ARTICLE
Auteur(s) : Lars
Norlen1,2
1Medical Nobel Institute, Department of Cellular and
Molecular Biology (CMB), Karolinska Institute, Stockholm,
Sweden
2Dermatology Clinic, Karolinska University Hospital,
Stockholm, Sweden
accepté le 5 Decembre 2007
In recent years high-resolution cryo-transmission electron
microscopy (CEMOVIS) and tomography (CETOVIS) of vitreous sections
has been used to visualize the native molecular organisation of
skin [1-5]. Micrographs of vitrified native skin obtained by
CEMOVIS show more detail and sometimes differ dramatically from
those obtained by conventional methods, in which the skin has been
dehydrated and stained. When combined with tomography, molecular
resolution (1.5-5 nm) 3D reconstructions of skin structures may be
obtained [4, 5].
Cryo-electron microscopy of vitreous skin sections
(CEMOVIS)
In conventional electron microscopy (EM) the specimen is dehydrated
during sample preparation, resulting in important loss and
aggregation of biological matter. Further, it is not the biological
structures themselves that are observed, but deposits of stain.
These problems can be overcome by cryo-electron microscopy of
vitreous sections (CEMOVIS), in which a biological specimen is
frozen under high pressure (~2000 bar) by means of ultra rapid
(~20 ms) cooling (below –140 °C). If freezing is fast
enough ice-crystals do not have time to form, and the tissue water
thereby becomes immobilized in a vitreous (i.e., amorphous) state
[6]. The vitrified sample is subsequently cut into ultrathin
(30-80 nm) sections in a cryo-microtome. The vitreous tissue
sections are then directly observed in a cryo-electron microscope.
In this way one can study the ultrastructure of skin in its native
fully hydrated state, without chemical fixation, dehydration or
staining. In fact, micrographs of human skin obtained by CEMOVIS
show not only more biological detail but also differ, sometimes
dramatically, from those obtained by conventional methods [1-3].
The major advantages of CEMOVIS and vitreous cryo-preservation
over conventional electron microscopy and sample preparation of
skin are that there is (i) no loss of tissue material (including
water), that there is (ii) a direct correlation between the optical
density of the recorded image and the local density of the
biological material of the skin sample, that (iii) vitrification of
the living skin is obtained within seconds after biopsy and that
(iv) the total vitrification time is in the millisecond range.
However, there are also drawbacks with CEMOVIS, such as possible
pressure-induced artifacts during vitrification, cutting-induced
tissue compression during sectioning [7], and high sensitivity to
electron beam damage, limiting the total electron dose that can be
applied (smoothing of surface defects starts at
10-30e–/Å2 for ~100 nm thick vitreous
sections [8]). Also, CEMOVIS cannot be applied to samples thicker
than about 200 microns, as sample vitrification otherwise will
fail.
The amount of information that can be extracted from
cryo-electron micrographs of vitreous epidermal sections is largely
limited by superposition of biological matter in the section
thickness dimension. This problem can, however, be circumvented if
CEMOVIS is combined with tomography (CETOVIS), by which a 3D image
of epidermis is reconstructed from a large series of tilted images
recorded with minimum electron dose.
Image contrast in electron microscopy
For high intensity data obtained from strongly scattering objects
such as stained plastic embedded skin samples, image contrast is
principally obtained by reducing the size of the diaphragm aperture
in the focal plane. However, for low intensity data obtained from
weakly-scattering objects such as non-stained fast-frozen skin, the
image contrast is mostly generated by an extra phase shift,
depending on defocus and lens imperfections such as chromatic and
spherical aberration [9].
Electron tomography
3D reconstruction from a tomographic tilt series is usually
performed using a mathematical technique called “weighted
back-projection” [10]. During a conventional weighted
back-projection a reconstructed density in real space is obtained
by integrating the reverse Fourier transforms of the
representations of the tilt images in the focal plane, the
so-called “central sections”. The representation of the image in
the focal plane is the Fourier transformed object multiplied with
the contrast transfer function (CTF) (i.e., the Fourier transformed
point spread function (PSF)). The CTF in the focal plane is mainly
determined by defocus and spherical lens aberration, but is damped
by an envelope function, essentially represented by the Fourier
transform of the aperture. The CTF is, however, further damped and
influenced by chromatic aberration, accelerating voltage and
astigmatism, respectively. The total imaging function is the sum of
the image PSF and the detector shot noise function. It contains all
Fourier transformed CTF envelopes, all pixelation envelopes in the
CCD and shot noise in the CCD.
The Nyquist frequency is the maximal frequency in Fourier space
that can be obtained from a discretely sampled data set of a
certain size (e.g., the number of pixels in the CCD). The inverse
of the Nyquist frequency defines the resolution limit in Real space
(the practical resolution limit is however further reduced by the
CTF of the optical system and the detector “modulation” transfer
function (MTF), i.e., the “modulation” of the CCD signal), which
for CCD sampling corresponds to two times the CCD pixel size.
Sampling in Real space must therefore take place at less than half
of the desired resolution, which is known as the Shannon sampling
theorem [9]. If no extra object symmetry information is available,
the CCD sampling thus sets the maximal resolution for a density
reconstruction.
At small tilt angle increments, which normally is the case
during tomographic sampling, R = pi*D/n, where R is the resolution,
D the section thickness and n the number of equally spaced views
between ± 90 degrees [9].
The alignment transformation (least-square refinement) defines
the geometry of the tilt series in terms of image-tilt, -rotation,
-translation and -scaling (defocus), by comparing the position of
each tilt-image’s fiducial markers to that of a theoretical
zero-tilt [11].
The missing region in a single axis ± 60 degrees tilt series
constitutes a ± 30 degrees “valley” in the Fourier space, which
mathematically can be represented by a “timeglass” function. The
multiplication in Fourier space between the timeglass function and
the tilt series data function corresponds in Real space to a
convolution between the reconstructed density and the reversed
Fourier transform of the timeglass function. The Fourier
transformed timeglass function is oval shaped and stretches in the
z-direction. Thus, the effect of the convolution in Real space is
that every point in the reconstruction is blurred asymmetrically to
stretch out over its nearby points in the z-direction [12].
In single tilt axis reconstructions, the resolution is
consequently different along the three axes. In the y-direction,
i.e., along the tilt axis, the resolution is solely limited by
image resolution (mainly set by defocus and spherical lens
aberration) and image alignment accuracy. In the x-direction the
resolution is also limited by the section thickness and by the
sampling density, i.e., the number of projections (R = pi*D/n, cf.
above). Finally, the resolution in the z-direction, i.e., along the
electron beam direction, is further reduced by the missing region
in the ± 30 degrees tilt range resulting in a ~50% reduced
resolution in the z-direction when compared to the y-direction.
For low-intensity data obtained from weakly scattering objects
such as vitreous skin sections, a large portion of the image signal
corresponds to background noise. Post data refinement methods
(e.g., SIRT, ART and COMET) are sometimes employed to remove such
noise from the 3D reconstructions and deconvolute the CTF. One
example is the regularization method, constrained maximum entropy
tomography (COMET) [13]. COMET produces the most featureless
reconstruction that fits the projection data within their
variances, minimizes the effects of incomplete primary data and
deconvolutes the effects of the CTF. The COMET algorithm ensures a)
that in each iterative step the entropy of a prior estimate (i.e.,
the available, normalized, positive density values of the
3D-reconstruction) is maximized relative to a low-pass filtered,
normalized prior prejudice density distribution, b) that the
normalization of the prior estimate is maintained, and c) that the
c2 statistic (i.e., the sum of the squared differences
between the projected 3D reconstruction and the corresponding
observed projections, divided by the variances of the observations)
is equal to 1. It is consequently crucial for COMET post data
refinement to possess good estimates of the variances of the
observed raw data. However, as CCD cameras have Poisson-distributed
counting statistics the variances are approximately proportional to
the observed optical densities. A COMET reconstruction has the
property of being maximally non-committal with respect to missing
information, e.g., effects caused by missing projections. Further,
by virtue of producing the least information containing
reconstruction that fits within the variances, COMET refined
reconstructions cannot be over-interpreted. COMET can be classified
as a tomographic reconstruction method as it utilises Real space
projections during the iterative refinement.
Exploring skin structure using cryo-electron microscopy
(CEMOVIS) and tomography (CETOVIS)
Examples of skin structures that have been visualized with higher
accuracy and resolution with CEMOVIS (~2 nm resolution) than with
conventional methods are desmosomes, keratin filaments,
keratohyaline granules, lamellar bodies, extracellular lipid
membranes and corneocyte lipid envelopes [2, 3]. Using CETOVIS, 3D
reconstructions of desmosomes, keratin filaments, “lamellar bodies”
and mitochondria have so far been obtained at ~5 nm resolution [4,
14, 15].
Figure 1
compares a CEMOVIS micrograph of an epidermal desmosome (A) with a
conventional electron micrograph (B). Figure 1A represents a
native, non-stained cryo-fixed skin sample while a conventionally
dehydrated and stained skin sample is visualized in (figure 1B). The plasma
membranes appear as low-density lines in (B) because the stain
poorly penetrates the lipid hydrocarbon chain tail regions (B, thin
white arrow). In contrast, the membranes are visualized as ~4 nm
wide high-density bilayers in the vitreous non-stained skin sample
(A). In the cytoplasm, a ~7 nm thick electron dense layer situated
~11 nm from the plasma membrane (A, thick white arrow), can be
seen. In contrast, this cytoplasmic layer is not resolved in the
plastic-embedded sample (B, thick white arrow). Further, in the
extracellular space, the cadherin molecules connecting the two
keratinocytes are poorly resolved and seem irregularly arranged
[16] (B, black arrow). However, in the vitreous skin sample (A) it
is revealed that the cadherins are ordered and possess a ~5 nm
periodicity (A, black arrow). Apart from the staining, the apparent
difference between the two images is also largely due to
aggregation of biological matter during specimen preparation in (B)
and the absence of the same in (A).
Figure 2
shows a tomographic reconstruction of an epidermal desmosome using
CETOVIS. Individual cadherin cell adhesion molecules are visualized
in 3D at ~5 nm resolution. COMET regularization gives smoother and
more continuous cadherin structures [14].
Figure 3
shows a CEMOVIS micrograph of keratin filaments in corneocytes.
Native keratin filaments appear as ~7.8 nm wide structures with a
centre-to-centre distance of ~16 nm, embedded in a comparatively
electron lucent matrix [2].
Figure 4
shows a 3D reconstruction of keratin filaments in stratum
granulosum using CETOVIS [4]. COMET regularization gives smoother
and more continuous 3D reconstructions of individual keratin
filaments (figure
5), when compared to conventional backprojection (figure 4D) [15].
figures 6 and
7 show CEMOVIS micrographs of the stratum corneum
extracellular lipid matrix. Electron dense lines correspond to
lipid headgroups.
From figures 1-7 it is
evident that important cell adhesion structures (like desmosomes),
cytoskeletal protein components (like keratin intermediate
filaments) and lipid membrane structures (like the stratum corneum
extracellular lipid matrix) are all better preserved, as well as
better visualized, with CEMOVIS than with conventional electron
microscopy. This allows for gain of new information, giving rise to
new morphological interpretations.
Present challenges for CETOVIS
Sample compression during cutting of vitreous tissue sections
constitutes, in our experience, the major drawback with CEMOVIS and
CETOVIS. Sample compression may affect the integrity of the
biological structures contained in the cryo-preserved skin, and
thereby complicate interpretations of the corresponding 3D
reconstructions. During tomographic 3D reconstruction it is
therefore mandatory to compare biological structures situated in
different orientations relative to the sectioning direction.
Optimising tilt series image alignment also constitutes a
challenge. Cross-correlation [17] and deposition of fiducial
markers onto the electron microscopy grid support film [18] have
classically been used for tomographic image alignment. However,
cross-correlation is unreliable for high-resolution work due to the
low signal-to-noise ratio of cryo-electron micrographs obtained
close to focus in the low-dose mode. Image alignment with the aid
of fiducial markers deposited on the support film is also
unpractical. This is because there is usually a substantial gap
between the vitreous cryo-section and the support film. Therefore,
only those parts of the section that happen to be in contact with
the support film can be aligned and reconstructed. However, a new
method for the deposition of fiducial markers directly onto the
vitreous sections was recently developed [4]. Using this method,
all parts of the section can potentially be aligned and
reconstructed. However, cryo-sections contain cutting artifacts,
such as crevasses, which may move during electron beam exposure,
causing marker drift. Generally, thinner sections are less prone to
cutting artefacts [7]. Thus, thinner sections may aid to increase
precision in tomographic image alignment. Also from the perspective
of the general tomographic theory, thinner sections are more
suitable for high-resolution work [9].
Another problem is gaps between the vitreous section and the
supporting electron microscopy grid due to section waviness. This
may cause section instability in the electron beam during data
collection.
A drawback of CETOVIS is a low signal-to-noise ratio due to the
low electron doses used. The tilt series are therefore often
collected at an underfocus of several micrometers to gain image
contrast. This reduces high frequency (high resolution) information
in the images. The choice of imaging conditions for tomography is
thus a compromise between contrast and resolution. Consequently,
the application of post-data refinement methodologies including
noise reduction and CTF deconvolution is necessary for
high-resolution cryo-electron tomographic analysis.
A resolution of 5 to 7 nm in tomographic reconstructions of
vitreous skin sections has been reported without post-data
refinement [4]. This is consistent with that expected from the
geometry of the data collection. This resolution is not high enough
to resolve fine details of skin molecules, such as the
subfilamentous 3D organization of keratin. However, it is most
probably possible to obtain a resolution below 5 nm by minimizing
section cutting artifacts and by optimizing 3D reconstruction
algorithms, including image alignment and post-data refinement
procedures.
Recent practical developments
Dual tilt axis tomography to increase 3D resolution
The reduced resolution along the x-direction with respect to the
y-direction (i.e., along the tilt axis) in tomographic 3D
reconstructions can be overcome if the specimen during data
acquisition is tilted around two orthogonal axes instead of around
a single tilt axis. Thus, unlike a single-axis tomogram, a
dual-axis tomogram shows good resolution for extended features at
any orientation in the plane of the specimen and, in fact, has also
improved resolution in the z-direction [19]. The difficulties
encountered when rotating the specimen at cryo-conditions inside
the electron microscope has until recently hampered dual-axis
CETOVIS. However, specimen cryo-holders with grid-rotating
capability are now under way (personal communication with Dr Jürgen
Plitzko, Max-Planck Institute, Martinsried, Germany).
Static charging/discharging deposition of vitreous sections
onto electron microscopy grids to increase section flattening
The non-flat nature of vitreous skin sections causes poor
attachment of the sections on the electron microscopy grids, and
can thereby cause section instabilities under the electron beam. It
further reduces the number of section areas suitable for data
collection due to “shadowing” at high tilt-angles. For this reason,
dual-axis CETOVIS will remain impractical until new techniques for
flattening and attachment of vitreous sections onto electron
microscopy grids have been developed. In fact, Simco©
(Amsterdam, Holland) has recently produced a static
charging/discharging device that promises to eliminate these
problems (personal communication with Professor Peter Peters,
Amsterdam University, Amsterdam, Holland).
Surface treatment of diamond knives and high-speed
sectioning
Section compression during cutting remains the main problem of
CEMOVIS and CETOVIS. A new surface film treatment of the diamond
knives reducing friction during sectioning has, however, reduced
section compression significantly (personal communication with Dr.
Helmuth Gnaegi, Diatome, Bern, Switzerland). Another important
measure to reduce friction during sectioning is to clean the
diamond knives after each usage to avoid lipid-contamination. Also,
sectioning of vitrified skin samples has traditionally been
performed at speeds below 1 mm/second. It has, however,
recently been proposed that high-speed sectioning
(> 50 mm/second) can reduce compression of the
vitreous sections (personal communication with Dr. Helmuth Gnaegi,
Diatome, Bern, Switzerland).
Increased gas-flow to reduces ice crystal contamination
Ice crystal contamination is often a practical problem during
handling of vitreous skin sections. Increased nitrogen gas flow
from 3.5 litres/hour (Leica cryo-microtome, standard) to 6
litres/per hour diminishes ice crystal contamination significantly.
If combined with a protective chamber wrapped around the
cryo-microtome, ice contamination can be almost completely avoided
(personal communication with Dr Ian Lambswood, Leica microsystems,
Wetzlar, Germany).
Gold nanoparticle alignment instead of quantum dots
Deposition of lead sulfide (PbS) quantum dots onto vitreous skin
sections revolutionized high-resolution CETOVIS [4]. PbS quantum
dots are, however, less electron dense than gold. Due to the
extreme low-dose conditions used during CETOVIS data collection,
higher fiducial marker electron density will contribute to decrease
the alignment error. Recently, an alignment error of 0.7 nm was
reported using oleylamine-capped gold nanoparticles as fiducial
markers instead of quantum dots (personal communication with
Manuela Gruska, Max-Planck Institute, Martinsried, Germany).
Conclusion
Basic science is essential to Dermatology [20]. The lack of a
molecular understanding, at the nanometer level, of skin biology
represents one major obstacle towards the advancement of basic skin
research. New methodologies have, however, recently been developed
for the visualization of native skin at nanometer resolution in
situ. Presently, ~2 nm 2D resolution and ~5 nm 3D resolution is
regularly achieved. In fast-frozen vitreous skin samples the native
biomolecular organization may be preserved down to atomic
resolution. CEMOVIS and CETOVIS are consequently ideally suited for
molecular skin research. Freeze-substitution electron microscopy,
with or without immunocytochemistry, may, however, represent a
powerful complement for supramolecular skin research.
Acknowledgements
The present work was made possible by the generous support from the
Wenner-Gren Foundations and the Edward Welander Foundation.
References
1 Norlén L, Al-Amoudi A. Dubochet J. A cryo-transmission
electron microscopy study of skin barrier formation. J Invest
Dermatol 2003; 120: 555-60.
2 Norlén L, Al-Amoudi A. Stratum corneum keratin
structure, function, and formation: the cubic rod-packing and
membrane templating model. J Invest Dermatol 2004; 123(4):
715-32.
3 Al-Amoudi A, Dubochet J, Norlén L.
Nanostructure of the epidermal extracellular space as observed by
cryo-electron microscopy of vitreous sections of human skin. J
Invest Dermatol 2005; 124: 764-77.
4 Masich S, Östberg T, Norlén L,
Shupliakov O, Daneholt B. A procedure to deposit fiducial
markers on vitreous cryo-sections for cellular tomography. J Struct
Biol 2006; 156: 461-8.
5 Norlén L, Masich S, Goldie KN, Hoenger A.
Structural analysis of vimentin and keratin intermediate filaments
by cryo-electron tomography. Exp Cell Res 2007; 313: 2217-27.
6 Dubochet J, Adrian M, Chang J-J, Homo J-C,
Lepault J, McDowall AW, Schultz P. Cryo electron
microscopy of vitrified specimens. Q Rev Biophys 1988; 21(2):
129-228.
7 Al-Amoudi A, Studer S, Dubochet J. Cutting
artifacts and cutting process in vitreous sections for
cryo-electron microscopy. J Struct Biol 2005; 150: 109-21.
8 Hsieh C, Marko M, Frank J, Manella CA.
Electron tomographic analysis of frozen-hydrated tissue sections. J
Struct Biol 2002; 138: 63-73.
9 Klug A. Image analysis and reconstruction in the electron
microscopy of biological macromolecules. Chem Scr 1979; 14:
245-56.
10 Radermacher M. Weighted back-projection methods. In:
Frank J, ed. Electron tomography – three-dimensional imaging
with the transmission electron microscope. Plenum Press, NY, 1992;
(Ch 5).
11 Lawrence MC. Least-square method of alignment using
markers. In: Frank J, ed. Electron tomography –
three-dimensional imaging with the transmission electron
microscope. Plenum Press, NY, 1992; (Ch 8).
12 Carazo JM. The fidelity of 3D reconstructions from
incomplete data and the use of restoration methods. In:
Frank J, ed. Electron tomography – three-dimensional imaging
with the transmission electron microscope. Plenum Press, NY, 1992;
(Ch 6).
13 Skoglund U, Öfverstedt L-G, Burnett R,
Bricogne G. Maximum entropy three-dimensional reconstruction
with deconvolution of the contrast transfer function: a test
application with adenovirus. J Struct Biol 1996; 117: 173-88.
14 Östeberg T, Masich S, Norlén L,
Daneholt B. Tomographic analysis of epidermal desmosomes using
vitreous cryo-sections. In: Marks R, Matts P,
Lévêque J-L, eds. Stratum corneum. Stratum Corneum Group,
Cardiff. 2007.
15 Nosek D, Masich S, Norlén L. Tomographic
analysis of epidermal keratin filaments using vitreous
cryo-sections. In: Marks R, Matts P, Lévêque J-L,
eds. Stratum corneum. Stratum Corneum Group, Cardiff. 2007.
16 He W, Cowin P, Stokes DL. Untangling
desmosomal knots with electron tomography. Science 2003; 302:
109-13.
17 Winkler H, Taylor KA. Accurate marker-free
alignment with simultaneous geometry determination and
reconstruction of tilt series in electron tomography.
Ultramicroscopy 2006; 106: 240-54.
18 Hsieh C, Leith A, Mannella CA, Frank J,
Marko M. Towards high-resolution three-dimensional imaging of
native mammalian tissue: Electron tomography of frozen-hydrated rat
liver sections. J Struct Biol 2006; 153: 1-13.
19 Mastronarde DN. Dual-axis tomography: an approach with
alignment methods that preserve resolution. J Struct Biol 1997;
120: 343-52.
20 Schwarz T. Dermatology and basic science. Eur J Dermatol
2007; 17: 182-4.
21 McKee PH. Pathology of the skin with clinical
correlations. London: Harper and Row, Gower Medical Publishing,
1990; (15).
|
Version imprimable
Version PDF
