Three-dimensional electron microscopy of entire cells.
ABSTRACT The digital processing of serial electron-microscope sections containing laser-induced topographical references allows a three-dimensional (3-D) reconstruction of entire cells at a depth resolution of 40-60 nm by the use of novel image analysis methods. The images are directly processed by a video-camera placed under the electron microscope in TEM mode or by the electron counting device in STEM mode. The deformations associated with the cutting of embedded cells are back-calculated by new computer algorithms developed for image analysis and treatment. They correct the artefacts caused by serial sectioning and automatically reconstruct the third dimension of the cells. Used in such a way, our data provide definitive information on the 3-D architecture of cells. This computer-assisted 3-D analysis represents a new tool for the documentation and analysis of cell ultrastructure and for morphometric studies. Furthermore, it is now possible for the observer to view the contents of the reconstructed tissue volume in a variety of different ways using computer-aided display techniques.
- [Show abstract] [Hide abstract]
ABSTRACT: The digital processing of electron microscopic images from serial sections containing laser-induced topographical references allows a 3-D reconstruction at a depth resolution of 30 to 40 nm of entire cells by the use of image analysis methods, as already demonstrated for Transmission Electron Microscopy (TEM) coupled with a video camera. We decided to use a Scanning Transmission Electron Microscope (STEM) to get higher contrast and better resolution at medium magnification. The scanning of our specimens at video frequencies is an attractive and easy way to link a STEM with an image processing system but the hysteresis of the electronic spools responsible for the magnetic deviation of the scanning electron beam induces deformations of images which have to be modelized and corrected before registration. Computer algorithms developed for image analysis and treatment correct the artifacts caused by the use of STEM and by serial sectioning to automatically reconstruct the third dimension of the cells. They permit the normalization of the images through logarithmic processing of the original grey level infonnation. The automatic extraction of cell limits allows to link the image analysis and treatments with image synthesis methods by minimal human intervention. The surface representation and the registered images provide an ultrastructural data base from which quantitative 3-D morphological parameters, as well as otherwise impossible visualizations, can be computed. This 3-D image processing named C.A.V.U.M. for Computer Aided Volumic Ultra-Microscopy offers a new tool for the documentation and analysis of cell ultrastructure and for 3-D morphometric studies at EM magnifications. Further, a virtual observer can be computed in such a way as to simulate a visit of the reconstructed object. © (1990) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only. Topics 3D modeling ; Algorithms ; Cameras ; Computing systems ; Digital signal processing ; Electron beams ; Electron microscopes ; Electron microscopy ; Electronic imaging ; ElectronsProc SPIE 05/1990;
- [Show abstract] [Hide abstract]
ABSTRACT: Images of a scene observed under a variable illumination or with a variable optical aperture are not identical. Does a privileged representant exist? In which physical setting? In which mathematical context? With which meaning and criterion? How to obtain it? The authors answer to such questions in the physical setting of logarithmic imaging processes. For such a purpose, they use the logarithmic image processing (LIP) model, known to be a compatible mathematical framework. After short recalls on this model, the paper presents two image transforms: one performs an optimal enhancement and stabilization of the overall dynamic range, and the other does of the mean dynamic range. The results obtained on X-ray images, as well as for some natural scenes, are shown. Also the implementation of the transforms is addressed.Signal Processing 01/1995; 41(2):225-237. · 2.24 Impact Factor
- Microscopy Microanalysis Microstructures 01/1996; 7.
Journal o f Microscopy, Vol. 157, Pr 1, January 1990, pp. 115-126.
Received 15 March 1989; revised 25 May 1989; accepted 9 June 1989
Three-dimensional electron microscopy of entire cells
byCPH. BRON*, P H . GREMILLETt, D. LAUNAYS,
H. P. G A U T S C H I ~ ,
TH. B A C H I ~
Centre for Retroviruses, and $Central Electron Microscopy Laboratory, Institute for
Immunology and Virology, Gloriastrasse 30, 8028-Ziirich, Switzerland, tLaboratoire
du Traitement du Signal et Instrumentation, Universite' de Saint-Etienne, 26 rue du
Dr Michelon, 42100 Saint-Etienne, France and Scornputimage GmbH, 6300 Zug,
*Swiss National and J . SCHUPBACH*,
K E Y WORD s . 3-dimensional, reconstructive electron microscopy, electron micro-
scopy, video electron microscopy, entire cells, serial sections, image analysis, image
processing, image synthesis, morphometry, 3-D ultrastructure, 3-D analysis.
The digital processing of serial electron-microscope sections containing laser-
induced topographical references allows a three-dimensional (3-D) reconstruction of
entire cells at a depth resolution of 4&6Onm by the use of novel image analysis
methods. The images are directly processed by a video-camera placed under the
electron microscope in TEM mode or by the electron counting device in STEM mode.
The deformations associated with the cutting of embedded cells are back-calculated by
new computer algorithms developed for image analysis and treatment. They correct
the artefacts caused by serial sectioning and automatically reconstruct the third dimen-
sion of the cells. Used in such a way, our data provide definitive information on
the 3-D architecture of cells. This computer-assisted 3-D analysis represents a new
tool for the documentation and analysis of cell ultrastructure and for morphometric
studies. Furthermore, it is now possible for the observer to view the contents of the
reconstructed tissue volume in a variety of different ways using computer-aided
Three-dimensional (3-D) reconstruction may directly provide new information
about the structure and, indirectly, the function of cells. While various methods such
as light or laser scanning microscopy are now available for 3-D reconstruction of entire
cells, no such tool has been available at the EM level. However, sections of cells or
tissues used for transmission electron microscopy (TEM) and scanning transmission
electron microscopy (STEM) convey only a limited amount of volume information,
because they are thin or ultrathin, and because the volume information is lost during
the cutting procedure of the polymer block. Sjostrand (1958) was the first to show that
1990 The Royal Microscopical Society
Cph. Bron et al.
serial section reconstruction with ultrathin sections is a feasible and necessary exten-
sion of ordinary EM analysis but the use of computer-aided techniques now enables
us to automatize a part of the reconstruction experiment, to back calculate the
deformations, to quantify precisely the 3-D images and to permit new visualizations
by the means of computer-aided virtual dissection. Until now, 3-D reconstruction,
whose aim is to retrieve the lost volume information, has been confronted by two
major problems: (i) the abundance of information to be processed. This problem has
been circumvented by restricting 3-D reconstruction to those structures of particular
interest whose shape was then defined by manual tracing (Ware & LoPresti, 1975).
The disadvantage is that most of the structural information in the block is lost in this
method. (ii) The deformations associated with cutting prevent a precise alignment of
corresponding structural elements on neighbouring sections. This aspect has tended
to be simply neglected and profiles have been digitized without consideration to these
deformations. The 'best fit', even when done with expert judgement has been no more
than a compromise and, consequently, detailed results are not fully reliable (Ware &
LoPresti, 1975; Briarty & Jenkins, 1984). In addition, manual tracing is very time-
consuming and represents the limiting step in all 3-D reconstruction experiments
(Ware & LoPresti, 1975; Levinthal, 1984).
In setting out to achieve a better solution for 3-D EM reconstruction, our first goal
was to define the deformations resulting from cutting mathematically. We chose to do
these studies with the Merkel cell, a rare cell of the epidermis (Saurat et al., 1983). This
cell possesses a particular feature: it contains thousands of spherical granules with
diameters of 100 nm. After serial cutting, these granules were used as internal corre-
lation points of neighbouring sections. We found that we could approximate the
deformations with a relatively simple mathematical model: a first-order affine function
with six parameters (Cph. Bron et al., unpublished data; Gremillet, 1987). A minimum
of three points of reference per section (fiducials) were sufficient to permit the recon-
struction of all types of cells or biological tissues. Such fiducials can be generated by
the use of an Excimer laser in cells or tissues otherwise devoid of internal correlation
MATERIALS AND METHODS
Preparation of the cells for electron microscopy
Tissues or cells were fixed with glutaraldehyde-acrolein, contrasted en bloc with
uranyl acetate, dehydrated, and embedded in Araldite or Epon.
The blocks were subjected to Excimer pulse laser treatment just before the cutting
procedure. Excimer lasers deliver a U.V. light of short wavelength (1931x11) with a
high surface energy which causes the disruption of intramolecular chemical bonds.
The U.V. light is produced by high-voltage electrical excitation of rare gases. Each
pulse increases the depth of a drilled cylinder in the polymer block. The diameter and
the position of the drilling were controlled with a microscope. Cylinders of 5 pm in
diameter and between 250 and 500 pm deep were drilled. These markers were placed
around the portion of tissue of interest; the distance between two holes was approxi-
mately 20pm depending on the object to be reconstructed.
Ultrathin serial cutting (40 and 60 nm) was done with an Ultracut E (Reichert- Jung)
and a Diatome cutting diamond. In a previous experiment the grids were exposed to
vaporization with platinum at an angle of 12". The section thickness is proportional to
the tangent of the vaporization's angle by the maximal width of the vaporized platinum
3-0 electron microscopy of entire cells
OPTICAL DISK UNIT
Fig. 1. Schematic view of the image analysis and processing system.
‘shadow’ in the laser holes as visualized by EM. The maximal deviation from the
expected results was in all cases within a range inferior or equal to 2.7nm. The
procedure showed that the mean section thickness was constant for each cut and
fairly reproducible. The section thicknesses mentioned above have been obtained by
the same procedure of cutting from block prepared in the same experiment. The
compression occurring during the cutting of Araldite-embedded cells was minimized
to some extent by flattening the sections with chloroform vapour, while they floated
in the boat. The sections were retrieved on carbon-reinforced, Formvar-coated, one-
hole specimen grids (Bakers) according to a slightly modified procedure of Galey and
Nilsson (Galey & Nilsson, 1966). This procedure allowed us to collect up to 500 serial
cuts with a loss of less than 2%.
Image acquisition and processing: hardware
Figure 1 gives a schematic view of the whole image analysis and processing system
used. The EM images taken with a Philips EM 400T were directly processed with an
analogue CF 152 EM camera coupled to a light intensifier (Sofretec, France). Images
were digitized to a density of 512 x 512 pixels and the following image analysis and
treatments were done with a PC-OEILTM card.* This electronic card contains all the
functions to digitize, analyse and process images and was plugged into an IBM PC AT
03TM or another compatible computer. The PC-OEILTM card is a completely self-
sufficient image-processing hardware and uses the IBM-PC bus architecture to
communicate with other processors. The PC drives the card and stores the digitized
data on optical disks (IBMTM 3363) of 200 Mb memory capacity each.
The processing of the images was based on the logarithmic imaging model developed
by Jourlin & Pinoli (1988) and will be described in detail elsewhere. Four different
operations were carried out:
*PC-OEIL card is a trademark of Electronique Lyonnaise. The software and hardware are commercially
available upon request from Computimage GmbH, Baarestrasse 43 CH-6300 Zug and MISIS-Image Sarl
10 bis, rue de la Productique F.42100 Saint-Etienne.
Cph. Bron et al.
Image restoration. In short, three images are needed to permit the calculations
necessary for image restoration. Image A is the image we want to study. Image B is
A at lower magnification; it should contain a minimum of one laser fiducial. Image C
contains all three fiducials (in STEM mode, the image B is also the image C).
From this set of three images, the computer calculates the position of the centres of
gravity of the laser holes. The deformations and the orientation deduced from these
calculation are then applied to each image A in order to fit it into the reconstructed
block. When the EM is not perfectly adjusted, a remaining translation may have to be
Image enhancement. Even with perfect cutting techniques, two adjacent sections
rarely have the same grey-level distribution. This is due to small variations in the
thickness of the cuts which results in different absorption of the electrons. The new
algorithms based on Jourlin & Pinoli’s model (1988) allow us to normalize the grey-
level distribution in all the imaged sections and to enhance the contrast of the images
in calculating their optimal homothesis. All images have the same grey-level distri-
bution after this automatic mathematical treatment, regardless of the distribution
before the treatment.
Image analysis. The cutting procedures and the magnification chosen resulted in
picture elements (pixels) of 40 nm square length. The restoration of the third dimen-
sion thus yielded isotropic volume elements (voxels) of 40 nm side length. The restored
3-D information of the whole reconstructed block is thus composed of 40-nm voxels.
The reconstructed block is called a virtual block since it is mathematically recon-
stituted. It contains all the information that would otherwise have been lost when only
individual sections of a block are evaluated.
Image reconstruction. An automatic feature extraction system specifically developed
for EM analysis will be described elsewhere. It already enables us to extract cell
borders and principal components of the cell automatically and gives correct perspec-
tive and depth cueing. In the same way, the parameters of the deformations calculated
at low magnification can be applied to high-magnification images. .The high-
magnification 3-D information is no longer isotropic since high magnification is
restricted to the first and second dimensions. This display is nevertheless useful for the
study of small details such as viruses.
Laser markers and image standardization
The determination of the centres of gravity of the laser holes is the basis for the
computation of the deformations in each section and the orientation with respect to the
other sections. The left row of Fig. 2 shows low-magnification EM images of two
successive sections. The laser markers in both pictures are clearly visible. On the right
side, the same two sections are shown after the corrections for deformation and the
orientation of the cuts have been applied. The laser marks are now in the same position
and distortions of the cell shapes have been corrected. The result of this step is an
almost perfect fit by superimposition.
Figure 3 shows the grey-level distribution of a representative cut (only a narrow
segment of the whole grey-level range is used) and the effect of grey-level optimization.
The diagram also indicates use of the whole grey-level range resulting in improved
contrast. In particular, the cells contrast much better against the background resin
and, within the cell, the uranyl-acetate-contrasted structures (nucleus membrane and
mitochondria) are considerably enhanced. Furthermore, correction is made for the
grey-level differences of individual cuts.
Figure 4 shows the combined results of geometric and grey-level standardization:
two consecutive sections that were positioned on two different grids, as they were
3-0 electron microscopy of entire cells
Fig. 2. Low-magnification EM images of two successive sections. The laser markers in both pictures are
clearly visible. On the right side, the same two sections are shown after the corrections for deformation and
the orientation of the cuts have been applied. The laser marks are now in the same position and distortions
of the cell shapes have been corrected.
digitized from the video camera, i.e. before the application of the corrections, are
shown with the same images after standardization, which included rotation of both
images, corrections for cell shapes and grey distribution. These restored and enhanced
images are used for the 3-D reconstruction.
The information contained in all the restored individual images was used for the
reconstruction of the whole cell block that was analysed here. This reconstructed block
is shown in Fig. 5. We see the three orthogonal faces of the block as conventional EM
sections but that offer a view of the inside of the cell. The computer programs allow
us to display the structure from any orientation, two such possibilities are shown
in views (a) and (b), which are taken at a rotation angle of 10". The upper face of the
block is formed by an original individual restored EM image. The vertical sections
were synthesized from the stored volume information of the block. Almost perfect
congruence was achieved in the centre of the block, as demonstrated by the unbroken,
smooth lines corresponding to the nuclear or cellular membranes. However, the match
is not perfect at the fringes of the block where deformations are due to the 'dome effect'
brought about by the analogue technology of the video camera.
As an early developmental step towards the reconstruction of complicated cellular
structures we reconstructed the nucleus of a Merkel cell. This reconstruction was done
interactively, i.e. by manual tracing of the nuclear shape from each restored section.
Cph. Bron et al.
3-0 electron microscopy of entire cells
Fig. 4. Combined results of geometric and grey-level standardization. The left side shows two consecutive
sections that were positioned on two different grids, as they were digitized from the video camera, i.e. before
the application of the corrections. The right side shows the same images after standardization which
included rotation of both images, corrections for cell shapes and grey distribution.
Figure 6(a-c) shows different views of this nucleus which is carved by deep fissures.
Nuclear pores are not visible. The indentation seen in the top left corner of the
nucleus, as best seen in (a) and (b), engulfs a mitochondrion (not shown). Programs
have now been created that permit the automatic reconstruction of major surfaces
of the cell. Figure 7 shows two different views of the automatically reconstructed
surface of an H9 cell, a malignant T-cell line used for the large-scale production of
the AIDS virus HIV. Qualitatively, this synthesis mode is not yet completely satisfac-
tory when compared to the result of interactive reconstruction, but improvements are
Figure 8 shows results of analysis at high magnification. For this, a group of H9 cells
infected with, and producing, HIV was used (H9/HTLV-IIIB). Two neighbouring
cells are shown. A complicated labyrinth system of microvilli and vesicles of the two
cells is seen reaching out into the intercellular space. This is the result of a viral
cytopathic effect. The structures have been enhanced by an artificial shadow cast upon
Fig. 8(a); individual sections were further enhanced by colour (b and c). In addition,
the cells are viewed by a virtual observer who is placed in the intercellular space (a-c).
Fig. 3. Grey-level distribution (diagram on top) of a representative cut (top left). Only a narrow segment
of the whole grey-level range is used. The bottom part of the figure shows the effect of grey-level
optimization. The diagram indicates how use of the whole grey-level range (bottom right) results in much
better contrast. Furthermore, correction is made for the grey-level differences of individual cuts.
Cph. Bron et al.
Fig. 5. Three orthogonal faces of the block as conventional EM sections offering a view of the inside of the
cell. Views (a) and (b) are taken at a rotational angle of 10". The vertical axes are 8040-
of 40 nm thickness).
long (201 sections
A calculated perspective view of the area corresponding to section marked red in (c)
is shown in Fig. 8(d).
The volume of the cell represented in Fig. 5 is composed of 19,419,030 voxels.
The surface of the cell membrane is composed of 496,186 pixels. These automatically
extracted values will allow later the precise measurement of volumes and surfaces with
3-0 electron microscopy of entire cells
Fig. 6. Different views of the Merkel cell nucleus carved by deep fissures. Nuclear pores are not visible
(done at the ETH Ziirich on a Vax and a Tektronic graphic station).
We have demonstrated that images directly digitized from a transmission electron
microscope can be used for 3-D reconstruction of entire cells of any complexity. By
using Excimer laser fiducials applied to the tissue block before sectioning we have
shown that the images of subsequent sections can be automatically corrected for
cutting-induced distortions. In addition, automated image-synthesis procedures have
been developed that enable us to lay open the reconstructed image block in any desired
orientation and to extract and visualize such structures as the nucleus, or the cell
membrane. We are aware of the fact that this latter procedure is not yet fully satisfac-
tory when compared with scanning EM, but we are working on its improvement.
Furthermore, these structures can be viewed as by a simulated observer placed at any
location w i t h i n , or outside of, the cell. Thus, otherwise impossible views can be
generated (Stevens & Trogadis, 1986). Currently, a depth resolution of 40 nm has been
achieved; studies under way suggest that a depth resolution of 15nm should be
possible. Finally, the procedures provide a new approach for obtaining precise
morphometric data. Taken together, these methods for which we propose to use the
designation CAVUM (computer-aided volumetric ultra-microscopy) should permit a
much better ultrastructural analysis of cells and, consequently, a better understanding
of cellular functions, intra- or intercellular interactions, or interactions between cells
and infectious pathogens such as viruses, bacteria, or protozoa: It is likely that these
procedures can also be used for the analysis of structures by light microscopy, in
Cph. Bron et al.
Fig. 7. Two different views of the automatically reconstructed surface of an H 9 cell.
3-0 electron microscopy of entire cells
Cph. Bron et al.
particular to elucidate complicated systems of cellvlar interactions, as present, e.g.
in the brain (Shantz & McCann, 1978). In addition, it is possible to use these two
optical tools interactively, e.g. to select areas of interest by light microscopy, and then
to do the fine analysis by EM.
A relatively simple and affordable personal computer system was used for this
work. This was made possible by both the dramatic progress in microcomputer and
microprocessor technology and the significant reduction in costs for memory capacity
and mass storage of data (Huijsmans et al., 1986). The complete reconstruction of a
block composed of 200 sections requires about half a week. Two days of this time are
needed for the processing of the sections under the EM. The calculations that lead to
the virtual block take 16 h. Virtual cuts to lay open the block at any desired position
are currently done in 6min. This may be improved further. We think that the near
future will bring such an advance in microcomputer technology that complicated
image synthesis can be done with low-cost/high-power dedicated electronic cards
plugged into a PC. New 3-D image treatment algorithms based on this concept are
now under development and will be of use in the automatic extraction of cellular
features. In combination with microcomputer-controlled electron microscopes that
are now also under development, such a system may possibly lead to fully automated
electron microscopes. Reconstruction of 3-D EM may then become a routine tool for
biological analysis (Stevens, 1986).
We wish to thank Lambda Physik for the assistance in Excimer-laser technology,
and Ms A. Sadlo for the help in photographic work. This work was supported by
grants from the Swiss Federal Office of Health, the Swiss Cancer League, the Zurich
Cancer League and the Roche Research Foundation.
Briarty, L.G. & Jenkins, P.H. (1984) GRIDSS an integrated suite of microcomputer programs for three-
dimensional graphical reconstruction from serial sections. 3. Microsc. 134, 121-124.
Galey, F.R. & Nilsson, S.E.G. (1966) A new method for transferring sections from the lipid surface of the
trough through staining solutions to the supporting f i l m of a grid. 3. Ultrasrruct. Res. 14,405-410.
Gremillet, P. (1987) Rapport de DEA informatique et analyse d’image. Universitk de Saint-Etienne, Paris.
Huijsmans, D.P., Lamers, W.H., Los, J.A. & Strackee, J. (1986) Toward computerized morphometric
facilities: a review of 58 software packages. Anatomical Record, 216,449-470.
Jourlin, M. & Pinoli, J.C. (1988) A model for logarithmic image processing LIP. 3. Microsc. 149, 21-35.
Levinthal, C. (1984) The formation of three-dimensional biological structures: computers uses and futures
needs. Ann. N.Y. Acad. Sci. 426, 171-180.
Polak, J.M. (1988) Monoclonal antibodies at the electron microscopical level. Znt. 3. Cancer, Supp. 2, 27.
Saurat, J.H., Chavaz, P., Carraux, P. & Didierjean, L. (1983) A human monoclonal antibody reacting with
Merkel cells: immunofluorescence, immunoperoxidase, and immunoelectron microscopy. 3. Znwest.
Dermatol. 81, 202-203.
Shantz, M.J. & McCann, G.D. (1978) Three-dimensional computer graphics for electron microscopy.
ZEEE Trans. on Biomedical Engineering Computational Morphology, BME 25, 99-103.
Sjostrand, F.S. (1958) 3. Ultrastruct. Res. 2, 122.
Stevens, J.K. & Trogadis, J. (1986) Reconstructive 3D electron microscopy: a routine biologic tool. AnaZyt.
Quant. Cytol. Histol. 8, 102-107.
Ware, R.W. & LoPresti, V. (1975) Three-dimensional reconstruction from serial sections. Znt. Rew. Cytol.