the displacement of each
patch in one tomogram relative to the corresponding patch in the reference
tomogram (typically the tomogram from the first axis is used as the reference
volume) (Fig. 4). As for fiducial tracking and image alignment for montaged tilt
series data, we have had to modify the standard approaches to determine the
optimal displacement solutions for matching and combining tomograms of large
cellular areas.
Although relatively small patches of overlap will provide suYcient information
for accurate combine solutions for small image areas, larger patch sizes are gener-
ally required to accurately match one volume to another for the large image areas
that result from digitally montaged tilt series. For an example of the dramatic eVectof modest changes in patch size specification during tomogram matching and
combination, compare the ‘‘patch vector models’’ presented in Fig. 4A and B.
It is also important to identify if there are regions within the areas to be combined
that should not be used for correlating the tomograms, either because they do not
contain suYcient material (e.g., the lumen of a vacuole) or because the reconstruc-
tion quality is poor (e.g., typically at the corners of large tomograms) (Fig. 4B).
solution erroneous vectors that result from areas in the tomogram that do not contain suYcient dense
material for correlation, or from areas at the edges or corners that demonstrate poor reconstruction
quality. (B) Such wayward vectors are evident in regions of the correlation model that correspond to the
lumens of vacuoles (e.g., Figs. 3B and C and 5A) and at the corners of the model. (C) After manually
editing out bad vectors from the model, only the remaining vectors will contribute to the correlated
combine solution for matching the two volumes accurately in 3D.
208 Brad J. Marsh
The patch vector model can then be edited so these areas are excluded from the
combine solution (see Fig. 4C and accompanying legend). Finally, volumetric
image data from each of the two tomograms is combined to produce a single,
high-resolution, dual-axis 3D reconstruction (Mastronarde, 1997) (Fig. 5A).
To further increase the cellular volume available for analysis, 3D reconstruc-
tions calculated in this manner from serial plastic sections can then be aligned to
one another in z as previously described (Ladinsky et al., 1999; Marsh et al.,
2001a), to produce a final single large volume for analysis and segmentation. The
interactive program MIDAS was used to generate a general linear transform that
accounted for rotation, distortion, and stretch from one section to its nearest
neighbor based on the best-fit alignment of subsets of slices extracted from the
bottom of one tomogram and the top of adjacent tomogram.
G. 3D Segmentation and Quantitative Analysis
Membranes of the Golgi complex, ER, compartments of the endosomal–
lysosomal pathway, and associated tubules/vesicles within the tomographic
volumes were segmented, extracted, and viewed with the IMOD package using
3dmod (Kremer et al., 1996). Typically, cellular tomograms are stored and viewed
as stacks of pixel-thick slices, oriented parallel to the plane of section. However,
one is also able to use the slicer tool that is part of the program 3dmod to arbitrarily
rotate the data in x, y, and z to visualize membranous connections between com-
partments or between membrane-bound compartments and the microtubule
cytoskeleton with minimal ambiguity (O’Toole et al., 1999; Marsh et al., 2004).
Although the pixel size—and hence the tomographic slice thickness—are deter-
mined by the magnification at which the data were collected (and whether the
image data were binned or interpolated), the actual depth of biological material to
which these data correspond is somewhat larger, since sections cut from plastic-
embedded specimens collapse on initial exposure to the electron beam (Kremer
et al., 1990; Luther et al., 1988; Mastronarde, 1997). Despite the fact that each
tomographic slice resembles a conventional EM image or ‘‘micrograph,’’ it corres-
ponds to slices of material far thinner than can be cut by ultramicrotomy. For
example, for a tomographic data set with a nominal pixel size of 2.3 nm, a single
pixel-thick slice parallel to the plane of section actually corresponds to �3.8 nm of
material prior to the specimen collapse (Marsh et al., 2001a). Consequently, we
typically rescale 3D surface-rendered models generated from segmented cellular
tomograms by an appropriate amount in the z dimension to account for this
phenomenon, and to more accurately represent and quantify the topology of the
structures in the specimen prior to data collection in the EM (Ladinsky et al., 1999;
Marsh, 2005).
Features of interest to us, such as membrane bilayers and microtubules con-
trasted with heavy metals, are usually modeled by an expert user who manually
segments/traces the membrane contours of an organelle or compartment through
Fig. 5 (A) Part of the extended Golgi ribbon of a glucose-stimulated islet beta cell is evident as the
‘‘stacks’’ of Golgi cisternae that wind through the cytoplasm in the dual-axis tomogram computed from
the data represented in Fig. 3, which yielded a reconstructed cellular volume of 5.92 � 5.92 � 0.35 mm3.
Tilt series images were acquired as a 3� 3 digital montage at 1�increments over approximately�63
�on
a Tecnai F30 IVEMoperated at 300 keV using a motorized tilt-rotate holder (Gatan). Individual images
at each tilt were brought into register by cross-correlation of image data in zones of overlap between
adjacent frames, and blended to generate a single image at each tilt. Tilt series images were first aligned
with one another by cross-correlation to yield a prealigned stack, and were then accurately align-
ed by tracking the positions of 10-nm-gold fiduciary markers as discussed in Section II. The tomograms
calculated from each set of aligned tilts were then brought into register and combined in 3D to produce
the single, dual-axis reconstruction presented here. (B) Membranes of the Golgi, insulin granules, and
other compartments were modeled by manual segmentation of the membrane bilayer through sequen-
tial tomographic slices (each 1.03-nm thick) in z using the program 3dmod that is part of the IMOD
software package. (C) Polygonal meshes fitted between adjacent contours provided surfaces both for
visualization and quantification in 3D. In addition to the stacked cisternal membranes of the Golgi that
are color coded as described in the legend for Fig. 1, a number of immature granules and/or condensing
vacuoles are visible (colored in light blue). Scale bars ¼ 1 mm.
8. Reconstructing Mammalian Cytomembranes by 3D EM 209
210 Brad J. Marsh
successive slices in z, or along an arbitrary plane in slicer, using computer tools in
the 3dmod program. Since each tomographic data set is stored arbitrarily as a set
of 2D slices in x and y (1 pixel thick in z), the user physically traces a given
membrane-bound compartment through successive slices in z (Fig. 5B). Triangular
or polygonal meshes fitted between adjacent contours then provide data for surface
meshing and visualization (Kremer et al., 1996; Ladinsky et al., 1999). Typically,
diVerent colors are used to define membrane contours that belong either to a given
compartment or compartment type. For example, in the case of a typical Golgi
region in one of our beta cell tomograms, a particular color might be used to
identify the membranes of a single continuous cisterna that occupies a certain
position or hierarchical level in the stack. The same color would also be used to
identify cisternae that can be determined as either functionally equivalent or
occupying an equivalent position in stacks that are spatially distinct but remain
part of the same Golgi ribbon for a given cell (Ladinsky et al., 1999; Marsh et al.,
2001a) (Fig. 5C).
In addition to providing surface data that can be used to generate a ‘‘skin’’ that
can simply be visualized in 3D, triangular meshes fitted between contours allow
one to compute quantitative parameters, such as the surface area and volume for
any segmented compartment, including accurate measurement of fenestrations or
holes in Golgi membranes (Ladinsky et al., 1999; Marsh et al., 2001a,b; Otegui
et al., 2001). In our case, since contours were drawn down the middle of the
membrane bilayer, the interior compartment volumes were computed by subtract-
ing the surface area � 3.5 nm (half of the membrane’s thickness) from the volumes
inside the contours. To ensure that quantitative data generated from 3D models
are as accurate as possible, the ends of closed compartments and vesicles are
‘‘capped’’—so that they are closed oV completely with a surface mesh—as well as
‘‘smoothed.’’ Smoothing routines work by fitting local polynomials to the sur-
faces and replacing each point with a corresponding point from the fitted surface
(Ladinsky et al., 1999). This operation helps to compensate for minor irregularities
in hand-drawn contours, due to user error, and results in surfaces that shift
smoothly from the plane of one contour to the next.
As noted above, and presented in Figs. 1 and 2, individual microtubules could
also be tracked in 3D as a set of points with a curvilinear trajectory to characterize
and quantify in situ associations between the microtubule cytoskeleton and the
membrane surface of diVerent organelles and compartments. This process was
frequently aided by use of the slicer tool in 3dmod, which allows the user to reorient
the data until the microtubule can be viewed clearly en face (for more detailed
explanation of the use of slicer for tracking and visualizing microtubules, please see
Chapt er 5 by O’Tool e, Chapt er 6 by Ho o g and Antony , and Chapt er 9 by Ote gui
and Austin, this volume).
Once all objects of interest have been segmented in the cellular reconstructions,
spatial relationships among the modeled objects in 3D can be quantified by
measuring distances between objects within a plane and computing an average
density of neighboring items as a function of distance between objects, using the
8. Reconstructing Mammalian Cytomembranes by 3D EM 211
programs nda and mtk (Marsh et al., 2001a). This is similar to previous quantita-
tive morphometric approaches to study the distribution of insulin granules and
spatial relationships with the beta cell microtubule cytoskeleton (Yorde and
KalkhoV, 1987). This technique, referred to as ‘‘neighbor density analysis,’’ re-
quires that each object of a given geometrical type (i.e., sphere, line, or complex/
convoluted surface defined by a mesh of triangles) serve in turn as a reference from
which the distances to all nearby neighboring objects of a particular kind are
measured (McDonald et al., 1992). Quantification of neighbor density was obtained
by dividing the number of items at a given range of distances by the approximate
total volume at those distances from all such reference objects, where volume
was estimated from the size of a spherical shell at that distance from the object
(Marsh et al., 2001a).
III. Discussion
The development and continued advancement of methods for reconstructing large
cellular areas in 3D at comparatively high resolution (5–7 nm) have been a prerequisite
forEMtomographic studies of themammalianGolgi ribbon,which frequently extends
over distances greater than 10 mm (Cooper et al., 1990; Rambourg and Clermont,
1997). As discussed above, these methods evolved from earlier tomographic studies
of large regions of the mammalian Golgi that were originally acquired on film and
subsequently digitized to provide a large cellular area for 3D reconstruction and
analysis. Key to the success of this method is the use of protocols that retain
suYcient resolution to readily identify protein coats such as clathrin (Ladinsky
et al., 1999; Marsh et al., 2001a). Importantly, these studies forced the develop-
ment of new mathematical approaches for handling large area tilt series data and
attempting to deal with the heterogeneous changes in specimen geometry that
occur over large areas; these had previously made it diYcult to accurately align
the image data without significantly compromising tomogram quality and resolu-
tion (Marsh, 2005; Marsh et al., 2001a). These methods have led directly to
tomographic studies of other extended cytomembranes such as somatic and
syncytial-type cell plate formation in plants (Otegui et al., 2001; Segui-Simarro
et al ., 2004 ) (also , see Chapter 9 by Otegui and Austin, this volume ). With large
format (>2K � 2K pixels) CCD cameras now readily available—and in fact often
the default for ET—tomographic studies of large cellular areas (even without
image montaging) are now more commonplace. Consequently, the application of
methods for dealing with distortions in image data over large areas has become a
requisite for the generation of high-quality cellular tomograms (Marsh et al.,
2001a).
The specific application of these techniques to elucidate structure–function
relationships among the key organelles involved in the biosynthesis and traYcking
of insulin, first in an immortalized cell culture model (Marsh et al., 2001a,b) and
then in glucose-stimulated beta cells preserved within intact pancreatic islets
212 Brad J. Marsh
isolated frommice (Marsh et al., 2004), has provided a number of new insights into
mammalian cell biology more generally, and the insulin biosynthetic pathway in
particular. In some instances, these high-resolution 3D data have provided the first
unequivocal evidence for concepts that had been previously received with skepticism
by the molecular membrane traYc community. One such example is the idea that
multiple, distinct trans-cisternae are consumed in the process of sorting and pack-
aging cargo for exit, originally a cornerstone of NovikoV’s Golgi-ER-lysosome
(GERL) hypothesis (NovikoV, 1964). Associated data that have also provided
convincing and clear evidence for other key aspects of the GERL hypothesis, such
as the extent of intimacy between specialized regions of the ER and trans-Golgi
cisternae (Ladinsky et al., 1999; Marsh et al., 2001a,b), have thus provoked serious
reconsideration of the complexity of events at and near the trans-Golgi. Moreover,
these studies have provided compelling evidence that all three major mechanisms
of intra-Golgi transport (cisternal progression-maturation, vesicle- and tubule-
mediated traYcking) play important roles in intra-Golgi traYcking, demonstrat-
ing that they can act in concert in the same region of the Golgi ribbon, and
revealing that one mechanism may appear dominant over another depending on
the level of protein/lipid traYc and the physiological state of the cell/tissue (Marsh
et al., 2001b, 2004; Mironov et al., 1997). Tomographic studies remain underway
to elucidate whether this kind of structure–function variation occurs along the
Golgi ribbon within a single cell.
In some cases, data from our studies have raised important questions about
fundamental concepts regarding maintenance of the sequential processing hierarchy
that is a considered a hallmark of Golgi organization. Using physiologically relevant
concentrations of glucose to stimulate proinsulin biosynthesis and insulin secretion
from beta cells in situ in primary cultured mouse islets, we have unequivocally
demonstrated the existence of direct connections between nonequivalent cisternae
in islet beta cells stimulated to traYc large amounts of secretory protein cargo
(Marsh et al., 2004). With short-term stimulation by extracellular glucose (1 h),
proinsulin biosynthesis occurs exclusively at the translational level (Wicksteed
et al., 2003), resulting in a rapid wave (within �15–30 min) of proinsulin moving
into and through the Golgi (Alarcon et al., 1993). Although tubular connections
between Golgi cisternae at their periphery or at branches in the Golgi ribbon had
been described by others previously (Rambourg and Clermont, 1997; Weidman,
1995), we observed membrane tubules that connected nonadjacent cisternae by
reaching directly through holes or fenestrae in the neighboring cisterna to eVec-tively ‘‘bypass’’ interceding cisternae (Marsh et al., 2004). Such direct intra-Golgi
connections could provide a continuous lumen for the expedited forward transit of
large amounts of newly synthesized proinsulin between nonequivalent cisternae
that are normally distinct from one another, and may also facilitate the rapid
retrieval of retrograde traYc (Marsh et al., 2004). These data have provided some
of the strongest evidence to date for the concept of lateral transfer in the Golgi
complex, in contrast to sequential linear processing.
8. Reconstructing Mammalian Cytomembranes by 3D EM 213
A. Future Directions
Despite the fact that our ‘‘large’’ reconstruction of a Golgi region (Fig. 1) only
covered �1% of the volume of a beta cell, it provided a rare opportunity to
visualize the dense packing of subcellular organelles, ribosomes, vesicles, and
microtubules in the cytoplasm of a mammalian cell. This view cannot yet be
aVorded by any other method (Marsh, 2005; Marsh et al., 2001a). The continued
development of extended methods for the accurate tomographic reconstruction of
large cellular areas, to the point where it becomes possible to generate a complete
set of 3D spatial coordinates for a mammalian cell in toto at high resolution, is
likewise expected to provide important and unique insights into cellular organiza-
tion that cannot come from other methods or even from lower resolution ET of
whole cells. Such a ‘‘Visible Cell’’ project would lead to the generation of a cellular
atlas with suYcient resolution to distinguish all subcellular compartments and
filaments of interest at once. Such a Visible Cell atlas will likely play a crucial
role in advancing our understanding of (mammalian) cellular systems biology
(Bork and Serrano, 2005; Lehner et al., 2005; Nickell et al., 2006), and should be
seen as a fundamental prerequisite to realistically simulating/predicting the spatio-
temporal coordinates of complex molecular membrane traYcking events in silico.
Meanwhile, comparatively large area cellular reconstructions carried out at rela-
tively high throughput/high resolution through the use of large format/fast readout
CCD cameras will provide important new insights into other key aspects of insulin
traYcking in the beta cell that occur after secretory cargo packaging and insulin
granule biogenesis at the trans-Golgi. Specifically, current and future studies are
aimed at elucidating the specifics of granule recruitment to the cell surface for
release into the bloodstream after glucose levels in the blood rise above normal.
Quantitatively determining how interactions between granules and the micro-
tubule and actin cytoskeletons change in islet beta cells exposed to low, normal,
and stimulatory concentrations of glucose for diVerent times will be essential to
develop a complete understanding of these events (Fig. 6). Microtubules play an
important role in potentiating the second, sustained phase of insulin release by
transporting insulin granules from the Golgi region (where they are formed) to the
plasma membrane, to replenish the readily releasable granule pool following a rise
in intracellular [Ca2þ] (Boyd et al., 1982; Donelan et al., 2002; Malaisse et al.,
1975). This process is believed to be dependent on the recruitment of the cyto-
plasmic motor protein, kinesin, to the granule membrane (Balczon et al., 1992;
Donelan et al., 2002; Varadi et al., 2002). Evidence also suggests that the binding of
small GTP-binding proteins such as Rab3A and Rab27A to yet uncharacterized
‘‘eVector’’ proteins plays a crucial role in regulating these events in islet beta cells
(Coppola et al., 2002; Kajio et al., 2001; Park et al., 2002; Yaekura et al., 2003;
Yi et al., 2002). By tracking the frequency of such interactions along the length of
any given microtubule over a relatively large distance, combined with a capacity
to scrutinize interactions between the microtubule and granule membrane at high
resolution, we expect to precisely establish the sequence of steps and molecular
Fig. 6 Our early study of the insulinoma cell line, HIT-T15, demonstrated that we could dissect the
beta cell modeled in 3D at�6-nm resolution as required, that is, each modeled object could be extracted
and viewed in any chosen orientation and in the context of any other object(s) to analyze structural
associations with confidence (Marsh et al., 2001a). (A) Shown are the membranes that comprise the
Golgi, together with the ER, ribosomes, microtubules, secretory granules, vesicles, endosomal–
lysosomal compartment, and mitochondria (for color key see legend for Fig. 1C). In that study,
microtubules with a total combined length of over 80 mm were segmented to accurately quantify
associations between the microtubule cytoskeleton and diVerent organelles and compartments in situ.
In that analysis each microtubule was divided into regularly spaced (10 nm) subsegments, and the
distances from each segment to all neighboring objects of a particular kind were measured. These data
revealed the distances of closest approaches from each subsegment to other objects, which in turn gave
indications of which objects interact with microtubules and which do not. (B) The paths of the
microtubules (bright green) can be followed more readily, as they are displayed only in the context of
insulin secretory granules (bright blue) and endosomal–lysosomal compartments (purple and red),
which are also likely to play an important role in proinsulin processing/insulin traYcking (Turner and
Arvan, 2000). Here, we demonstrate the ability to pinpoint sites of close approach betweenmicrotubules
and insulin containing compartments in tomograms of beta cells. Using the programmtk, microtubules
and insulin granules shown in (C) were analyzed in 3D to determine all distances of approach between
them. After examining the histogram of distances of close approach between all microtubules and
granules, we limited our search of the modeled data to approaches on the order of �20 nm. This
identified the subset of microtubules highlighted in red in (D). (E) Using visible ‘‘markers’’ placed by the
software between the granules and the nearby microtubules, we were able to identify the structures at
sites of close approach (�20 nm) in the modeled data (F). Such sites represent the best chance of directly
identifying microtubule-associated proteins and cytoplasmic motors involved in microtubule-dependent
insulin transport in situ. Scale bar ¼ 500 nm.
214 Brad J. Marsh
8. Reconstructing Mammalian Cytomembranes by 3D EM 215
machinery involved in moving insulin from the peri-Golgi region to the cell
surface, and to better understand the nature of the subcellular defects that accom-
pany states of impaired insulin intracellular traYcking and exocytosis/secretion.
IV. Summary
In cooperation with David Mastronarde at the Boulder Laboratory for 3D Elec-
tron Microscopy of Cells, we have developed methods for high resolution, large
area ET of thick, stained plastic sections for the 3D reconstruction and analysis of
mammalian membrane architecture at high fidelity in the insulin-secreting beta cells
of the endocrine pancreas. This approach, when employed in conjunction with high-
pressure freezing and freeze-substitution, has allowed us to generate tomograms of
comparatively large cellular volumes (each typically measuring at least 4 � 4 �0.4 mm3) with which to study the key organelles and compartments involved in
the synthesis, processing, and traYcking of insulin with unprecedented reliability.
These data have aVorded us new insights into beta cell biology and function
that include: (1) the complexity of structural relationships among the Golgi, ER,
and compartments of the endosomal–lysosomal system (Marsh et al., 2001a);
(2) evidence that multiple transport mechanisms act in concert in the same region
of the Golgi ribbon (Marsh et al., 2001b); (3) a role for the ER in regulating
membrane traYc/sorting at the trans-Golgi, the presumptive site where proinsulin
is sorted and packaged into nascent secretory granules (Marsh et al., 2001a,b);
(4) evidence that multiple, distinct trans-cisternae, frequently referred to as the
TGN, detach and fragment as membrane is consumed in the process of packaging
secretory cargo for exit (Marsh et al., 2001a); and (5) the unequivocal demonstra-
tion of direct intra-Golgi connections that facilitate ‘‘cisternal bypass’’ to provide a
continuous lumen for the expeditious transit of Golgi traYc (Marsh et al., 2004).
Acknowledgments
We thank the entire staff of The Boulder Laboratory for 3D Electron Microscopy of Cells for their
personal support and professional assistance from the very onset of these studies, and Kathryn Howell
and John Hutton of the University of Colorado Health Sciences Center for additional tutelage and
intellectual support. This work was supported by P41-RR00592 to J.R.M., GM42629 and P01-
GM61306 to K. E. H., and a Juvenile Diabetes Research Foundation International (JDRF) Postdoc-
toral Fellowship (3-1999-538) to B.J.M. B.J.M. is currently supported by JDRF (2-2004-275) and NIH/
NIDDK (DK-71236) funding and is a senior research affiliate of the ARC Special Research Centre for
Functional and Applied Genomics. I would like to especially thank Adam Costin, Janette Galea, Garry
Morgan, and Peter van der Heide for generating and segmenting data that are presented here.
The Advanced Cryo-ElectronMicroscopy Laboratory housed at the Institute forMolecular Bioscience
is a major node of the federally funded Nanostructural Analysis Network Organisation’s Major National
Research Facility (NANO-MNRF), and is jointly supported by the Queensland State government’s
‘‘Smart State Strategy’’ initiative. We thank Dr. Jamie Riches of the NANO-MNRF for instrument
calibrations and upkeep, and nanoTechnology Systems (Greensborough, VIC, Australia) for critical
maintenance and upgrades of the 300 keV Tecnai F30 EM within the laboratory.
216 Brad J. Marsh
References
Alarcon, C., Lincoln, B., and Rhodes, C. J. (1993). The biosynthesis of the subtilisin-related proprotein
convertase PC3, but no that of the PC2 convertase, is regulated by glucose in parallel to proinsulin
biosynthesis in rat pancreatic islets. J. Biol. Chem. 268, 4276–4280.
Arvan, P., and Castle, D. (1998). Sorting and storage during secretory granule biogenesis: Looking
backward and looking forward. Biochem. J. 332, 593–610.
Balczon, R., Overstreet, K. A., Zinkowski, R. P., Haynes, A., and Appel, M. (1992). The identification,
purification, and characterization of a pancreatic beta-cell form of the microtubule adenosine tripho-
sphatase kinesin. Endocrinology 131, 331–336.
Biel, S. S., Kawaschinski, K., Wittern, K. P., Hintze, U., and Wepf, R. (2003). From tissue to cellular
ultrastructure: Closing the gap between micro- and nanostructural imaging. J. Microsc. 212, 91–99.
Bonner-Weir, S. (1988). Morphological evidence for pancreatic polarity of beta-cell within islets of
Langerhans. Diabetes 37, 616–621.
Bonner-Weir, S. (1989). Pancreatic islets: Morphology, organization, and physiological implications.
In ‘‘Molecular and Cellular Biology of Diabetes Mellitus. Vol. I. Insulin Secretion’’ (B. Draznin, S.
Melmed, and D. LeRoith, eds.), pp. 1–11. Alan R. Liss, Inc., New York.
Bork, P., and Serrano, L. (2005). Towards cellular systems in 4D. Cell 121, 507–509.
Boyd, A. E., III, Bolton, W. E., and Brinkley, B. R. (1982). Microtubules and beta cell function: EVect
of colchicine on microtubules and insulin secretion in vitro by mouse beta cells. J. Cell Biol. 92,
425–434.
Breant, B., Lavergne, C., Astesano, A., Ferrand, N., Asfari, M., Boissard, C., Anteunis, A., and
Rosselin, G. (1992). Development of the beta cells. Mt. Sinai J. Med. 59, 175–185.
Breuza, L., Halbeisen, R., Jeno, P., Otte, S., Barlowe, C., Hong, W., and Hauri, H. P. (2004).
Proteomics of endoplasmic reticulum-Golgi intermediate compartment (ERGIC) membranes from
brefeldin A-treated HepG2 cells identifies ERGIC-32, a new cycling protein that interacts with
human Erv46. J. Biol. Chem. 279, 47242–47253.
Cooper, M. S., Cornell-Bell, A. H., Chernjavsky, A., Dani, J. W., and Smith, S. J. (1990). Tubulove-
sicular processes emerge from trans-Golgi cisternae, extend along microtubules, and interlink adja-
cent trans-golgi elements into a reticulum. Cell 61, 135–145.
Coppola, T., Frantz, C., Perret-Menoud, V., Gattesco, S., Hirling, H., and Regazzi, R. (2002). Pancre-
atic beta-cell protein granuphilin binds Rab3 and Munc-18 and controls exocytosis. Mol. Biol. Cell
13, 1906–1915.
Cortizo, A., Espinal, J., and Hammonds, P. (1990). Vectorial insulin secretion by pancreatic beta-cells.
FEBS Lett. 272, 137–140.
Donelan, M. J., Morfini, G., Julyan, R., Sommers, S., Hays, L., Kajio, H., Briaud, I., Easom, R. A.,
Molkentin, J. D., Brady, S. T., and Rhodes, C. J. (2002). Ca2þ-dependent dephosphorylation of
kinesin heavy chain on beta-granules in pancreatic beta-cells. Implications for regulated beta-granule
transport and insulin exocytosis. J. Biol. Chem. 277, 24232–24242.
Dubochet, J. (1995). High-pressure freezing for cryoelectron microscopy. Trends Cell Biol. 5, 366–368.
Elsner, M., Hashimoto, H., and Nilsson, T. (2003). Cisternal maturation and vesicle transport: Join the
band wagon! (Review). Mol. Membr. Biol. 20, 221–229.
Farquhar,M. G., and Palade, G. E. (1998). TheGolgi apparatus: 100 years of progress and controversy.
Trends Cell Biol. 8, 2–10.
Gilkey, J. C., and Staehelin, L. A. (1986). Advances in ultrarapid freezing for the preservation of cellular
ultrastructure. J. Electron Microsc. Tech. 3, 177–210.
Gleason, C. E., Gonzalez, M., Harmon, J. S., and Robertson, R. P. (2000). Determinants of glucose
toxicity and its reversibility in the pancreatic islet beta-cell line, HIT-T15. Am. J. Physiol. Endocrinol.
Metab. 279, E997–E1002.
Golgi, C. (1898). Sur la structure des cellules nerveuses des ganglions spinaux. Arch. Ital. Biol. 30,
278–286.
8. Reconstructing Mammalian Cytomembranes by 3D EM 217
Goodge, K. A., and Hutton, J. C. (2000). Translational regulation of proinsulin biosynthesis and
proinsulin conversion in the pancreatic beta-cell. Semin. Cell Dev. Biol. 11, 235–242.
GriYths, G. (2004). Ultrastructure in cell biology: Do we still need it? Eur. J. Cell Biol. 83, 245–251.
GriYths, G., McDowall, A., Back, R., and Dubochet, J. (1984). On the preparation of cryosections for
immunocytochemistry. J. Ultrastruct. Res. 89, 65–78.
GriYths, G., and Simons, K. (1986). The trans Golgi network: Sorting at the exit site of the Golgi
complex. Science 234, 438–443.
Gu, F., Crump, C. M., and Thomas, G. (2001). Trans-Golgi network sorting. Cell. Mol. Life Sci. 58,
1067–1084.
Guest, P. C., Bailyes, E. M., and Hutton, J. C. (1997). Endoplasmic reticulumCa2þ is important for the
proteolytic processing and intracellular transport of proinsulin in the pancreatic beta-cell. Biochem. J.
323, 445–450.
Hama, K., Arii, T., and Kosaka, T. (1994). Three-dimensional organization of neuronal and glial
processes: High voltage microscopy. Microsc. Res. Tech. 29, 357–367.
Ho, H. C., Tang, C. Y., and Suarez, S. S. (1999). Three-dimensional structure of the Golgi apparatus in
mouse spermatids: A scanning electron microscopic study. Anat. Rec. 256, 189–194.
Howell, S. L., and Bird, G. S. (1989). Biosynthesis and secretion of insulin. Br. Med. Bull. 45, 19–36.
Howell, S. L., and Tyhurst, M. (1984). Insulin secretion: The eVector system. Experientia 40, 1098–1105.
Kajio, H., Olszewski, S., Rosner, P. J., Donelan, M. J., Geoghegan, K. F., and Rhodes, C. J. (2001).
A low-aYnity Ca2þ-dependent association of calmodulin with the Rab3A eVector domain inversely
correlates with insulin exocytosis. Diabetes 50, 2029–2039.
Katsumoto, T., Inoue, M., Naguro, T., and Kurimura, T. (1991). Association of cytoskeletons with the
Golgi apparatus: Three-dimensional observation and computer-graphic reconstruction. J. Electron
Microsc. (Tokyo) 40, 24–28.
Kowluru, A., and Morgan, N. G. (2002). GTP-binding proteins in cell survival and demise: The
emerging picture in the pancreatic beta-cell. Biochem. Pharmacol. 63, 1027–1035.
Kremer, J. R., Mastronarde, D. N., and McIntosh, J. R. (1996). Computer visualization of three-
dimensional image data using IMOD. J. Struct. Biol. 116, 71–76.
Kremer, J. R., O’Toole, E. T., Wray, G. P., Mastronarde, D. N., Mitchell, S. J., and McIntosh, J. R.
(1990). Characterization of beam-induced thinning and shrinkage of semi-thick sections in the
HVEM. In ‘‘Proc. XII Int. Congr. Electr. Microsc.,’’ 3, 752–753.
Kuliawat, R., and Arvan, P. (1994). Distinct molecular mechanisms for protein sorting within immature
secretory granules of pancreatic beta-cells. J. Cell Biol. 126, 77–86.
Ladinsky, M. S., Kremer, J. R., Furcinitti, P. S., McIntosh, J. R., and Howell, K. E. (1994). HVEM
tomography of the trans-Golgi network: Structural insights and identification of a lace-like vesicle
coat. J. Cell Biol. 127, 29–38.
Ladinsky,M. S., Mastronarde, D. N., McIntosh, J. R., Howell, K. E., and Staehelin, L. A. (1999). Golgi
structure in three dimensions: Functional insights from the normal rat kidney cell. J. Cell Biol. 144,
1135–1149.
Lehner, B., Tischler, J., and Fraser, A. G. (2005). Systems biology:Where it’s at in 2005.Genome Biol. 6,
338.
Lindsey, J. D., and Ellisman, M. H. (1985). The neuronal endomembrane system. I. Direct links
between rough endoplasmic reticulum and the cis element of the Golgi apparatus. J. Neurosci. 5,
3111–3123.
Lippincott-Schwartz, J., Roberts, T. H., and Hirschberg, K. (2000). Secretory protein traYcking and
organelle dynamics in living cells. Annu. Rev. Cell Dev. Biol. 16, 557–589.
Lippincott-Schwartz, J., Snapp, E., and Kenworthy, A. (2001). Studying protein dynamics in living
cells. Nat. Rev. Mol. Cell. Biol. 2, 444–456.
Lombardi, T., Montesano, R., and Orci, L. (1986). Loss of polarization of plasma membrane domains
in transformed pancreatic endocrine cell lines. Endocrinology 119, 502–507.
Luther, P. K., Lawrence, M. C., and Crowther, R. A. (1988). A method for monitoring the collapse of
plastic sections as a function of electron dose. Ultramicroscopy 24, 7–18.
218 Brad J. Marsh
Malaisse, W. J., Malaisse-Lagae, F., Van Obberghen, E., Somers, G., Devis, G., Ravazzola, M., and
Orci, L. (1975). Role of microtubules in the phasic pattern of insulin release. Ann. NY Acad. Sci. 253,
630–652.
Marsh, B. J. (2005). Lessons from tomographic studies of the mammalian Golgi. Biochim. Biophys. Acta
1744, 273–292.
Marsh, B. J., and Howell, K. E. (2002). The mammalian Golgi-complex debates. Nat. Rev. Mol. Cell.
Biol. 3, 789–795.
Marsh, B. J., Mastronarde, D. N., Buttle, K. F., Howell, K. E., andMcIntosh, J. R. (2001a). Organellar
relationships in the Golgi region of the pancreatic beta cell line, HIT-T15, visualized by high
resolution electron tomography. Proc. Natl. Acad. Sci. USA 98, 2399–2406.
Marsh, B. J., Mastronarde, D. N., McIntosh, J. R., and Howell, K. E. (2001b). Structural evidence for
multiple transport mechanisms through the Golgi in the pancreatic beta-cell line, HIT-T15. Biochem.
Soc. Trans. 29, 461–467.
Marsh, B. J., Volkmann, N., McIntosh, J. R., and Howell, K. E. (2004). Direct continuities between
cisternae at diVerent levels of the Golgi complex in glucose-stimulated mouse islet beta cells. Proc.
Natl. Acad. Sci. USA 101, 5565–5570.
Mastronarde, D. N. (1997). Dual-axis tomography: An approach with alignment methods that preserve
resolution. J. Struct. Biol. 120, 343–352.
Mastronarde, D. N. (2005). Automated electron microscope tomography using robust prediction of
specimen movements. J. Struct. Biol. 152, 36–51.
McDonald, K., and Morphew, M. K. (1993). Improved preservation of ultrastructure in diYcult-to-fix
organisms by high pressure freezing and freeze substitution: I. Drosophila melanogaster and Stron-
gylocentrotus purpuratus embryos. Microsc. Res. Tech. 24, 465–473.
McDonald, K. L., O’Toole, E. T., Mastronarde, D. N., and McIntosh, J. R. (1992). Kinetochore
microtubules in PTK cells. J. Cell Biol. 118, 369–383.
McIntosh, J. R. (2001). Electron microscopy of cells. A new beginning for a new century. J. Cell Biol.
153, F25–F32.
Mellman, I., and Simons, K. (1992). The Golgi complex: In vitro veritas? Cell 68, 829–840.
Mironov, A. A., Weidman, P., and Luini, A. (1997). Variations on the intracellular transport theme:
Maturing cisternae and traYcking tubules. J. Cell Biol. 138, 481–484.
Molinete, M., Irminger, J. C., Tooze, S. A., and Halban, P. A. (2000). TraYcking/sorting and granule
biogenesis in the beta-cell. Semin. Cell Dev. Biol. 11, 243–251.
Nickell, S., Kofler, C., Leis, A. P., and Baumeister, W. (2006). A visual approach to proteomics. Nat.
Rev. Mol. Cell. Biol. 7, 225–230.
NovikoV, A. B. (1964). GERL, its form and function in neurons of rat spinal ganglia. Biol. Bull. 127,
358.
NovikoV, A. B., Yam, A., and NovikoV, P. M. (1975). Cytochemical study of secretory process in
transplantable insulinoma of Syrian golden hamster. Proc. Natl. Acad. Sci. USA 72, 4501–4505.
O’Toole, E. T., Winey, M., and McIntosh, J. R. (1999). High-voltage electron tomography of spindle
pole bodies and early mitotic spindles in the yeast Saccharomyces cerevisiae. Mol. Biol. Cell 10,
2017–2031.
Ohgawara, H., Miyazaki, J., Karibe, S., Tashiro, F., Akaike, T., and Hashimoto, Y. (1995). Embedded-
culture of pancreatic beta-cells derived from transgenic mouse insulinoma as a potential source for
xenotransplantation using a diVusion chamber. Cell Transplant. 4, 307–313.
Orci, L. (1986). The insulin cell: Its cellular environment and how it processes (pro)insulin. Diabetes
Metab. Rev. 2, 71–106.
Orci, L., Malhotra, V., Amherdt, M., Serafini, T., and Rothman, J. E. (1989a). Dissection of a single
round of vesicular transport: Sequential intermediates for intercisternal movement in the Golgi stack.
Cell 56, 357–368.
Orci, L., Ravazzola, M., and Perrelet, A. (1984). (Pro)insulin associates with Golgi membranes of
pancreatic B cells. Proc. Natl. Acad. Sci. USA 81, 6743–6746.
8. Reconstructing Mammalian Cytomembranes by 3D EM 219
Orci, L., Thorens, B., Ravazzola, M., and Lodish, H. F. (1989b). Localization of the pancreatic beta cell
glucose transporter to specific plasma membrane domains. Science 245, 295–297.
Orci, L., Vassalli, J. D., and Perrelet, A. (1988). The insulin factory. Sci. Am. 259, 85–94.
Otegui, M. S., Mastronarde, D. N., Kang, B. H., Bednarek, S. Y., and Staehelin, L. A. (2001). Three-
dimensional analysis of syncytial-type cell plates during endosperm cellularization visualized by high
resolution electron tomography. Plant Cell 13, 2033–2051.
Palade, G. E. (1988). Cell fractionation: Importance to cell-free systems development. Prog. Clin. Biol.
Res. 270, xix–xx.
Park, J. B., Kim, J. S., Lee, J. Y., Kim, J., Seo, J. Y., and Kim, A. R. (2002). GTP binds to Rab3A in a
complex with Ca2þ/calmodulin. Biochem. J. 362, 651–657.
Penczek, P., Marko, M., Buttle, K., and Frank, J. (1995). Double-tilt electron tomography.Ultramicro-
scopy 60, 393–410.
Poitout, V., Olson, L. K., and Robertson, R. P. (1996). Insulin-secreting cell lines: Classification,
characteristics and potential applications. Diabetes Metab. 22, 7–14.
Poitout, V., Stout, L. E., Armstrong, M. B., Walseth, T. F., Sorenson, R. L., and Robertson, R. P.
(1995). Morphological and functional characterization of beta TC-6 cells—an insulin-secreting cell
line derived from transgenic mice. Diabetes 44, 306–313.
Porter, K. R., Claude, A., and Fullam, E. F. (1945). A study of tissue culture cells by electron
microscopy: Methods and preliminary observations. J. Exp. Med. 81, 233–246.
Rambourg, A., and Clermont, Y. (1997). Three-dimensional structure of the Golgi apparatus in mam-
malian cells. In ‘‘The Golgi apparatus’’ (E. G. Berger and J. Roth, eds.), pp. 37–61. Birkhauser-
Verlag, Basel.
Rambourg, A., Marraud, A., and Thiery, G. (1974). Scanning electron microscopy by transmission: Its
value for tridimensional study of cell organelles. CR Acad. Sci. Hebd. Seances. Acad. Sci. D 279,
283–284.
Rios, R. M., and Bornens, M. (2003). The Golgi apparatus at the cell centre. Curr. Opin. Cell Biol. 15,
60–66.
Rothman, J. E., and Orci, L. (1992). Molecular dissection of the secretory pathway. Nature 355,
409–415.
Rutter, G. A. (1999). Insulin secretion: Feed-forward control of insulin biosynthesis? Curr. Biol. 9,
R443–R445.
Sandler, S., and Andersson, A. (1984). The significance of culture for successful cryopreservation of
isolated pancreatic islets of Langerhans. Cryobiology 21, 503–510.
Santerre, R. F., Cook, R. A., Crisel, R. M., Sharp, J. D., Schmidt, R. J., Williams, D. C., and Wilson,
C. P. (1981). Insulin synthesis in a clonal cell line of simian virus 40-transformed hamster pancreatic
beta cells. Proc. Natl. Acad. Sci. USA 78, 4339–4343.
Segui-Simarro, J. M., Austin, J. R., II, White, E. A., and Staehelin, L. A. (2004). Electron tomographic
analysis of somatic cell plate formation in meristematic cells of Arabidopsis preserved by high-
pressure freezing. Plant Cell 16, 836–856.
Slot, J. W., and Geuze, H. J. (1983). Immunoelectron microscopic exploration of the Golgi complex.
J. Histochem. Cytochem. 31, 1049–1056.
Storrie, B., and Nilsson, T. (2002). The Golgi apparatus: Balancing new with old. TraYc 3, 521–529.
Tanaka, K., Mitsushima, A., Fukudome, H., and Kashima, Y. (1986). Three-dimensional architecture
of the Golgi complex observed by high resolution scanning electron microscopy. J. Submicrosc. Cytol.
18, 1–9.
Taylor, K. A., Reedy, M. C., Cordova, L., and Reedy, M. K. (1984). Three-dimensional reconstruction
of rigor insect flight muscle from tilted thin sections. Nature 310, 285–291.
Tsuboi, T., and Rutter, G. A. (2003). Insulin secretion by ‘kiss-and-run’ exocytosis in clonal pancreatic
islet beta-cells. Biochem. Soc. Trans. 31, 833–836.
Turner, M. D., and Arvan, P. (2000). Protein traYc from the secretory pathway to the endosomal
system in pancreatic beta-cells. J. Biol. Chem. 275, 14025–14030.
220 Brad J. Marsh
Varadi, A., Ainscow, E. K., Allan, V. J., and Rutter, G. A. (2002). Involvement of conventional kinesin
in glucose-stimulated secretory granule movements and exocytosis in clonal pancreatic beta-cells.
J. Cell Sci. 115, 4177–4189.
Wang, R. N., and Rosenberg, L. (1999). Maintenance of beta-cell function and survival following islet
isolation requires re-establishment of the islet-matrix relationship. J. Endocrinol. 163, 181–190.
Weidman, P. J. (1995). Anterograde transport through the Golgi complex: Do Golgi tubules hold the
key? Trends Cell Biol. 5, 302–307.
Wicksteed, B., Alarcon, C., Briaud, I., Lingohr, M. K., and Rhodes, C. J. (2003). Glucose-induced
translational control of proinsulin biosynthesis is proportional to preproinsulin mRNA levels in islet
beta-cells but not regulated via a positive feedback of secreted insulin. J. Biol. Chem. 278,
42080–42090.
Wilson, C. J., Mastronarde, D. N.,McEwen, B., and Frank, J. (1992).Measurement of neuronal surface
area using high-voltage electron microscope tomography. Neuroimage 1, 11–22.
Wu, C. C., MacCoss, M. J., Mardones, G., Finnigan, C., Mogelsvang, S., Yates, J. R., III, and Howell,
K. E. (2004). Organellar proteomics reveals Golgi arginine dimethylation. Mol. Biol. Cell 15,
2907–2919.
Yaekura, K., Julyan, R., Wicksteed, B. L., Hays, L. B., Alarcon, C., Sommers, S., Poitout, V., Baskin,
D. G., Wang, Y., Philipson, L. H., and Rhodes, C. J. (2003). Insulin secretory deficiency and glucose
intolerance in Rab3A null mice. J. Biol. Chem. 278, 9715–9721.
Yates, J. R., III, Gilchrist, A., Howell, K. E., and Bergeron, J. J. (2005). Proteomics of organelles and
large cellular structures. Nat. Rev. Mol. Cell. Biol. 6, 702–714.
Yi, Z., Yokota, H., Torii, S., Aoki, T., Hosaka, M., Zhao, S., Takata, K., Takeuchi, T., and Izumi, T.
(2002). The Rab27a/granuphilin complex regulates the exocytosis of insulin-containing dense-core
granules. Mol. Cell. Biol. 22, 1858–1867.
Yorde, D. E., and KalkhoV, R. K. (1987). Morphometric studies of secretory granule distribution and
association with microtubules in beta-cells of rat islets during glucose stimulation. Diabetes 36,
905–913.