expedited approaches to whole cell electron tomography and ... · expedited approaches to whole...

Available online at www.sciencedirect.comJournal of

ARTICLE IN PRESS

www.elsevier.com/locate/yjsbi

Journal of Structural Biology xxx (2007) xxx–xxx

StructuralBiology

Expedited approaches to whole cell electron tomography andorganelle mark-up in situ in high-pressure frozen pancreatic islets

Andrew B. Noske a,b, Adam J. Costin a,1, Garry P. Morgan a,1, Brad J. Marsh a,b,c,*

a Institute for Molecular Bioscience, Queensland Bioscience Precinct, The University of Queensland, Brisbane, Qld 4072, Australiab ARC Centre of Excellence in Bioinformatics, Queensland Bioscience Precinct, The University of Queensland, Brisbane, Qld 4072, Australia

c Centre for Microscopy and Microanalysis and School of Molecular and Microbial Sciences, The University of Queensland, Brisbane, Qld 4072, Australia

Received 13 May 2007; received in revised form 28 August 2007; accepted 11 September 2007

Abstract

We have developed a simplified, efficient approach for the 3D reconstruction and analysis of mammalian cells in toto by electronmicroscope tomography (ET), to provide quantitative information regarding ‘global’ cellular organization at �15–20 nm resolution.Two insulin-secreting beta cells—deemed ‘functionally equivalent’ by virtue of their location at the periphery of the same pancreaticislet—were reconstructed in their entirety in 3D after fast-freezing/freeze-substitution/plastic embedment in situ within a glucose-stim-ulated islet of Langerhans isolated intact from mouse pancreata. These cellular reconstructions have afforded several unique insights intofundamental structure–function relationships among key organelles involved in the biosynthesis and release of the crucial metabolic hor-mone, insulin, that could not be provided by other methods. The Golgi ribbon, mitochondria and insulin secretory granules in each cellwere segmented for comparative analysis. We propose that relative differences between the two cells in terms of the number, dimensionsand spatial distribution (and for mitochondria, also the extent of branching) of these organelles per cubic micron of cellular volumereflects differences in the two cells’ individual capacity (and/or readiness) to respond to secretagogue stimulation, reflected by an apparentinverse relationship between the number/size of insulin secretory granules versus the number/size of mitochondria and the Golgi ribbon.We discuss the advantages of this approach for quantitative cellular ET of mammalian cells, briefly discuss its application relevant toother complementary techniques, and summarize future strategies for overcoming some of its current limitations.� 2007 Elsevier Inc. All rights reserved.

Keywords: Cellular tomography; 3D reconstruction; Mammalian; Insulin secretion; Beta cell; Islets of Langerhans; Montaging; Segmentation; Autom-ation; Transforms; Insulin; Electron microscope tomography; Golgi; Pancreas; Secretory granules; Trans-Golgi; Endocrine; Exocytosis; Cytoskeleton

1047-8477/$ - see front matter � 2007 Elsevier Inc. All rights reserved.

doi:10.1016/j.jsb.2007.09.015

Abbreviations: 3D, three-dimensions/-dimensional; 2D, two-dimen-sions/-dimensional; EM, electron microscope/microscopy/microscopic;ET, electron microscope tomography; TGN, trans-Golgi network; ER,endoplasmic reticulum; keV, kilo electron volts; NCS, newborn calf ser-um; FBS, fetal bovine serum; LM, light microscope/microscopy/micro-scopic; SA, surface area; V, volume.

* Corresponding author. Address. Institute for Molecular Bioscience,Queensland Bioscience Precinct, The University of Queensland, Brisbane,Qld 4072, Australia. Fax: +61 7 3346 2101.

E-mail address: [email protected] (B.J. Marsh).1 Both authors contributed equally to this work.

Please cite this article in press as: Noske, A.B. et al., Expedited appr(2007), doi:10.1016/j.jsb.2007.09.015

1. Introduction

Studying cell biology/physiology at the electron micro-scopic (EM) level, particularly in the context of intactnative tissue with biomedical relevance to human/publichealth, has emerged as a major goal of the complex systemsand computational biology research communities over thepast few years (Arita et al., 2005; Bork and Serrano, 2005;Burrage et al., 2006; Coggan et al., 2005). Moreover, build-ing on a precise spatio-temporal scaffold in all four dimen-sions (4D) is increasingly regarded as a basic tenet todeveloping a sufficient understanding of the cell as a uni-tary complex system, and as a prerequisite for in silicoefforts at cell simulation (Bork and Serrano, 2005; Lehner

oaches to whole cell electron tomography and ..., J. Struct. Biol.

mailto:[email protected]

2 A.B. Noske et al. / Journal of Structural Biology xxx (2007) xxx–xxx

ARTICLE IN PRESS

et al., 2005; Nickell et al., 2006). Indeed, much of the workcarried out over the past 40 years or so to elucidate theinsulin biosynthetic pathway within the beta cells of theendocrine pancreas has relied heavily on a rigorous struc-ture–function based approach using conventional 2D EMtechniques as well as rudimentary 3D studies of serial par-affin sections (Bonner-Weir, 1988, 1989; Bonner-Weir andOrci, 1982; Greider et al., 1969; Howell and Tyhurst,1974; Howell et al., 1969; Orci, 1974; Orci, 1976a,b, 1985;Orci et al., 1984).

Together, these studies have contributed significantly toour overall understanding of the key steps involved in insu-lin manufacture and release. However, real gaps stillremain in our basic knowledge of the insulin biosyntheticpathway from molecule-to-cell, and from cell-to-tissue.This led us several years ago to develop techniques forthe improved preservation and subsequent 3D imaging ofinsulin-secreting cells by EM tomography (ET), first usingimmortalized beta cell lines (Marsh et al., 2001b), then withbeta cells still resident in situ in pancreatic islets isolatedfrom mice (Marsh et al., 2004). Although the tomogramsgenerated from these studies have provided (and continueto provide) exciting new insights into structure–functionrelationships among organelles of the insulin biosyntheticpathway at comparatively high (�5–7 nm) resolution(Marsh et al., 2001a, 2004, 2001b; Mogelsvang et al.,2004), they are limited in that they normally only allowthe detailed examination of a relatively small percentage(i.e., 61%) of the total cell volume due to the significanttechnical challenges associated with high-resolution ET oflarge cellular volumes (Marsh, 2005; Marsh, 2007).

To both qualitatively and quantitatively assess relation-ships among the key organelles involved in insulin produc-tion in the context of the whole cell, we recently undertookthe development and application of an expedited ETapproach that would allow us to image, reconstruct andanalyze entire pancreatic beta cells in detail (620 nm) in3D, at an order of magnitude better resolution than stan-dard light microscopy (LM) and at least twice the resolutionof the most promising approaches in 3D and 4D LM (Egneret al., 2004; Hara et al., 2006; Ma et al., 2004; Michael et al.,2004). While limited by the fact that when compared to livecell imaging we are restricted to viewing a single static ‘‘3Dsnapshot’’, we retain the advantage of having sufficient res-olution to visually/morphologically distinguish most com-partments/organelles of interest at once (based on awealth of elegant ultrastructural studies of mammalianmembrane traffic over the past six decades), as opposed toonly being able to visualize those compartments that trafficand/or house a protein(s) of interest tagged with a fluoro-phore(s). Our primary criteria for establishing an appropri-ate schema was that it must provide a means forcomparative whole cell studies of islet/beta cell biology byET in a way that was fiscally/temporally practical, andwould lend itself to attempts at automation. Thus, weimaged and reconstructed two beta cells from the same glu-cose-stimulated mouse islet by single axis, serial section ET


at magnifications of 4700· and 3900·, respectively, thatresulted in whole cell tomograms with a final resolution of�15–20 nm. In addition, we developed several new methodsfor the abbreviated segmentation of cellular organelles/compartments that allowed us to segment the entire Golgiribbon as well as both cells’ full complement of mitochon-dria and insulin secretory granules for comparative analy-sis, on a timescale of weeks rather than months. Wepropose that although relatively crude, this approachaffords important new insights into global cellular organiza-tion and comparative cell biology in 3D that cannot beattained by other methods.

2. Materials and methods

2.1. Islet isolation and culture

Islets of Langerhans were isolated intact from mousepancreas by intra-ductal collagenase injection/digestionand purified as previously described (Gotoh et al., 1985;Marsh et al., 2004; Nicolls et al., 2002). After manuallypicking individual islets to minimize exocrine tissue con-tamination, islets were cultured in RPMI medium contain-ing 10% (v/v) newborn calf serum (NCS) or fetal bovineserum (FBS) (Invitrogen, Australia) and 7 mM D-glucoseequilibrated with 5% CO2 at 37 �C. The medium was addi-tionally supplemented with L-glutamine and 2-mercap-toethanol (Sigma Chemical Co., St. Louis, MO, USA).Following culture for 2–3 h for tissue recovery and to pro-mote re-initiation of protein synthesis, islets were trans-ferred to RPMI medium containing low glucose (3 mM)and cultured overnight. Islets were then transferred to med-ium containing a physiologically relevant stimulatory con-centration (11 mM) of D-glucose for one hour before highpressure freezing to maximally upregulate proinsulinbiosynthesis.

2.2. High-pressure freezing and freeze-substitution fixation

Islet cultures were maintained at 37 �C in Hepes-buf-fered (10 mM) RPMI medium (Sigma Chemical Co., St.Louis, MO, USA) containing either NCS or FBS 10%(v/v) (Invitrogen, Australia) prior to freezing. Immediatelyprior to freezing, 10–30 islets (depending on size) weremanually transferred by pipette into interlocking brassplanchettes (Swiss Precision Inc., CA, USA), also pre-warmed to 37 �C, under a dissecting microscope (Olympus,Australia). Islets were high-pressure frozen using a BalzersHPM010 high-pressure freezer (BAL-TEC AG, Liechten-stein) and stored under liquid nitrogen. Specimens werefreeze-substituted and plastic-embedded essentially asdescribed previously (Marsh et al., 2001b).

2.3. Microtomy and preparation for EM

Ribbons of thin (40–60 nm) or thick (300–400 nm) sec-tions were cut on a microtome (Leica-Microsystems) for


Fig. 1. Intact islets of Langerhans isolated from adult mice were culturedovernight in vitro, stimulated with elevated extracellular glucose (11 mM)for 60 min, and preserved for EM by high-pressure freezing, freeze-substitution and plastic embedding. (Inset) 2D survey showing the wholeislet in cross-section, acquired by stage-shifted montaging of 12 · 12images at 4700· magnification using the data acquisition programSerialEM (Mastronarde, 2003, 2005). Boxed regions highlight the twoislet beta cells (designated ‘ribbon01’ and ‘ribbon02’) in the islet imagedand reconstructed by ET. (Main image) 0� (untilted) view taken from oneof the tilt series of islet beta cell ribbon02. Scale bar: 1 lm.

A.B. Noske et al. / Journal of Structural Biology xxx (2007) xxx–xxx 3

ARTICLE IN PRESS

conventional 2D survey at 80–100 keV to assess the qualityof islet preservation and for 3D studies by ET at 300 keVon Tecnai T12 and F30 microscopes, respectively (FEICompany). Typically, ribbons of serial thick sections col-lected onto formvar-coated copper (2 · 1 mm) slot grids(Electron Microscopy Sciences) and post-stained with 2%aqueous uranyl acetate (UA) and Reynold’s lead citraterequire an additional carbon-coating step to minimizecharging/movement in the electron beam, particularlywhen tilted for tomography (Marsh, 2005, 2007). Slot gridsare normally used because they have—to date—affordedthe maximal possible unobstructed viewing area when thespecimen is tilted beyond 60� in the EM. However, sinceribbons of serial sections are essentially laid down thelength of the grid parallel to the slot itself, tomographic ‘tiltseries’ collected around two orthogonal axes are often lim-ited to tilt angles lower than 60� because the metal of thespecimen support grid obstructs the electron beam at hightilt for at least one of the two tilt series. The limiting factorto using grids that provide larger viewing areas for high tilttomography has been the strength and durability of theplastic support films that are commercially available.

In the present study, we also extensively trialed andemployed a new ‘‘gridless’’ aperture/support film combina-tion in which polyimide plastic support films (tested atthicknesses of 388 A or 605 A) span an ‘‘open aperture’’of uniform 2 mm diameter (Luxel Corp., Friday Harbor,WA, USA). The polyimide support film was also testedat thicknesses of 388 A, 400 A, 552 A, 605 A and 741 Ausing standard 1 · 2 mm slot grids. Most notably, the dra-matically reduced shift/drift/charging we have been affor-ded through use of these ‘‘gridless’’/open aperturesupports (typically <0.8 pixel with successive measure-ments) in the EM at all tilts—in the absence of carboncoating—when compared to sections from the same speci-men blocks prepared in parallel and imaged on regularformvar-coated slot grids carbon-coated both sides, pro-vided us with (1) a significant improvement in efficiency(an overall 2- to 4-fold improvement in speed of tilt seriesacquisition) for semi-automated image data collection forET and (2) the ability to reproducibly tilt to angles beyond60� essentially regardless of the positioning of the ribbon(s)of serial sections. These factors played an important role infacilitating the rapid acquisition of tilt series data for anentire cell. Colloidal gold particles of 10 nm diameter werethen deposited on both surfaces of these sections for use asfiducial markers during subsequent image alignment(Marsh et al., 2001b).

2.4. Whole cell tomography

Initially, over 40 serial thick sections through awell-preserved mouse islet were surveyed in 2D bystage-shifted montaging of 12 · 12 images at 4700·magnification using the data acquisition program Seria-lEM (Mastronarde, 2003, 2005). These montages allowedus to identify a number of candidate cells for further


study by whole cell ET and to determine how many sec-tions each cell spanned. In addition, it allowed us tocheck what magnification and orientation were necessaryto encompass each cell fully in cross-section within a sin-gle 2K · 2K CCD image (4K · 4K CCD, binned ·2), andto check that sections were not damaged or contaminatedin a way that would prevent potential ‘target’ cells frombeing imaged in toto.

3D reconstruction by ET of each of the two beta cells(designated ‘ribbon01’ and ‘ribbon02’) was accomplishedby single axis tomography of 46 and 27 consecutive serialthick (300–400 nm) sections imaged at magnifications of4700· or 3900·, respectively, using a Tecnai F30 intermedi-ate voltage EM (FEI Company) and motorized tilt–rotatespecimen holders (Models 650 and CT3500TR; GATANInc.) (Fig. 1). Tilt series data were digitally recorded usingsemi-automated methods for CCD (charge-coupled device)image montaging, data acquisition and image alignment asthe sections were serially tilted by 2� increments over arange of ±62� using the microscope control program Seria-

lEM. For ribbon01, a total of 46 adjacent serial thick sec-tions were imaged at 4700· magnification (pixelsize = 5.156 nm) to generate a final single volume spanning�10 · 10 · 20 lm3 (Figs. 2A and 3). Tilt series image datafor ribbon02 was collected in a similar manner from 27


Fig. 2. Tomographic slices extracted from the whole cell tomograms generated for datasets ribbon01 and ribbon02. The boxed areas reveal the relativemembrane clarity for each of the two datasets at higher magnification (inset). (A) Slice 939 of 2217 from the dataset ribbon01 (pixel size = 5.156 nm;imaged at 4700· magnification) shows segmentation of the Golgi ribbon (gray), TGN (red), mitochondria (green), mature insulin granules (dark blue),nucleus (yellow), and plasma membrane (purple). (B) Slice 556 of 1153 from ribbon02 (pixel size = 6.066 nm; imaged at 3900· magnification). The colorcoding scheme is as described above for (A). Scale bars: 1 lm. (For interpretation of color mentioned in this figure the reader is referred to the web versionof the article.)


ARTICLE IN PRESS

serial thick sections (Figs. 2B and 4). However, because thecell was wider in cross-section, the magnification used inthe tilt series was reduced down one step to 3900· (pixelsize = 6.066 nm). The timecourse for tilt series data acqui-sition and reconstruction was between 2 and 3 weeks intotal for each cell.

Tilt series images (2K · 2K) acquired to an Ultra-Scan� 4000 (USC4000, Model 895; GATAN Inc.) werefirst brought into register with one another by cross-cor-relation. Tracking a limited number (�40) of the 10 nmgold fiducial markers on both surface(s) of the sectionsacross the entire field of view became an absolute prere-quisite so that the program TILTALIGN could be usedto solve for global fiducial alignment and distortion cor-rection. Fiducial-less alignment procedures generallyresulted in grossly distorted and uninterpretable tomo-grams because of the significant and inherent geometricalchanges and distortions across such large cellular areas.Tomograms from each of the serial sections through thecell(s) calculated by R-weighted backprojection from eachset of aligned tilts were joined along the Z axis essentiallyas described previously (Ladinsky et al., 1999; Marshet al., 2001b; Shoop et al., 2002; Sosinsky et al., 2005)using the interactive program MIDAS to generate a gen-eral linear transform that accounted for rotation, transla-tion and stretch of one section relative to its nearestneighbor. 3D cellular reconstructions generated in thisway using the IMOD software package incorporatingthe eTomo and 3dmod graphical user interfaces were thenanalyzed (Kremer et al., 1996).


2.5. Segmentation and quantitative 3D analysis

Compartments/organelles within the tomographic vol-umes were segmented, extracted and viewed using IMODsoftware package (Kremer et al., 1996). To quantify mem-brane surface area and volume for the entire Golgi ribbonin each cell expeditiously (Fig. 6), only every third slice in Z

was modeled for cis–medial-cisternae (which were seg-mented/modeled as a single object) and likewise fortrans-Golgi cisternae, the latter frequently/collectivelyreferred to as the trans-Golgi network (TGN). In ribbon01,the penultimate trans-Golgi cisterna and trans-most Golgicisterna were modeled as separate objects to demarcatefunctional exit sites along the length of the Golgi ribbon.

The plasma cell membrane (purple) and nucleus (yellow)were delineated using only �4 contours per section, butstill provided a clear definition of their boundaries. Mito-chondria were segmented by drawing a series of connectedspheres centered along the length of each. In this way, wewere able to mark-up all of the mitochondria in less than15 h per cell. The numerous mature granules (blue) weresegmented by precisely fitting a sphere. However, forimmature granules (light blue), approximately half weresegmented as a sphere while those more irregular in shapewere segmented using contours (�1 in every 6 slices), aswere multi-vesicular compartments. All spheres—includingthose used to approximate mitochondria—were centered asaccurately as possible and individually resized.

Spatial relationships (such as the proximity distributionanalysis between mitochondria and the surface of the Golgi


Fig. 3. Comparison views along the X–Z axis of the surface-rendered 3D model data segmented following serial section ET reconstruction of ribbon01 andribbon02. (A,C) Tomographic volumes that encompassed each of the beta cells designated ribbon01 and ribbon02 were generated by joining 46 and 27 serialsection tomograms, respectively. (B,D) 3D models revealing key compartments involved in insulin production and release by beta cells were generated foreach of ribbon01 and ribbon02 using an expedited segmentation approach that allowed for the efficient yet accurate mark-up of organelles such as the Golgiribbon and TGN, mitochondria and insulin granules (both immature and mature). Here, however, the 3D model data are presented with the dark bluespheres demarking the mature insulin granule populations omitted so that the other compartments such as immature granules and multi-granular/vesicular bodies can be more readily seen (for detailed visualization of mature insulin granules, see Fig. 4). Shown is the: plasma membrane (purple),nucleus (yellow), main Golgi ribbon (gray), trans-most Golgi cisterna (red), penultimate trans-Golgi cisterna (gold), mitochondria (green), multi-vesicularbodies (orange) and immature granules (light blue). Scale bars: 1 lm.


ARTICLE IN PRESS

ribbon presented in Fig. 8) among the modeled objects in3D were assessed and quantified essentially as describedelsewhere using the program mtk (Marsh et al., 2001b).Because sections cut from plastic (epon) resin are knownto collapse by �40% in the EM on initial exposure to theelectron beam (Luther et al., 1988), the 3D surface-ren-dered model data were re-expanded by a factor of 1.7 inZ to more accurately represent the original topology ofsubcellular structures under study in the plastic section


prior to imaging in the EM (Marsh et al., 2001b). Mem-brane surface area and volumes were computed from thecontours and triangular meshes essentially as describedpreviously (Marsh et al., 2001a,b).

2.6. Supplementary material and multimedia file generation

Additional figures and movies provided as supportingmaterial for the data presented in Figs. 1–8 accompany


Fig. 4. Panel showing the relative number and distribution of both mature and immature insulin granules in both beta cell reconstructions. Mature insulingranules are demarked as dark blue spheres, while immature secretory granules are colored light blue (for unobstructed views of the number anddistribution of immature granule pools in each cell, see Fig. 3; due to the crowded appearance of the cytosol because of the numerous insulin granules, thisfigure panel is best viewed electronically). (A,C) show top and side views, respectively, for ribbon01, which contained 3370 mature insulin granules (averagediameter 280 nm) and �700 immature granules. Likewise, (B,D) reveal top and side views, respectively, for ribbon02, which contained 8250 mature insulingranules (average diameter 245 nm) and �520 immature secretory granules. Scale bars: 1 lm.


ARTICLE IN PRESS

the online version of this article. Movies of the 3D datawere compiled using Apple’s QuickTime� software fromsequences of individual TIF files generated using the 3dmod

viewer in IMOD, and files average around 10 MB each insize. For videos of tomograms and/or overlaid model/seg-mentation contours, images were generated as the tomo-gram was advanced along the Z axis. Movies of the 3Dsurface-rendered models were generated from segmented


tomograms as the data were incrementally rotated and/orzoomed to provide unambiguous views of the structurespresented.

2.7. Statistical analysis

Results are expressed as means ± SEM. Statistical anal-ysis was performed using the program ‘R’ (http://www.r-


http://www.r-project.org

Fig. 5. The 3D organization of the entire Golgi ribbon was analyzed in both glucose-stimulated islet beta cells reconstructed in toto. Membranes of theGolgi complex were segmented in ribbon01 (A) and ribbon02 (B) to distinguish between the main Golgi ribbon (i.e., the ‘ribbon proper’), comprised ofstacked cis- and medial-cisternae (gray), the trans-most Golgi cisterna (red) and the penultimate trans-cisterna (gold). The nucleus (yellow) is partiallyvisible in the figure background for reference; the orientation of the Golgi relative to the cell is the same for ribbon01 and ribbon02 as for Figs. 3 and 4,respectively. (B, inset) A portion of the segmented stacked cis–medial-Golgi cisternae (highlighted in green) shows how the main Golgi ribbon wassegmented.

Fig. 6. Variations in mitochondrial morphology are presented by way of example from ribbon01. As described in Section 2, mitochondria were segmentedusing an abbreviated approach aimed at expediting analysis whilst still providing important and accurate quantitative information indicative of changes inmitochondrial structure–function relationships in situ. Consequently, both intra- and inter-cellular variations were quantified with regard to the extent ofmitochondrial branching, length and diameter. For visual simplification, the main lengths of mitochondria are shown here in light blue, while branches arecolored mauve so that they can be readily distinguished; mitochondrial branch points are highlighted in red. (A–C) Representative examples are providedfor either single (A) or multiple (B) mitochondrial branches, and for regions where the mitochondrial diameter thins dramatically (C), suggestive of activemitochondrion fission or fusion. (D) A cartoon demonstrating how mitochondrial length and branching was assessed and measured.


ARTICLE IN PRESS

project.org) by unpaired Students t-test, where p < 0.05was considered significant.

2.8. Database deposition/accession information

Original EM tilt series image stacks, together with theresulting tomographic reconstructions and 3D surface-ren-dered model data associated with this work have been


deposited in the Cell Centered Database (CCDB; http://ncmir.ucsd.edu/CCDB/) (Martone et al., 2002). AssociatedQuickTime� movies and high-resolution images derivedfrom the tomograms and 3D rendered models publishedhere are publicly available for download directly from theCCDB or from the Journal of Structural Biology website,or can be provided from the authors on CD-ROM orDVD-ROM on request.


http://www.r-project.org

http://ncmir.ucsd.edu/CCDB/

http://ncmir.ucsd.edu/CCDB/

Fig. 7. Mitochondrial organization and distribution in ribbon01 (A, C, E) and ribbon02 (B, D, F). (A, B) The relative number and distribution of branched(main lengths: light blue; branches: mauve) versus non-branched (green) mitochondria are shown; branch points are highlighted with red spheres. Alsorefer to Table 2. Scale bars: 1 lm. (C–F) A proximity distribution analysis of mitochondria versus the Golgi ribbon was undertaken to assess relativedifferences in the extent of mitochondrial clustering to the Golgi between cells. (C,D) The subset of mitochondria that came within an arbitrary (400 nm)zone of close approach to any surface of the Golgi ribbon are highlighted in yellow, while the remainder are colored green. (E,F) Tomographic slices fromribbon01 (E) and ribbon02 (F) that show the arbitrary 400 nm proximity zone around the Golgi ribbon (shaded; dotted white line). The segmented Golgi(gray), proximal (yellow) and distal (green) mitochondria are visible. Scale bars: 400 nm.


ARTICLE IN PRESS

Please cite this article in press as: Noske, A.B. et al., Expedited approaches to whole cell electron tomography and ..., J. Struct. Biol.(2007), doi:10.1016/j.jsb.2007.09.015

Fig. 8. A comparative analysis of the variation in mitochondrial length between ribbon01 (A) and ribbon02 (B). Please note that the length of individualbranches has been graphed separately to the main lengths of mitochondria; also see Table 2 for a summary of additional quantitative data related tobranched versus non-branched mitochondria in ribbon01 versus ribbon02. Inset: The cartoon shows how the profile of a mitochondrion of average lengthcompares to the longest mitochondrion measured for each cell if they were straightened.


ARTICLE IN PRESS

3. Results

Although cellular ET has now almost reached the statusof a ‘mainstream’ research tool, much remains to belearned about how best to utilize this powerful techniqueto answer significant research questions in mammalian celland molecular biology. In our case, we have previouslyavoided imaging our insulin-secreting beta cells at lower


magnifications (and thus, lower resolutions) to accommo-date larger cellular areas for analysis, because the bulk ofour studies—like many in the 3D EM community—haverelied heavily on working toward developing the capacityto simultaneously visualize organelles and large macromo-lecular protein complexes together in situ in 3D within thesame high-resolution tomograms, to bridge the resolutiongap with molecular microscopy and X-ray crystallography



ARTICLE IN PRESS

(Marsh, 2006; Medalia et al., 2002; Nickell et al., 2006).However, single axis tomography of serial thick (1–2 lm)sections cut from mammalian cells and tissue has in thepast proved extremely useful for bridging the resolutiongap in the other direction: from cell-to-tissue (Soto et al.,1994; Wilson et al., 1992). Moreover, efforts to directly cor-relate EM data in 2D and/or 3D over a range of resolu-tions against image data from the light microscope (LM)(both live cell imaging and indirect immunofluorescencedata) promise to emerge as increasingly important as wemove toward developing a more ‘‘holistic’’ understandingof how changes at the molecular level manifest at the levelof cell and/or tissue (Biel et al., 2003; Gaietta et al., 2006;Hoog et al., 2007; Kurner et al., 2005; Medalia et al.,2007; Sosinsky et al., 2007).

3.1. 3D reconstruction of whole islet cells at �15–20 nm

resolution—an acceptable compromise between information

quality and efficiency?

Recently, in an effort to understand how basic changesin beta cell/islet physiology are reflected in terms of orga-nelle number, size and distribution at the level of the wholecell—and to work toward a schema that would lend itselfto a rapid, automated approach for the comparative anal-ysis of multiple whole tissue cells in the future—we carriedout serial sectioning/single axis ET of two entire beta cellsfrom the same pancreatic islet of Langerhans that allowedus to image, reconstruct and analyze key aspects of cellularorganization in 3D on the timescale of weeks (Fig. 1).Moreover, because we undertook these studies of differentcells within the same unit tissue but at different magnifica-tions, it afforded us the opportunity to directly compare thequality of the data generated, to determine a ‘minimum’magnification requirement for upcoming whole cell studiesby ET (Fig. 2).

Although we initially aimed to use fiducial-less align-ment for the rapid reconstruction of individual volumesfor the 46 and 27 serial sections that were imaged for eachof ribbon01 and ribbon02, respectively, it became apparentthat we absolutely needed to track a small number (�40) ofthe fiducials to solve for global alignment so that distortionin the final tomograms was reduced to a point where thedata were of sufficient quality for a robust and accurateanalysis (Supplemental Data S1). However, at a magnifica-tion of 3900· used to image ribbon02, the 10 nm gold fidu-cials were often difficult to visualize and track. Acombination of less accurate tracking of fiducial beads,together with a reduced magnification, resulted in finaltomograms of noticeably lower clarity for ribbon02 whencompared with ribbon01 (Fig. 2). Although the use of fidu-cial markers with a larger diameter (e.g., 15 nm) could cir-cumvent some aspects of the difficulties we experiencedwith bead tracking at a magnification of 3900·, it is worthnoting that at a magnification of 4700· we experiencedalmost no difficulties tracking the 10 nm colloidal gold par-ticles. Since one of the aims of this study was to develop a


schema that would promote efforts to work with the samecells/tissue across multiple scales/resolutions by ET, the useof fiducial gold beads of a single diameter (10 nm) amena-ble to tracking at both low and high magnification waspreferred.

Although the quality of reconstructions was substan-tially lower than can be obtained with high-resolution dualaxis tomograms that we routinely collect at 20,000· orhigher magnification, we estimated that the final resolutionwas still in target range of �15–20 nm. At this resolution,and despite limiting data collection to a single axis, we werestill able to clearly see and identify most of the organellesthat we typically study in our higher resolution tomograms(Marsh et al., 2004, 2001b), with the notable exception ofmicrotubules. By way of example, individual cisternae ofthe Golgi ribbon were easily distinguished in each slice(Fig. 2, inset), although segmenting them through Z wasless trivial. As a consequence, we have more recentlyattempted to find sets of imaging parameters that wouldallow us to image whole mammalian cells by ET quicklyyet maintaining sufficient resolution to adequately visualizeand track microtubules (data not shown). To this end, wehave collected data at a number of intermediate magnifica-tions (e.g. 9400· and 12,000·) in combination with limitedtiling/montaging (necessary to capture entire cells in cross-section at these magnifications) and additional image bin-ning (to expedite electronic data readout from the CCD).Unfortunately, however, we have not managed to identifyimaging parameters that would facilitate tomographicreconstruction of mammalian cells in toto in a suitablyrapid manner as well as provide sufficient detail to trackmicrotubules reliably. Any potential benefits of data collec-tion at the higher magnifications were lost as a consequenceof over-binning the tilt series images in order to acquire themontaged data in a temporally practical manner. Thisnotion is reinforced by an excellent recent study by Hoogand co-workers that specifically focused on elucidatingmicrotubule organization in whole (fission) yeast cells byET, and required them to collect tilt series data with a pixelsize of �1.5 nm (Hoog et al., 2007). This compares wellwith our own experience tracking microtubules reliably inhigh-resolution reconstructions with pixel sizes of�2.5 nm or smaller. In our own trials outlined above, wewere able to follow microtubules in single axis tomogramsfrom data collected at 9400· magnification (binned ·3;final pixel size of 3.94 nm) more easily than in dual axistomograms from tilt series collected at 12,000· magnifica-tion (binned ·4; final pixel size of 4.17 nm) (data notshown).

Although the effective resolution in Z was considerably(�1.7 times) worse than the resolution in X and Y, a moresignificant problem during segmentation resulted from mis-alignment of adjacent sections. Due to uneven specimencollapse in the EM together with heterogeneous sectionthinning and distortion during tilt series acquisition (Kre-mer et al., 1990; Luther et al., 1988; Marsh, 2005; vanMarle et al., 1995), the anisotropic distortions from one



ARTICLE IN PRESS

section to the next made it impossible to perfectly alignserial sections using global uniform transformations(Fig. 3; Supplemental Data S2). This problem is primarilyan inherent consequence of our abbreviated approach,since sectional misalignment has been negligible (less thana few nm) in our previous high-resolution ET studies wheredual axis tomograms have been joined from serial thicksections essentially without issue (Ladinsky et al., 1999;Marsh et al., 2001a,b; McEwen and Marko, 1999). Here,in our lower resolution reconstructions of cellular areas> 10 lm2 in X and Y, the problem of anisotropic distortionof sections was greatly exacerbated due to the size of theareas captured and as a consequence of employing singlerather than dual axis ET data. For example, in ribbon02many organelles are offset by up to 90 nm (�15 pixels)between adjacent section boundaries in the worst-case sce-narios. For these poorly aligned sections, additional carehad to be taken to track organelles reliably (Fig. 3; Supple-mental Data S2). However, in most instances, the align-ment of serial section tomograms was sufficientlyaccurate that even convoluted membrane-bound organ-elles, such as mitochondria and the Golgi ribbon, couldbe followed unambiguously for reliable segmentation overtheir entire length and in Z.

3.2. Abbreviated segmentation techniques for rapid insights

into 3D cell biology

Over the past few years, many discussions have takenplace in our group with regard to the potential merits ofusing a heavily abbreviated approach for the rapid segmen-tation of large cellular volumes. After all, even ‘stick andball’ models of atomic structure convey vital clues aboutthe structure–function characteristics of molecules. In thecurrent study, we decided to trial an expedited schemefor the segmentation of different organelles/compartmentsof interest that would allow us to extract crucial spatialand volumetric information about those structures as fastas possible, without sacrificing important informationregarding differences/changes in compartmentmorphology.

The development of tools for the expedited segmenta-tion and analysis of complex 3D cellular data has mostlyfailed to keep pace with the development of data acquisi-tion hardware and reconstruction software. As demon-strated here, an enormous volume of ET data can nowbe acquired and reconstructed relatively quickly (i.e. 2–3weeks), but methods for extracting useful spatial and quan-titative information from such data have remained rela-tively slow. Normally, to extract valuable informationregarding structure–function relationships for and amongorganelles, the tomogram is typically segmented manuallyto delineate cellular compartments; outlining objects ofinterest on every successive tomographic slice allows oneto generate a ‘contour map’ of subcellular topology whichis then converted into a high fidelity 3D meshed model.This process represents the major bottleneck in our ET


‘processing pipeline’ for cellular data as it typically takesweeks to months per tomogram. The 3.1 · 3.2 · 1.2 lm3

volume of part of the Golgi region in an insulin-secretingHIT-T15 cell reconstructed at �5–7 nm resolution in a pre-vious study (Marsh et al., 2001b) required approximately 9months of manual segmentation followed by �3 months ofadditional editing for detailed analysis. Although this rep-resents one of the largest tomographic volumes recon-structed from a mammalian cell at comparatively high(�5 nm) resolution, the mean volume of a murine beta cellvolume has been estimated at �1400 lm3 (Dean, 1973).Taken together, this reiterates the need for a ‘multi-resolu-tion’ approach to segmenting 3D cellular data that dependson the biological question at hand.

In the present study, we were able to readily segmentmost membrane-bound organelles of interest in both cells,with the notable exception of the endoplasmic reticulum(ER), small vesicles and microtubules (Figs. 3 and 4; Sup-plemental Data S3). Although both beta cells were selectedbased on equivalent locations (Fig. 1, inset) on oppositesides at the periphery of the same pancreatic islet, theirorganization was quite different. Ribbon01 was �610 lm3

in volume and had an estimated total surface area of683 lm2, with less than 10% (�5.3%) of the total plasmamembrane facing the ‘extra-islet space’ (Fig. 1, inset). Incontrast, ribbon02 was �698 lm3 in volume (1.14· largerthan ribbon01), with a total surface area of �490 lm2, ofwhich a comparatively large proportion (�23%) of theplasma membrane faced the islet’s exterior (Fig. 1, inset;Supplemental Data S2).

3.3. Comparative quantification of insulin granule number,

size and distribution

Each mature insulin granule was segmented by fittingand re-sizing a sphere that approximated its surface areaand volume in 3D as described previously (Marsh et al.,2001b) (Fig. 4; Supplemental Data S3). Surprisingly,although ribbon01 and ribbon02 had similar cell volumes(�610 lm3 versus �698 lm3), ribbon01 contained 2.4·fewer mature granules (3370 = 5.54 per lm3) than ribbon02

(8250 = 11.8 per lm3). In addition, the average diameter ofindividual granules was 1.14· larger in ribbon01 thanin ribbon02 (mean diameters of 280 nm versus 245 nm,respectively) (Table 1). Consequently, although ribbon01

contained fewer mature granules, the cytoplasmic volumeoccupied by those (mature) granules was disproportion-ately large in ribbon01 (8.5%) compared to ribbon02

(14%) (Table 1; Supplemental Data S5).Typically, islet beta cells each contain �9000 mature

granules (Dean, 1973; Olofsson et al., 2002; Rorsmanand Renstrom, 2003), with an estimated half-life of �3–5days (Halban, 1991; Halban and Wollheim, 1980). More-over, it is predicted that beta cells secrete only 1–2% oftheir insulin granule content per hour under stimulatoryconditions (Rorsman and Renstrom, 2003). Although wehad not predicted such differences in mature granule num-


Table 1A comparative summary of quantitative data derived for insulin granules,multi-vesicular bodies and the Golgi ribbon in datasets ribbon01 andribbon02

ribbon01 ribbon02

Mature granules: total number 3370 8250Mean diametera 280 ± 0.84 nm 245 ± 0.43 nmEstimated total volume 42.6 lm3 68.4 lm3

Immature granules: total number 700 520Estimated total volume 5.6 lm3 5.0 lm3

Multi-vesicular bodies: total number 28 12Estimated total volume 0.39 lm3 3.7 lm3

Golgi: estimated volume of main stack 3.6 lm3 3.1 lm3

Estimated volume of penultimate trans-cisterna

0.35 lm3 N/A

Estimated volume of trans-most cisterna 1.9 lm3 0.53 lm3

Estimated total volume 5.8 lm3 3.7 lm3

Estimated membrane SA of main Golgiribbon

310 lm2 230 lm2

Estimated membrane SA of penultimatetrans-cisterna

13.4 lm2 N/A

Estimated membrane SA of trans-mostcisterna

52.7 lm2 30.4 lm2

Estimated total membrane SA 380 lm2 260 lm2

a Results are expressed as means ± SEM.


ARTICLE IN PRESS

ber and size between beta cells within the same islet, evi-dence exists for functional heterogeneity at the level of anindividual islet beta cell’s capacity to exocytose insulingranules in response to stimulation with elevated extracel-lular glucose (Bennett et al., 1996; Heimberg et al., 1993;Leung et al., 2005). Inversely, ribbon01 contained more(1.35·) immature insulin granules than ribbon02 (�700 ver-sus �520) (Supplemental Data S3). Multi-granular/-vesicu-lar bodies—compartments of lysosomal origin thatfunction in the intracellular degradation/dynamic turnoverof ‘aged’ insulin granules—were more abundant (2.33·)in ribbon02 relative to ribbon01 (28 versus 12, respectively)(Bommer et al., 1976; Borg and Schnell, 1986;Halban, 1991; Halban and Wollheim, 1980; Halbanet al., 1980, 1987; Orci et al., 1984) (Table 1; SupplementalData S5).

3.4. Organization of the Golgi ribbon

The Golgi ribbon—assessed here in toto in 3D at �15–20 nm resolution by ET for the first time in mammaliancells—exhibited a similar reticulated morphology overallin ribbon01 and ribbon02 (Fig. 5; Supplementary DataS3). In each case, the Golgi ribbon resembled a complexfenestrated structure in which the trans-face of the Golgiwas oriented inward, while the cis-Golgi was usually ori-ented facing the plasma membrane. To expedite segmenta-tion of the entire Golgi ribbon in each cell whilst stillproviding general information with regard to total mem-brane surface area and volume for the Golgi ribbon, themain stack—comprised of cis–medial-cisternae (gray mesh;green model contours in the Fig. 5, inset)—was segmentedas a single object using bounding polygons on every third


tomographic slice. To more accurately calculate theapproximate volume and surface area of the main Golgiribbon, 50 slices in each cell were randomly selected andall cisternae were individually segmented (Fig. 5B andinset) to provide a more accurate estimate of cisternal vol-ume versus bounding volume and cisternae surface areaversus bounding surface area. Thus, this ‘limited sampling’approach helped guide and refine the accuracy of our mea-surements for the Golgi ribbon without requiring us to seg-ment each individual cisterna.

In ribbon01, the Golgi had a total estimated volume (V)of approximately 5.8 lm3 and membrane surface area (SA)of 380 lm2—calculated by summing together the mainstack (V �3.6 lm3, SA �310 lm2), penultimate trans-cis-terna (V �0.35 lm3, SA �13.4 lm2) and trans-most cis-terna (V �1.9 lm3, SA �52.7 lm2). In contrast, inribbon02 the Golgi ribbon comprised an estimated totalvolume of approximately 3.7 lm3 and membrane SA of260 lm2—as calculated by summing together the mainstack (V �3.1 lm3, SA �230 lm2) plus trans-most cisterna(V �0.53 lm3, SA �30.4 lm2). The penultimate Golgi inribbon02 was not segmented separately from the mainstacks/ribbon because it could not be identified reliablyas was the case for ribbon01, both due to reduced resolu-tion in ribbon02 and because the cisterna itself was ‘sparse’(Table 1). Significantly, the Golgi ribbon accounted foralmost twice (1.85·) the total non-nuclear/cytoplasmic cellvolume in ribbon01 compared to ribbon02 (1.15% versus0.62%, respectively).

3.5. Mitochondria: relating structure to function

In the present study, each mitochondrion was seg-mented by drawing a series of connected spheres alongits length rather than our usual lengthy process of man-ually drawing contours tomographic slice by slice (Marshet al., 2001b) (Supplemental Data S3 & S4). Individualnon-branched mitochondria are colored green (Figs. 6and 7A and B). For branched mitochondria, the mainlengths are colored light blue, the branches are coloredmauve and the branch points highlighted in red (Figs.6 and 7A and B). The main length was defined as thestraightest path along the length of a given mitochon-drion. However, it should be noted that the classificationof main length versus branch(es) was sometimes arbi-trary. For example, in Fig. 6A it is unclear exactly whichportion of the mitochondrion could be considered thebranch. Moreover, it remains somewhat open for debatewhether these kinds of mitochondria should be consid-ered as a single mitochondrion, two mitochondria inthe process of fusion or fission, or three separate mito-chondrial lengths (Chen and Chan, 2005; Frazier et al.,2006; Karbowski and Youle, 2003; Okamoto and Shaw,2005; Perkins and Frey, 2000).

Here, for the purposes of simplicity, we count the totalnumber of mitochondria per cell as the number of single(unbranched) mitochondria (S) plus the number of main



ARTICLE IN PRESS

lengths (M) for the branched mitochondria. For abranched mitochondrion with n branches, we calculate itslength as equal to the end-to-end length of the main length(m) end plus the length of all its branches (b1+ . . .+bn)(Fig. 6D).

A summary of quantitative data derived for mitochon-dria in ribbon01 and ribbon02 has been compiled in Table2. In ribbon01, there were a total of 249 mitochondria, ofwhich 10% had branches. Of the 26 branched mitochon-dria, most (17; 65%) had only a single branch (Fig. 6A);8 (31%) had two branches (Fig. 6B) but only one (4%)had three branches. In ribbon02, there were 168 mitochon-dria, of which only 6% had branches. Of these 10 branchedmitochondria, 9 had only a single branch while just onehad two branches. The significant difference in the relativeproportion of branched mitochondria per cell and the aver-age number of branches per branched mitochondrionbetween the two cells suggests that the mitochondrial pop-ulation in ribbon01 were more functionally active with ahigher rate of fission/fusion than ribbon02 (Chen andChan, 2005; Chen et al., 2005) (Table 2). Notably, the aver-age cumulative length of branched mitochondria was con-siderably greater than that of non-branched mitochondriafor both cells (2.5· versus 3.7· for ribbon01 and ribbon02,respectively), consistent with the idea that shorter (single)mitochondria are more stable and less likely to be involvedin fusion/fission events, whereas longer (branched) mito-chondria provide a greater surface where they can poten-tially form new branches or merge/fuse with othermitochondria (Chen et al., 2005) (Fig. 8). It is also note-worthy that the length of branched as well as non-branchedmitochondria was significantly greater for ribbon01 thanribbon02 (1.35· versus 1.97·), suggesting that the increasedproportion of branched mitochondria in ribbon01 itselfdoes not account for the difference in mitochondrial lengthbetween the two cells (Fig. 8; Table 2; Supplemental DataS4).

Table 2A comparative summary of quantitative data derived for mitochondria inribbon01 and ribbon02

ribbon01 ribbon02

Total number of mitochondria 249 168Number of branched mitochondria 26 (10%) 10 (6%)Mean number of branches per branched

mitochondria1.38 1.1

Mean length of non-branchedmitochondriaa,b

2300 ± 107 nm 1170 ± 86 nm

Mean diameter of non-branchedmitochondria

304 ± 2.8 nm 296 ± 3.18 nm

Mean length of branched mitochondria 5870 ± 468 nm 4320 ± 635 nmMean diameter of branched mitochondria 297 ± 4.0 nm 280 ± 7.2 nmEstimated total volume of all

mitochondria49 lm3 21 lm3

Estimated membrane SA of allmitochondria

635 lm2 301 lm2

a Results are expressed as means ± SEM.b Indicates a significant difference in the length of branched mitochon-

dria in ribbon01 versus ribbon02 (p < 0.0005).


As noted elsewhere and above, mitochondria are in aconstant state of dynamic flux between fusion and fissionevents, with mitochondrial respiration and metabolismspatially and temporally regulated by the morphologyand positioning of the organelle (Chen et al., 2005;McBride et al., 2006). With its increased proportion ofbranching mitochondria and presumably increased inci-dence of fission/fusion, ribbon01 provided us with a num-ber of opportunities to identify sites along the lengths ofmitochondria that thinned considerably—and tentativelyrepresent static ‘snapshots’ of such events (Fig. 6C).

The global organization of mitochondria within each ofthe cells was itself quite striking (Fig. 7; Supplemental DataS4). In ribbon01, a large proportion of the mitochondriawere closely apposed to the Golgi ribbon; 95 (�38%) ofall mitochondria in the cell clustered within 400 nm ofthe Golgi (Fig. 7C and E; Supplemental Data S4), despitethe fact that this space only accounts for �14% of the cyto-plasmic volume. In comparison, in ribbon02, only 35(�20%) mitochondria approached within 400 nm of theGolgi ribbon (Fig. 7D and F; Supplemental Data S4).

4. Discussion

In the present study, we have employed methods forefficient tomographic reconstruction and analysis of wholecells to compare structure–function relationships betweentwo ‘equivalent’ insulin-secreting beta cells (designated‘ribbon01’ and ‘ribbon02’) that were similar in size(�610 lm3 and �698 lm3, respectively) and located atcomparable positions at the periphery of an (intact) iso-lated mouse islet of Langerhans. This was important,since functional heterogeneity in terms of responsivenessto stimulation with elevated extracellular glucose as wellas capacity for the rapid exocytic release of insulin gran-ules has been reported for beta cells of different sizes and/or positions within the same pancreatic islet (Leung et al.,2005; Rocheleau et al., 2004).

Collectively, the relative differences between the twocells in terms of the number, dimensions and distributionof insulin granules, mitochondria and the Golgi, stronglysuggest inverse relationships among these organelles thatvary depending on the level of biosynthetic activity withinthe cell. Moreover, our data provide structure–functionevidence that upon comparison of the two cells, ribbon01

represents a beta cell that has responded more vigorouslyto glucose-stimulation and appears to have discharged alarge proportion (>60%) of its intracellular insulin granulestores (based on previous estimates of islet beta cell granulecontent) (Dean, 1973; Olofsson et al., 2002; Rorsman andRenstrom, 2003). Conversely, ribbon02 appears to have dis-charged <10% of its total insulin granule stores.

In addition, it is important to note that insulin secretedfrom the beta cell is rapidly replaced due to a specific glu-cose-induced increase in (pro)insulin biosynthesis (Rhodes,2004). Typically, i.e., following short-term glucose stimula-tion (<4 h), (pro)insulin synthesis occurs exclusively at the



ARTICLE IN PRESS

translational level (Wicksteed et al., 2003). Indeed, the factthat the threshold glucose concentration required to stimu-late (pro)insulin biosynthesis (2–4 mM) is lower than thatrequired to stimulate insulin granule exocytosis (4–6 mM)ensures that insulin production is maintained under condi-tions where insulin secretion is negligible (Ashcroft, 1980),and promotes the beta cell’s bias towards continual replen-ishment of its insulin stores (Ashcroft, 1980; Rhodes,2004). Notably, the ratio of mature: immature granuleswas approximately 5:1 and 16:1 for ribbon01 and ribbon02,respectively.

Here, taken together with our finding that the intracellu-lar granule pool in ribbon01 appears substantially depletedcompared to ribbon02, the marked increase between thetwo cells in terms of the difference in Golgi volume/mem-brane surface area (5.8 lm3; 380 lm2 versus 3.7 lm3;260lm2), particularly with respect to trans-Golgi cisternaealone (2.25 lm3; 66.1 lm2 versus 0.53 lm3; 30.4 lm2) (Sup-plemental Data S5), strongly indicated an upregulation ofprotein biosynthetic activity in ribbon01. Likewise, the rel-ative increases in mitochondrial number, length, extent ofbranching as well as the proportion of mitochondria thatwere clustered in close proximity to the Golgi ribbon in rib-

bon01 compared to ribbon02 were all commensurate withan increased level of mitochondrial function/activity inresponse to additional energy demands stemming from ele-vated (pro)insulin biosynthesis (Fig. 8; Supplemental DataS5). Indeed, in ribbon01 mitochondria accounted for 8.0%(2.2· more) of the cytoplasmic volume compared with3.6% in ribbon02 (see Supplemental Data S5).

Finally—consistent with the idea that regulated secre-tory cells such as the beta cells of the endocrine pancreasmust dynamically regulate the balance between the biosyn-thesis of new insulin granules versus the exocytic/extracel-lular release of insulin or the intracellular digestion of oldergranules (Bommer et al., 1976; Borg and Schnell, 1986;Halban, 1991; Halban and Wollheim, 1980; Halbanet al., 1980; Halban et al., 1987; Orci et al., 1984) to ensurethat cellular insulin stores are maintained at appropriatelevels—it is noteworthy that the number of identifiable deg-radative compartments in ribbon02 is more than 2-foldgreater than ribbon01 (Supplemental Data S5). Also consis-tent with the notion that ribbon02 housed a ‘more mature’granule population was our finding that the average diam-eter of insulin granules was smaller in ribbon02 than in rib-

bon01 (245 nm versus 280 nm), since it has been reportedelsewhere that insulin granules shrink as they age (Hutton,1989).

Taken together, our data suggest that ribbon01 had dis-charged the bulk of its insulin granule stores in response tostimulation with extracellular glucose and was in the throesof preparing for subsequent rounds of new insulin granuleproduction at the time the islet was frozen. In contrast, rib-

bon02 appeared not to have responded strongly to the glu-cose stimulus prior to islet freezing, with its Golgi ribbonand mitochondrial population apparently maintainedunder steady-state conditions.


5. Conclusions

We have successfully developed an expedited approach(on a timescale of weeks rather than months) that hasallowed us to reconstruct and view mammalian cells indetail (at �15–20 nm resolution) in their entirety in 3Dby ET. In parallel, we have developed/applied a numberof abbreviated segmentation techniques that have enabledus to extract useful quantitative and spatial informationregarding structure–function relationships of organelles ata cellular level on a timescale of months rather than years.Taken together, these approaches for the efficient 3Dreconstruction and analysis of whole mammalian cells pro-vide us with a set of tools that—although comparativelycrude relative to conventional methods for subcellulartomography—offer a level of insight into the complex 3Dorganization of mammalian cells not currently affordedby other techniques. Notably, these methods are notintended to replace high-resolution ET approaches thatyield 3D cellular information at approximately 5 nm reso-lution or better, but with concomitant demands on timeand resources.

That said, we continue to explore a number of optionsfor improving the accuracy of quantitative data obtainedfrom future cellular reconstructions, but with the aim ofmaintaining a compromise between the efficiency of wholecell tomographic data collection and resolution (Marsh,2005). First, to improve isotropy whilst minimizing thestrength of noise artifacts that originate from data collectedaround a single axis, we propose to assess the contributionof a limited set of tilts from a second tilt series orthogonal tothe first (Mastronarde, 1997; Penczek et al., 1995). Second,to minimize the extent to which errors accumulate as a con-sequence of using uniform transforms to align volumes gen-erated from large sections, we are currently testing methodsfor non-uniform distortion and alignment of serial volumes.Third, we are developing additional tools to compensate formissing material that typically occurs at the top and bottomof tomograms following 3D reconstruction through theaddition of ‘ghost’ tomographic slices to substitute for miss-ing cellular volume. Finally, we are working to implement anumber of ‘smart’ yet interactive interpolative tools thatwill offer both improved speed and accuracy for segmenta-tion on a day-to-day basis.

Acknowledgments

We gratefully acknowledge the assistance of PamelaHollis from the Luxel Corporation (Friday Harbor, WA,USA), who has orchestrated collaborative support in theform of provision and quality control of new specimen sup-ports for testing for cryo-EM and ET. This work was sup-ported by an Australian Postgraduate Award (APA)scholarship to ABN, and grants from the Juvenile DiabetesResearch Foundation International (2-2004-275) and theNational Institutes of Health (DK-71236) to B.J.M.B.J.M. is a Senior Research Affiliate of the ARC Special



ARTICLE IN PRESS

Research Centre for Functional and Applied Genomics.The Advanced Cryo-Electron Microscopy Laboratoryhoused at the Institute for Molecular Bioscience is a majornode of the National Collaborative Research Infrastruc-ture Strategy-funded ‘National Facility for AdvancedMicroscopy and Microanalysis’, and is jointly supportedby the Queensland State government’s ‘Smart State Strat-egy’ initiative. We thank nanoTechnology Systems(Greensborough, Vic., Australia) for critical maintenanceand upgrades of the 300 keV Tecnai F30 EM within thelaboratory. The Cell Centered Database (CCDB) is sup-ported by NIH grants from NCRR RR04050, RR08605and the Human Brain Project DA016602 from the Na-tional Institute on Drug Abuse, the National Institute ofBiomedical Imaging and Bioengineering and the NationalInstitute of Mental Health. Finally, we sincerely thankthe three anonymous reviewers for their time and construc-tive criticisms that led to an improved final manuscript.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.jsb.2007.09.015.

References

Arita, M., Robert, M., Tomita, M., 2005. All systems go: launching cellsimulation fueled by integrated experimental biology data. Curr. Opin.Biotechnol. 16, 344–349.

Ashcroft, S.J., 1980. Glucoreceptor mechanisms and the control of insulinrelease and biosynthesis. Diabetologia 18, 5–15.

Bennett, B.D., Jetton, T.L., Ying, G., Magnuson, M.A., Piston, D.W.,1996. Quantitative subcellular imaging of glucose metabolism withinintact pancreatic islets. J. Biol. Chem. 271, 3647–3651.

Biel, S.S., Kawaschinski, K., Wittern, K.P., Hintze, U., Wepf, R., 2003.From tissue to cellular ultrastructure: closing the gap between micro-and nanostructural imaging. J. Microsc. 212, 91–99.

Bommer, G., Schafer, H.J., Kloppel, G., 1976. Morphologic effects ofdiazoxide and diphenylhydantoin on insulin secretion and biosynthesisin B-cells of mice. Virchows Arch. A Pathol. Anat. Histol. 371,227–241.

Bonner-Weir, S., 1988. Morphological evidence for pancreatic polarity ofbeta cell within islets of Langerhans. Diabetes 37, 616–621.

Bonner-Weir, S., 1989. Pancreatic islets: morphology, organization, andphysiological implications. In: Draznin, B. et al. (Eds.), Molecular andCellular Biology of Diabetes Mellitus. Alan. R. Liss, Inc., New York,pp. 1–11.

Bonner-Weir, S., Orci, L., 1982. New perspectives on the microvasculatureof the islets of Langerhans in the rat. Diabetes 31, 883–889.

Borg, L.A., Schnell, A.H., 1986. Lysosomes and pancreatic islet function:intracellular insulin degradation and lysosomal transformations.Diabetes Res. 3, 277–285.

Bork, P., Serrano, L., 2005. Towards cellular systems in 4D. Cell 121,507–509.

Burrage, K., Hood, L., Ragan, M.A., 2006. Advanced computing forsystems biology. Brief Bioinform. 7, 390–398.

Chen, H., Chan, D.C., 2005. Emerging functions of mammalian mito-chondrial fusion and fission. Hum. Mol. Genet. 14 (2), R283–R289.

Chen, H., Chomyn, A., Chan, D.C., 2005. Disruption of fusion results inmitochondrial heterogeneity and dysfunction. J. Biol. Chem. 280,26185–26192.


Coggan, J.S., Bartol, T.M., Esquenazi, E., Stiles, J.R., Lamont, S.,Martone, M.E., Berg, D.K., Ellisman, M.H., Sejnowski, T.J., 2005.Evidence for ectopic neurotransmission at a neuronal synapse. Science309, 446–451.

Dean, P.M., 1973. Ultrastructural morphometry of the pancreatic b-cell.Diabetologia 9, 115–119.

Egner, A., Verrier, S., Goroshkov, A., Soling, H.D., Hell, S.W., 2004. 4Pi-microscopy of the Golgi apparatus in live mammalian cells. J. Struct.Biol. 147, 70–76.

Frazier, A.E., Kiu, C., Stojanovski, D., Hoogenraad, N.J., Ryan, M.T.,2006. Mitochondrial morphology and distribution in mammalian cells.Biol. Chem. 387, 1551–1558.

Gaietta, G.M., Giepmans, B.N., Deerinck, T.J., Smith, W.B., Ngan, L.,Llopis, J., Adams, S.R., Tsien, R.Y., Ellisman, M.H., 2006. Golgitwins in late mitosis revealed by genetically encoded tags for live cellimaging and correlated electron microscopy. Proc. Natl. Acad. Sci.USA 103, 17777–17782.

Gotoh, M., Maki, T., Kiyoizumi, T., Satomi, S., Monaco, A.P., 1985. Animproved method for isolation of mouse pancreatic islets. Transplan-tation 40, 437–438.

Greider, M.H., Howell, S.L., Lacy, P.E., 1969. Isolation and properties ofsecretory granules from rat islets of Langerhans. II. Ultrastructure ofthe beta granule. J. Cell. Biol. 41, 162–166.

Halban, P.A., 1991. Structural domains and molecular lifestyles ofinsulin and its precursors in the pancreatic beta cell. Diabetologia34, 767–778.

Halban, P.A., Wollheim, C.B., 1980. Intracellular degradation of insulinstores by rat pancreatic islets in vitro. An alternative pathway forhomeostasis of pancreatic insulin content. J. Biol. Chem. 255,6003–6006.

Halban, P.A., Wollheim, C.B., Blondel, B., Renold, A.E., 1980. Long-term exposure of isolated pancreatic islets to mannoheptulose:evidence for insulin degradation in the beta cell. Biochem. Pharmacol.29, 2625–2633.

Halban, P.A., Mutkoski, R., Dodson, G., Orci, L., 1987. Resistance of theinsulin crystal to lysosomal proteases: implications for pancreatic B-cell crinophagy. Diabetologia 30, 348–353.

Hara, M., Dizon, R.F., Glick, B.S., Lee, C.S., Kaestner, K.H., Piston,D.W., Bindokas, V.P., 2006. Imaging pancreatic beta cells in theintact pancreas. Am. J. Physiol. Endocrinol. Metab. 290,E1041–E1047.

Heimberg, H., De Vos, A., Vandercammen, A., Van Schaftingen, E.,Pipeleers, D., Schuit, F., 1993. Heterogeneity in glucose sensitivityamong pancreatic beta cells is correlated to differences in glucosephosphorylation rather than glucose transport. EMBO J. 12,2873–2879.

Hoog, J.L., Schwartz, C., Noon, A.T., O’Toole, E.T., Mastronarde, D.N.,McIntosh, J.R., Antony, C., 2007. Organization of interphase micro-tubules in fission yeast analyzed by electron tomography. Dev. Cell 12,349–361.

Howell, S.L., Tyhurst, M., 1974. Cryo-ultramicrotomy of islets ofLangerhans. Some observations on the fine structure of mammalianislets in frozen sections. J. Cell. Sci. 15, 591–603.

Howell, S.L., Kostianovsky, M., Lacy, P.E., 1969. Beta granule formationin isolated islets of langerhans: a study by electron microscopicradioautography. J. Cell. Biol. 42, 695–705.

Hutton, J.C., 1989. The insulin secretory granule. Diabetologia 32,271–281.

Karbowski, M., Youle, R.J., 2003. Dynamics of mitochondrial morphol-ogy in healthy cells and during apoptosis. Cell Death Differ. 10,870–880.

Kremer, J.R., Mastronarde, D.N., McIntosh, J.R., 1996. Computervisualization 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., McIntosh, J.R., 1990. Characterization of beam-induced thinningand shrinkage of semi-thick sections in the HVEM. Proc. XII Int.Congr. Electr. Microsc. 3, 752–753.


http://dx.doi.org/10.1016/j.jsb.2007.09.015

http://dx.doi.org/10.1016/j.jsb.2007.09.015


ARTICLE IN PRESS

Kurner, J., Frangakis, A.S., Baumeister, W., 2005. Cryo-electron tomog-raphy reveals the cytoskeletal structure of Spiroplasma melliferum.Science 307, 436–438.

Ladinsky, M.S., Mastronarde, D.N., McIntosh, J.R., Howell, K.E.,Staehelin, L.A., 1999. Golgi structure in three dimensions: functionalinsights from the normal rat kidney cell. J. Cell. Biol. 144, 1135–1149.

Lehner, B., Tischler, J., Fraser, A.G., 2005. Systems biology: where it’s atin 2005. Genome Biol. 6, 338.

Leung, Y.M., Ahmed, I., Sheu, L., Tsushima, R.G., Diamant, N.E., Hara,M., Gaisano, H.Y., 2005. Electrophysiological characterization ofpancreatic islet cells in the mouse insulin promoter-green fluorescentprotein mouse. Endocrinology 146, 4766–4775.

Luther, P.K., Lawrence, M.C., Crowther, R.A., 1988. A method formonitoring the collapse of plastic sections as a function of electrondose. Ultramicroscopy 24, 7–18.

Ma, L., Bindokas, V.P., Kuznetsov, A., Rhodes, C., Hays, L., Edwardson,J.M., Ueda, K., Steiner, D.F., Philipson, L.H., 2004. Direct imagingshows that insulin granule exocytosis occurs by complete vesiclefusion. Proc. Natl. Acad. Sci. USA 101, 9266–9271.

Marsh, B.J., 2005. Lessons from tomographic studies of the mammalianGolgi. Biochim. Biophys. Acta 1744, 273–292.

Marsh, B.J., 2006. Toward a Visible Cell and beyond. AustralianBiochemist 37, 5–10.

Marsh, B.J., 2007. Reconstructing mammalian membrane architecture bylarge area cellular tomography. Methods Cell Biol. 79C, 193–220.

Marsh, B.J., Mastronarde, D.N., McIntosh, J.R., Howell, K.E., 2001a.Structural evidence for multiple transport mechanisms through theGolgi in the pancreatic beta cell line, HIT-T15. Biochem. Soc. Trans.29, 461–467.

Marsh, B.J., Volkmann, N., McIntosh, J.R., Howell, K.E., 2004. Directcontinuities between cisternae at different levels of the Golgi complexin glucose-stimulated mouse islet beta cells. Proc. Natl. Acad. Sci.USA 101, 5565–5570.

Marsh, B.J., Mastronarde, D.N., Buttle, K.F., Howell, K.E., McIntosh,J.R., 2001b. Organellar relationships in the Golgi region of thepancreatic beta cell line, HIT-T15, visualized by high resolutionelectron tomography. Proc. Natl. Acad. Sci. USA 98, 2399–2406.

Martone, M.E., Gupta, A., Wong, M., Qian, X., Sosinsky, G., Ludascher,B., Ellisman, M.H., 2002. A cell-centered database for electrontomographic data. J. Struct. Biol. 138, 145–155.

Mastronarde, D.N., 1997. Dual-axis tomography: an approach withalignment methods that preserve resolution. J. Struct. Biol. 120,343–352.

Mastronarde, D.N., 2003. SerialEM: A program for automated tilt seriesacquisition on Tecnai microscopes using prediction of specimenposition. Microsc. Microanal. 9, 1182–1183.

Mastronarde, D.N., 2005. Automated electron microscope tomographyusing robust prediction of specimen movements. J. Struct. Biol. 152,36–51.

McBride, H.M., Neuspiel, M., Wasiak, S., 2006. Mitochondria: more thanjust a powerhouse. Curr. Biol. 16, R551–R560.

McEwen, B.F., Marko, M., 1999. Three-dimensional transmission elec-tron microscopy and its application to mitosis research. Methods Cell.Biol. 61, 81–111.

Medalia, O., Weber, I., Frangakis, A.S., Nicastro, D., Gerisch, G.,Baumeister, W., 2002. Macromolecular architecture in eukaryotic cellsvisualized by cryoelectron tomography. Science 298, 1209–1213.

Medalia, O., Beck, M., Ecke, M., Weber, I., Neujahr, R., Baumeister, W.,Gerisch, G., 2007. Organization of actin networks in intact filopodia.Curr. Biol. 17, 79–84.

Michael, D.J., Geng, X., Cawley, N.X., Loh, Y.P., Rhodes, C.J., Drain,P., Chow, R.H., 2004. Fluorescent cargo proteins in pancreatic betacells: design determines secretion kinetics at exocytosis. Biophys. J. 87,L03–L05.

Mogelsvang, S., Marsh, B.J., Ladinsky, M.S., Howell, K.E., 2004.Predicting function from structure: 3D structure studies of themammalian Golgi complex. Traffic 5, 338–345.


Nickell, S., Kofler, C., Leis, A.P., Baumeister, W., 2006. A visualapproach to proteomics. Nat. Rev. Mol. Cell. Biol. 7, 225–230.

Nicolls, M.R., Coulombe, M., Beilke, J., Gelhaus, H.C., Gill, R.G., 2002.CD4-dependent generation of dominant transplantation toleranceinduced by simultaneous perturbation of CD154 and LFA-1 pathways.J. Immunol. 169, 4831–4839.

Okamoto, K., Shaw, J.M., 2005. Mitochondrial morphology and dynam-ics in yeast and multicellular eukaryotes. Annu. Rev. Genet. 39,503–536.

Olofsson, C.S., Gopel, S.O., Barg, S., Galvanovskis, J., Ma, X., Salehi, A.,Rorsman, P., Eliasson, L., 2002. Fast insulin secretion reflectsexocytosis of docked granules in mouse pancreatic B-cells. PflugersArch 444, 43–51.

Orci, L., 1974. A portrait of the pancreatic B-cell. The Minkowski AwardLecture delivered on July 19, 1973, during the 8th Congress of theInternational Diabetes Federation, held in Brussels, Belgium. Diabet-ologia 10, 163–187.

Orci, L., 1976a. The microanatomy of the islets of Langerhans. Metab-olism 25, 1303–1313.

Orci, L., 1976b. Some aspects of the morphology of insulin-secreting cells.Acta Histochem. 55, 147–158.

Orci, L., 1985. The insulin factory: a tour of the plant surroundings and avisit to the assembly line. The Minkowski lecture 1973 revisited.Diabetologia 28, 528–546.

Orci, L., Ravazzola, M., Amherdt, M., Yanaihara, C., Yanaihara, N.,Halban, P., Renold, A.E., Perrelet, A., 1984. Insulin, not C-peptide(proinsulin), is present in crinophagic bodies of the pancreatic B-cell. J.Cell. Biol. 98, 222–228.

Penczek, P., Marko, M., Buttle, K., Frank, J., 1995. Double-tilt electrontomography. Ultramicroscopy 60, 393–410.

Perkins, G.A., Frey, T.G., 2000. Recent structural insight into mitochon-dria gained by microscopy. Micron 31, 97–111.

Rhodes, C.J., 2004. Processing the insulin molecule. In: LeRoith, D. et al.(Eds.), Diabetes Mellitus, A Fundemental and Clinical Text. Lippin-cott-Raven Publishers, Philadelphia, PA, pp. 27–50.

Rocheleau, J.V., Walker, G.M., Head, W.S., McGuinness, O.P., Piston,D.W., 2004. Microfluidic glucose stimulation reveals limited coordi-nation of intracellular Ca2+ activity oscillations in pancreatic islets.Proc. Natl. Acad. Sci. USA 101, 12899–12903.

Rorsman, P., Renstrom, E., 2003. Insulin granule dynamics in pancreaticbeta cells. Diabetologia 46, 1029–1045.

Shoop, R.D., Esquenazi, E., Yamada, N., Ellisman, M.H., Berg, D.K.,2002. Ultrastructure of a somatic spine mat for nicotinic signaling inneurons. J. Neurosci. 22, 748–756.

Sosinsky, G.E., Giepmans, B.N., Deerinck, T.J., Gaietta, G.M., Ellisman,M.H., 2007. Markers for correlated light and electron microscopy.Methods Cell. Biol. 79, 575–591.

Sosinsky, G.E., Deerinck, T.J., Greco, R., Buitenhuys, C.H., Bartol,T.M., Ellisman, M.H., 2005. Development of a model for microphys-iological simulations: small nodes of ranvier from peripheral nerves ofmice reconstructed by electron tomography. Neuroinformatics 3,133–162.

Soto, G.E., Young, S.J., Martone, M.E., Deerinck, T.J., Lamont, S.,Carragher, B.O., Hama, K., Ellisman, M.H., 1994. Serial sectionelectron tomography: a method for three-dimensional reconstructionof large structures. Neuroimage 1, 230–243.

van Marle, J., Dietrich, A., Jonges, K., Jonges, R., de Moor, E., Vink, A.,Boon, P., van Veen, H., 1995. EM-tomography of section collapse, anonlinear phenomenon. Microsc. Res. Tech. 31, 311–316.

Wicksteed, B., Alarcon, C., Briaud, I., Lingohr, M.K., Rhodes, C.J., 2003.Glucose-induced translational control of proinsulin biosynthesis isproportional to preproinsulin mRNA levels in islet beta cells but notregulated via a positive feedback of secreted insulin. J. Biol. Chem.278, 42080–42090.

Wilson, C.J., Mastronarde, D.N., McEwen, B., Frank, J., 1992. Mea-surement of neuronal surface area using high-voltage electron micro-scope tomography. Neuroimage 1, 11–22.


expedited approaches to whole cell electron tomography and ... · expedited approaches to whole...

Documents