Soft X-ray microscopy with a cryo scanning transmission X

Soft X-ray microscopy with a cryo scanning transmissionX-ray microscope: II. Tomography

Y. WANG 1, C. JACOBSEN, J. MASER1 & A. OSANNA

Department of Physics and Astronomy, State University of New York at Stony Brook, Stony Brook,

NY 11794-3800, U.S.A.

Key words. frozen hydrated, soft X-ray microscopy, synchrotron radiation,

tomography, zone plates.

Summary

Using a cryo scanning transmission X-ray microscope

(Maser, et al. (2000) Soft X-ray microscopy with a cryo

scanning transmission X-ray microscope: I. Instrumenta-

tion, imaging and spectroscopy. J. Microsc. 197, 68±79), we

have obtained tomographic data-sets of frozen hydrated

mouse 3T3 ®broblasts. The ice thickess was several

micrometres throughout the reconstruction volume, pre-

cluding cryo electron tomography. Projections were

acquired within the depth of focus of the focusing optics,

and the three-dimensional reconstruction was obtained

using an algebraic reconstruction technique. In this ®rst

demonstration, 100 nm lateral and 250 nm longitudinal

resolution was obtained in images of unlabelled cells, with

potential for substantial further gains in resolution. Future

efforts towards tomography of spectroscopically highlighted

subcellular components in whole cells are discussed.

Introduction

Three-dimensional imaging plays a growing role in

investigations of biological structure, where it is often

important to understand the relationship between compo-

nents in a cell that may be distributed in depth. Ideally this

is done on a specimen which is minimally modi®ed from its

living state. We describe here the use of soft X-ray

microscopy to record tomographic data sets of eukaryotic

cells which have been removed from cell culture medium

and rapidly plunge-frozen with no intermediate preparation

steps except for removal of medium in excess of about

10 mm in thickness. Specimens of this thickness are likely to

have many overlapping structures, so some form of 3D

imaging becomes highly desirable to resolve the complexity.

Good contrast of many subcellular structures is obtained

without labelling, at 100 nm transverse and 250 nm

longitudinal resolution in this ®rst demonstration. It is

feasible to improve the resolution to 50 nm or better in all

three dimensions, and it is also possible to highlight speci®c

structures based on their intrinsic chemistry using X-ray

near-absorption-edge resonances.

Soft X-ray tomography has been demonstrated at sub-

100 nm resolution on microfabricated test objects Haddad

et al. (1994), and on dry, mineralized sheaths of soil

bacteria (Lehr, 1997). However, we are not aware of

previous demonstrations of soft X-ray tomography of

hydrated specimens. The use of X-ray near-absorption-

edge resonances is well established in studies of dehydrated

(Ade et al., 1992; Zhang et al., 1996; Buckley et al., 1997)

and some hydrated (NeuhaÈusler et al., 1999) specimens and

efforts towards spectromicroscopy of frozen hydrated speci-

mens are underway (Maser et al., 2000).

Optical microscopy is the most widely used method for

3D imaging of cells. Using approaches such as confocal and

two-photon confocal microscopy (Hell et al., 1997),

through-focus deconvolution methods (Agard & Sedat,

1983; Carrington et al., 1995), or interferometric methods

(Gustafsson et al., 1999), images of eukaryotic cells have

been obtained with a resolution as high as 100 nm in all

three dimensions. Living specimens have been examined in

some cases, although it is more common to study

chemically ®xed specimens. An element to all of these

approaches is the use of ¯uorescent labels because the

resulting incoherent emission process is key to the method.

This is frequently of great advantage in selectively imaging

identi®able subcellular components, and sometimes multi-

ple ¯uorophores can be used simultaneously. However, it is

not generally possible to also image the unlabelled features

in 3D at the same resolution.

Electron microscopy has also been used with great

success for 3D imaging of biological structures. For speci-

mens which have been ®xed, embedded and dried, serial

sections can be cut and their images assembled to obtain a

Journal of Microscopy, Vol. 197, Pt 1, January 2000, pp. 80±93.

Received 10 February 1999; accepted 30 May 1999

q 2000 The Royal Microscopical Society80

Correspondence to: Chris Jacobsen.1Present address: Advanced Photon Source, Argonne National Laboratory,

Argonne, IL 60439, U.S.A.

3D reconstruction, or thick single sections can be rotated to

collect a tomographic data set. A more recent development

is the use of cryo methods to allow tomography with rapidly

plunge-frozen prokaryotic cells at 10 nm resolution (Grimm

et al., 1998). This method avoids the signi®cant biochemical

(Coetzee & van der Merwe, 1984) and structural (O'Toole

et al., 1993) changes associated with the usual preparation

techniques for electron microscopy of dehydrated speci-

mens. However, electron tomography of frozen hydrated

specimens is ultimately limited by the electron mean free

path for inelastic scattering in ice (even if energy ®ltering is

used (Grimm et al., 1998; Jacobsen et al., 1998)), such that

thicknesses greater than about 400 nm at 120 keV or 1 mm

at 400 keV become exceedingly challenging.

We therefore believe that X-ray microscopy complements

the capabilities of other approaches for 3D imaging of cells.

Because of radiation damage (Bennett et al., 1993; Williams

et al., 1993; Kirz et al., 1995), we believe it will not be

possible for soft X-ray microscopy to match the capabilities

of optical microscopy to study living specimens. However,

the resolution in this particular demonstration of soft X-ray

tomography is comparable to that of 3D optical microscopy,

and can be expected to improve towards the 30 nm level, as

demonstrated in other X-ray microscopy experiments

(Spector et al., 1997; Schneider, 1998). Labelling methods

are only in their infancy in X-ray microscopy (Jacobsen et al.,

1993; Chapman et al., 1996; Moronne, 1999), yet it is

useful in many cases to be able to observe unlabelled

structures in 3D as well (especially to understand the 3D

relationships between different structures in a cell). Finally,

500 eV soft X-rays are able to penetrate through 10 mm

layers of ice with good contrast for small features, whereas

electron microscopy of unsectioned frozen hydrated eukaryotic

cells faces severe challenges because of specimen thickness.

Tomographic data recording with a cryo scanningtransmission X-ray microscope

Soft X-rays have properties which make them well suited for

tomographic imaging of cells. By operating at a photon

energy # 540 eV (the oxygen K absorption edge), the 1/e

penetration distance in water/ice is about 9 mm. At the same

time, for high resolution tomography one would like to have

strong contrast for thin organic structures. This is

accomplished by operating at an energy above 284 eV (the

carbon K absorption edge), where , 5 nm thick layers of

protein can be expected to give about 1% absorption

contrast. Because X-ray interactions are dominated by

photoelectric absorption with very little atomic elastic or

inelastic scattering (Kirz et al., 1995; Hubbell et al., 1980),

multiple scattering effects are negligible and quantitative

image interpretation is simpli®ed.

Strong absorption of ionizing radiation implies high

radiation dose; the radiation dose delivered to a hydrated

biological specimen in the course of acquiring a single

50 nm resolution image with a signal-to-noise ratio of 5 : 1

is about 100 Mrad or 106 Gray (Sayre et al., 1977). Even if

one were to lower the dose (e.g. by using phase contrast,

which can lower the dose by as much as an order of

magnitude (Schmahl et al., 1994)), radiation damage would

lead to changes in the specimen during the collection of a

3D data set unless one were to consider schemes involving

multiple beams from a single, fast-pulse X-ray source

(Howells et al., 1994). This is true both of living specimens,

and also of chemically ®xed specimens at room temperature.

One solution is to use the cryo methods developed in the

electron microscopy community (Taylor & Glaeser, 1974;

Dubochet et al., 1988), where the sample is rapidly frozen to

liquid nitrogen temperatures so as to obtain ice in an

amorphous state and thereby avoid ice crystal artefacts.

Using the cryo technique, it has been shown that frozen

hydrated cells can be imaged with no apparent change to

the specimen at the ,50 nm resolution level following doses

as high as 1010 Gray (Schneider et al., 1995; Maser et al.,

1998; Schneider & Niemann, 1998; Maser et al., 2000).

Therefore, the specimen appears stable to doses of about 104

times the required single-image dose, which is suf®cient for

tomography. It should be noted that the exposure require-

ment for tomography is in principle only modestly higher

than that for 2D imaging (Hegerl & Hoppe, 1976; McEwen

et al., 1995) in order to achieve the same signal-to-noise

ratio.

X-ray focusing

The highest resolution optics currently available for soft X-

ray focusing are Fresnel zone plates and the highest

resolution zone plates at present have been fabricated by

electron beam lithography (Schneider et al., 1995; Spector

et al., 1997). With suf®cient care in electron beam

lithography, zone plates can be fabricated with negligible

aberrations. In a scanning transmission X-ray microscope,

one seeks to illuminate the zone plate with suf®cient

coherence so that a diffraction limited probe is formed with

X-rays of wavelength l. For zone plates with more than

about 100 zones (Michette, 1986), the light amplitude is

distributed according to the well known Airy pattern

A�2J1(n)/n, where n�1´22p is the location of the ®rst

amplitude zero-crossing at a radius of (1´22/2)l/NA from

the optical axis, with NA denoting the numerical aperture.

One then has a transverse resolution dt of

dt � 0:61l

NA�1�

and a longitudinal resolution (Born & Wolf, 1980) of

dl � 1:22l

NA2: �2�

q 2000 The Royal Microscopical Society, Journal of Microscopy, 197, 80±93

SOFT X-RAY MICROSCOPY WITH A CRYO STXM 81

The numerical aperture of a Fresnel zone plate is

determined by diffraction from the outermost zones which

have a half-period or width of drN, giving NA� l/(2drN). We

can therefore write

dt � 1:22drN�3�

dl � 4:88drN

drN

l� 4dt

drN

l: �4�

The zone plate used in the experiments reported here had

an outermost zone width drN �60 nm and was used at a

wavelength of l�2´4 nm; using Rayleigh criteria its

theoretical transverse resolution is dt �73 nm, and its

longitudinal resolution is dl �7´3 mm. We therefore see

that the longitudinal resolution is poor compared to the

transverse resolution, so that with X-ray microscopy it is

not favourable to attempt optical sectioning or through-

focus deconvolution as is done in visible light microscopy.

Instead, poor longitudinal resolution translates into large

depth of focus, so that a better approach for 3D imaging is

to rotate the specimen through the focus and acquire

projection images such as is needed for tomography

(Haddad et al., 1994; Lehr, 1997).

How much is the resolution spoiled at the edges of the

depth of focus? The Airy pattern describes the amplitude at

the plane of focus; because a coherent probe is formed, one

can calculate the amplitude at other planes by propagating

the Airy pattern A(r, 0). Free-space propagation of a

wave®eld with wavelength l by a distance z involves

convolution with a kernel h� exp[±ipr2/(lz)] (Goodman,

1968). This can be written as

A�r; z� � A�r;0� � h �r; z�

� Bÿ1�B �A�B�h�� 5�

where * means convolution, and B is a Fourier±Bessel

transform (the cylindrical coordinates equivalent of a 2D

Fourier transform). Because this calculation involves a

simple integration over one variable, it can be rapidly

calculated with a very ®ne grid spacing.

For a scanning transmission microscope with large

detector aperture and a specimen with no phase contrast,

the image is formed by a convolution of the probe intensity

jA(r,z)j2 with the specimen absorption distribution. One can

then calculate the modulation transfer function MTF from

the Fourier transform of the probe intensity at various

distances z (Fig. 1 left). We can also calculate the image of a

simple absorption object by convolution with the probe

intensity at various distances z (Fig. 1 right). From these

simulations we see that we can obtain good contrast for

high resolution features throughout a longitudinal distance

of 2dl, or 14 mm for the example considered here. These

®gures can be scaled to other choices of outermost zone

Fig. 1. High resolution projections of specimens several micrometres thick can be obtained in a soft X-ray microscope. We show here the

optical performance versus defocus calculated at a wavelength l�2´5 nm for a drN �45 nm zone plate with a central stop over 44% of

its diameter. (Such a zone plate has a theoretical Rayleigh resolution of dt �55 nm, a spatial frequency cutoff of 1/(2 ´ 22´5 nm) or

22 mmÿ1.) On the left is shown a shaded contour plot of the modulation transfer function versus spatial frequency and defocus; the MTF

applies only to a purely absorptive object. On the right is shown a simulated defocus image series of bars of width 33, 71, and 148 nm.

One-dimensional lines were extracted from two-dimensional calculations at each defocus image to make up the representation shown.

These calculations can be scaled to other choices of X-ray wavelength l and zone plate outermost zone width drN by scaling transverse dimen-

sions by drN, and longitudinal dimensions by drN2/l, and a depth of focus of 2dl �7´9 mm.

82 Y. WANG ET AL.



Fig. 2. An experimental demonstration of the depth of ®eld in our cryo STXM. This image shows not the usual x-y representation of the

specimen but rather an x-z representation; that is, the horizontal scan direction is in the object's plane as usual, while the vertical scan direc-

tion represents defocus. The object was a microfabricated test pattern with a variable grating pattern. The depth of ®eld for this image with a

drN �45 nm zone plate at a wavelength of l�2´4 nm is consistent with the calculated value of 2dl �8´2 mm.

Fig. 3. A simpli®ed schematic of the optical system of the microscope. Soft X-rays from an undulator at the National Synchrotron Light

Source are delivered by a grating monochromator in the X1A beamline (Winn et al., 1996) with horizontal and vertical foci coinciding

at the exit slit. This beam then coherently illuminates a Fresnel zone plate objective lens to produce a high resolution focus (isolated

using an OSA) through which the specimen is scanned; the transmitted ¯ux is detected by a phosphor/photomultiplier detector.


width drN and wavelength l by scaling the transverse spatial

frequencies or distances with drN, and the longitudinal

distances with drN2/l.

The above considerations can be veri®ed experimentally

by taking images where the horizontal image direction

corresponds to a transverse position, and the vertical image

direction corresponds to a longitudinal position. In images

of this type, a pinhole gives an hourglass-type appearance

as the in-focus plane is located at the `neck' where the

image is the narrowest and sharpest. In practice, it is

usually straightforward to determine the proper focus and

estimate the depth of focus from such images. An image of

this type acquired using a drN�45 nm zone plate at

l�2´4 nm is shown in Fig. 2, which shows a depth of

focus of about 8 mm when Eq. (4) predicts 8´2 mm.

Specimen tilt in a cryo scanning transmission X-raymicroscope

We have recorded tomographic data using a cryo scanning

transmission X-ray microscope (cryo STXM). The micro-

scope is built around transmission electron microscope type

cryo specimen holders (E. A. Fischione, Inc. and Oxford

Instruments) which are scanned by a combination of piezo

transducers and stepping motors for high resolution and

large range (Maser et al., 2000). The specimen holder is

inserted into an airlock for rapid pumpout, and then

inserted into the microscope chamber which is maintained

at a vacuum of 10ÿ6 to 10ÿ7 Torr. The airlock can be driven

manually to rotate the specimen holder about a single tilt

axis, and a dial is used to indicate the tilt angle to a

precision of about 0´58.

A spatially coherent, tunable X-ray beam with an

adjustable energy resolving power of E/(DE)�1000±

5000 is focused by a Fresnel zone plate to a point focus

through which the specimen is scanned (see Fig. 3). For the

experiments reported here, a Fresnel zone plate with a

diameter of d�160 mm and an outermost zone width of

drN�60 nm was used. The zone plate had a 80 mm diameter

central stop, so that to isolate the focused light in the ®rst

diffractive order we placed a 70 mm pinhole as an order

sorting aperture (OSA) at a distance of 0´6 focal lengths from

the zone plate. The OSA was fabricated on a 4 ´ 0´5 mm

aperture mount orientated with its long axis parallel to the

specimen holder, thereby minimizing interference at high tilt

angles. For the experiments reported here, the X-ray energy

was 516 eV so that the focal length of the zone plate was

f�4´00 mm. A schematic of how the specimen was able to

tilt about the zone plate is shown in Fig. 4. Further work on

this general approach with appropriately modi®ed cryo

specimen holders should allow us to obtain a tilt range of

6 608 with zone plate focal lengths as short as 1´5 mm.

Specimen preparation for cryo tomography

For our ®rst demonstration experiments of tomography in

a cryo STXM, we used mouse 3T3 ®broblasts. They were

grown directly on 200 mesh gold electron microscope

grids with formvar support ®lm (EMS). The cells were

cultured in a medium consisting of 90% Dulbecco modi®ed

Eagle medium, 9% Hyclone bovine serum, and 1%

antibiotics; the cultures were maintained at 37 8C in a

saturated humidity atmosphere with 5% CO2. The cells

typically reached a near-con¯uent state 2±3 days after

being plated, although coating the EM grids with

®bronectin or poly-L-lysine improved cell attachment and

marginally reduced the growth time. When the cell density

on a grid reached a sub-con¯uent level, the grids were

removed from the culture medium with a ®ne-point forceps

and the forceps were mounted on a plunge-freezing

Fig. 4. Geometry of the zone plate, OSA, and sample when the sample holder is tilted by 458. A thin strip OSA is used so that the sample

holder can tilt around it, thereby increasing the angular tilt range.

84 Y. WANG ET AL.


workstation. The excess culture medium on the grid was

carefully blotted by draining from the edges with ®lter

paper. The forceps holding the grid were then propelled by a

preloaded spring into liquid ethane at about 3 m sÿ1 entry

speed, following a procedure known to produce vitri®ed

specimens for cryo electron microscopy (Dubochet et al.,

1988; Echlin, 1992). The frozen grid was then mounted in

the cryo specimen holder and maintained in a cryogenic

state for imaging. Note that in this procedure, the cells were

not intentionally modi®ed until they were rapidly frozen so

that the frozen state should resemble near-physiological

conditions. After a sample grid was loaded into the

specimen holder, the holder was inserted into the micro-

scope for imaging. Images could be taken within 15 min of

cell freezing. A rapid evaluation of the three-dimensional

character of a specimen could be obtained by acquiring and

examining a stereo pair, such as is shown in Fig. 5.

Image acquisition and alignment

To acquire a tomographic dataset, we ®rst took an image at

08 tilt, followed by images from ÿ458 to �45 tilt at 58

increments, and then tilt angles beyond 6 458. This

procedure was used so that a full data set with restricted

tilt range could be obtained before running the risk of

having the sample holder bump into optics at larger tilt

angles. At each tilt angle, the centring and focus of the

image was evaluated by the microscope operator and

adjusted manually prior to imaging. A ®eld size of

400 ´ 300 pixels of 100 nm size was chosen to allow the

entire cell to be contained within a subsequently cropped

350 ´ 256 pixel ®eld (Nyquist sampling at the spatial

frequency cutoff of the drN �60 nm zone plate used would

require 15 nm step size, but such a data set would

overwhelm the computational facilities we had available).

This sampling allowed us to use well optimized fast Fourier

transform algorithms on 256 pixel slices, and a

350 ´ 256 ´ 256 pixel volume (involving 92 Mb of 32-bit

¯oating point numbers) could be accommodated within our

available computational facilities. For these ®rst experi-

ments, we used a per-pixel dwell time of 20 msec, so that

acquiring each image took approximately half an hour.

Collection of the entire data set took over 24 h, although we

expect this time to be reduced in the future.

The acquisition procedure described above yielded a set of

24 images to be processed (Fig. 6). Because each image was

manually aligned during data acquisition, there was a

considerable shift in position from one image to the next.

We therefore identi®ed a single, highly absorbing, 2-mm-

diameter vesicle which was clearly visible near the vertical

centre of all images and used it as an alignment marker. The

sub-region about this vesicle was selected from each image

A, and the relative shift of an image Ai�1 compared to its

predecessor Ai was calculated by determining the location of


Fig. 5. A stereo pair of a chemically ®xed and frozen hydrated mouse 3T3 cell at the M-phase of its division. The ®xation was carried out with

3´7% paraformadehyde at room temperature for 15 min prior to plunge-freezing and imaging. Each image consists of 300 ´ 400 pixels at

100 nm step size and the angular separation between the two images is 208.


the peak of the cross correlation function F ÿ1|( FA i�1)

( FAi)*|, where F refers to a fast Fourier transform and *

denotes the complex conjugate. This shift was subsequently

corrected, yielding a set of images with a tilt axis de®ned to

pass through the chosen vesicle. Using this method, two

consecutive images can be aligned to 1 pixel, or 100 nm

accuracy (we also plan to use gold beads for alignment

assistance in the future when the complexity of the specimen

structure is increased). For tomographic reconstruction, the

images should represent linear maps of the total overlapping

material in each projection. The Lambert±Beer law of X-ray

absorption by a ®lm of thickness t can be written as

I � I0 exp ÿ2lre

NA

Af2� rt�

� �; �6�

where l is the wavelength, re �2´818 ´ 10ÿ15 m is the

classical radius of the electron, NA �6´02 ´ 1023 is Avoga-

dro's number, A is the atomic weight, f2 is an X-ray optical

coef®cient which is tabulated for all elements (Henke et al.,

1993), and r is the density. If the density varies as r(z) along

a distance z, the net transmission is given by

I � I0 exp ÿ2lre

NA

Af2

� z

0r�z 0

�dz 0� �

: �7�

We therefore see that to obtain a linear mapping propor-

tional to the total material�

z0 r(z0) dz0 in a projection column

such as is required for tomography, we must take the

logarithm of the normalized data log[I/I0]. Projections of

log-normalized data were then cropped about the reference

marker to an array size of 350 slices of 256 pixels about the

rotation axis.

Fig. 6. A series of 24 images of a frozen

hydrated mouse 3T3 ®broblast obtained

over a tilt range between ÿ558 to �608

in 58 tilt intervals. Each image contains

400 ´ 300 pixels at 100 nm step size.

86 Y. WANG ET AL.


Tomographic reconstruction of the cell dataset

Reconstruction has been attempted using both ®ltered

backprojection (FBP) (Bracewell & Riddle, 1967) and

algebraic reconstruction techniques (ART) (Kaczmarz,

1937; Tanabe, 1971). Figure 7 shows a comparison between

reconstructions with the two methods using simulated data

projected from a Shepp±Logan (Shepp & Logan, 1974)

phantom. With the type of limited tilt data acquired in our

experiment, reconstructions using ART show higher contrast

and better ®delity to the original data, especially in the

direction of missing projection data. In future work, we

expect to be able to reduce the data acquisition time and take

advantage of dose fractionation in data recording (Hegerl &

Hoppe, 1976; McEwen et al., 1995). In this case, the

improved signal-to-noise ratio shown by ART for limited

photon statistics data will also become useful (see Fig. 7).

The computing time cost of using ART rather than FPB is

increased from about 30 min to about 10 h (using a 200-

MHz Pentium Pro processor) on our 24 projections of

350 ´ 256 pixels when 25 ART iterations are computed.

Convergence in ART is reached typically after 15±20

iterations with our data set.

The ART reconstruction of the 3T3 ®broblast projection

data provides a three-dimensional view of the map of

absorption coef®cient times density f2r in the frozen hydrated

cell. Our convention is that in the 08 projection, the X-ray

beam travels towards the viewer in the�z direction, and that

the horizontal and vertical axes in a 08 projection are in


Fig. 7. Reconstruction using simulated pro-

jections from a Shepp±Logan phantom

(Shepp & Logan, 1974) indicate that alge-

braic reconstruction techniques (ART)

give better reconstructions than ®ltered

backprojection (FBP) on projection data

sets with the characteristics described

here. At the top are shown reconstructions

using 25 projections in 58 intervals over a

ÿ608 to �608 angular range, similar to

that of Fig. 6. The results from the ®ltered

backprojection method show slightly

higher resolution in the directions with

®lled Fourier components, while ART repro-

duce features in the un®led Fourier space

better and show higher overall contrast.

At the bottom are shown reconstructions

from simulated projection data with noise

added using ®ltered backprojection and

ART, respectively. The added Gaussian

noise was signal independent with a peak-

to-peak signal-to-noise ratio of 20 : 1. In

this case, 37 simulated projections were

used with 58 increments over a ÿ908 to

�908 range. The ART reconstructions

were calculated with 25 iterations.

Fig. 8. Schematic representation of how the ®broblast grew on an

electron microscope grid. One end of the cell body attached to the

top surface of the grid bars, while another end attached to the sup-

port membrane below. The directions of the x, y and z axes are

shown.


the�x and �y directions, respectively. It is easiest to ®rst

view the reconstruction by examining slices at speci®c y

planes, with horizontal and vertical axes in the�x and �z

directions, respectively (see Fig. 9). These slices reveal that

the left side of the cell is located high up in the reconstruction

volume, while the `tail' of the cell at the right side is located at

the bottom of the reconstruction volume (see Fig. 8). The cell

was attached to the formvar support ®lm at right, and to the

tops of the grid bars (13 mm away) above the `well' at left. (At

the point of cell attachment to the formvar ®lm, one can see a

void in the mass density; see especially the x�27´4 mm slice

in Fig. 10.) The general orientation of the cell is also shown

in the x plane slices of Fig. 10, where the cell body can be

seen to move towards the bottom of the reconstruction

volume (larger values of z along the beam axis) as slices are

taken further to the right. This can also been seen in the

original projection data of Fig. 6. Another interesting feature

visible at large positive tilt angles in Fig. 6 is a small `tentacle'

at the upper edge of the cell. This `tentacle' can be observed

in the reconstruction volume in Fig. 11 at the slices

z�17´6 mm and z�20´0 mm, and in Fig. 9 at the slices

y�17´2 mm and y�18´1 mm.

In the reconstruction volume, one can see a number of

small vesicles which overlap in the z direction and therefore

are impossible to identify as being separate when looking

only at the 08 projection (see especially the x�5´8 mm slice

in Fig. 10 and the y�15´6 mm slice in Fig. 9). These vesicles

are presumably round, so their oblong appearance in the

reconstruction volume provides a means of estimating the z

resolution to be about 250 nm, along with 100 nm in the x

and y directions. We believe this is the highest resolution

three-dimensional reconstruction of unlabelled features in

an unsectioned, hydrated eucaryotic cell.

Discussion

In the demonstration reported here, the maximum tilt angle

was 608. As can be seen in Fig. 4, this is limited by the short

working distance of the zone plate optics used, and by the

thickness of the cryo specimen holder. To increase the

working distance with improved resolution zone plates,

larger zone plate diameters will be required (the electron

beam lithography machine used for zone plate fabrication

has a laser intereferometer for large ®eld positioning, so this

should not present a problem). It may be possible to thin the

cryo holder further; alternative schemes which could be

developed include systems like the Philips Compustage and

the Gatan ultra high tilt holder, or mounting a specimen at

the end of a cold pipette.

With a tomographic reconstruction such as is shown in

Figs 9±11, it should be possible to measure the volume and

mass of various components of cells. In ¯uorescence

microscopy, the task is made somewhat easier since one

can set an intensity threshold to determine where the

¯uorophore is present. In Fig. 12, we show a rendering of

the cell reconstruction using a threshold to attempt to

highlight the large dense vesicles present in the reconstruc-

tion of Figs 9±11. Reconstructions of objects with the

complexity of a cell will bene®t from the application of

more sophisticated methods of object identi®cation and

segmentation.

Because the cryo STXM is operated at a synchrotron

Fig. 9. Reconstruction slices in the y plane. On the left is shown the 08 projection image (shown also in Fig. 6), with the horizontal and ver-

tical axes indicated in the x and y directions, respectively. In this projection, the�z direction goes into the paper away from the viewer. White

lines are shown at the location of slices in the y direction, which are then shown on the right; these slices are through the reconstruction

volume with horizontal and vertical axes in the x and z directions, respectively.

88 Y. WANG ET AL.


radiation beamline with an energy resolution of , 0´1 eV, it

should also be possible to combine tomography with X-ray

absorption spectroscopy. Images can be obtained on both

side of X-ray absorption edges of certain low-Z elements

such as carbon and calcium and used to map the

concentration of this element, provided it is present at a

local concentration of about 1% or more (EngstroÈm, 1946;

Bigler et al., 1983). In addition, X-ray absorption near-edge

structure (XANES) resonances appear at X-ray energies

within a few eV of the elemental absorption edge, and can be

used to map different regions with high concentrations of

certain chemical states of an element. The intrinsic width of

such XANES peaks is , 0´2 eV, so that cryo TXM systems

(Schneider, 1998) with an energy resolution of 1±2 eV are

less well suited to their exploitation. In a scanning

transmission X-ray microscope, XANES imaging has been

used to map protein and DNA in sperm (Zhang et al., 1996)

and collagen and apatite in bone (Buckley et al., 1997), and

to map phase separation in polymer blends (Ade et al.,

1992). Such contrast mechanisms could be exploited in


Fig. 10. Reconstruction slices in the x plane. At the top is shown the 08 projection image (shown also in Fig. 6), with the horizontal and

vertical axes indicated in the x and y directions, respectively. White lines are shown at the location of slices in the x direction, which are

then shown at right; these slices are through the reconstruction volume with horizontal and vertical axes in the y and z directions,

respectively.


tomography by taking two or more images at different

photon energies at each projection, and then carrying out

spectroscopic analysis on the reconstructed volume mea-

surements of X-ray absorption coef®cient. While we have

shown that mass loss in frozen hydrated specimens is not

signi®cant at the radiation dose levels required for

tomography (Maser et al., 2000), it may be that XANES

peaks (which represent bond strength rather than overall

Fig. 11. Reconstruction slices in the z plane. Each slice has its horizontal and vertical axes in the x and y direction, respectively, and therefore

resembles the 08 projection shown in Fig. 6. However, what is shown here is only one slice (with 100 nm thickness) in the reconstruction

volume at the z distance indicated.

Fig. 12. An isosurface rendering of a density threshold chosen to highlight vesicles surrounding the cell nucleus. The isosurface rendering is

shown along with slices x�4´1 mm, y�18´1 mm, z�21´6 mm in the reconstruction volume (of the type shown in Figs. 9±11) from three

orthogonal planes. The rendering was created using the IDL slicer routine.

90 Y. WANG ET AL.


electron density) will prove more sensitive; we plan to

address this question in future work. The same resonances

can also be exploited by electron energy loss spectroscopy

(EELS) at signi®cantly higher spatial resolution. However,

EELS analysis requires # 100 nm thick specimens and is

therefore incompatible with electron tomography; further-

more, most EELS instruments have lower energy resolution,

and the energy-loss near-edge structure appears on top of a

large plural inelastic scattering background.

This ®rst demonstration of tomography on a frozen

hydrated eucaryotic cell required considerable operator

attention to tilt, centre and focus for each projection. The

time required to acquire each projection was about 45 min

to one hour. We expect this to be reduced considerably, in

part due to the use of more ef®cient zone plates and

detectors that are under development in our laboratory.

Finally, we are aware of the considerable success that the

electron microscope community has had in automating the

acquisition of tomographic data (Dierksen et al., 1992;

Koster et al., 1992) and efforts towards more automated

tomography dataset acquisition with cryo STXM are underway.

Some challenges arise as the spatial resolution is

improved. From Eq. (4), we see that improvements in

transverse resolution by decreasing the zone plate outer-

most zone width drN lead to decreases in the depth of ®eld,

which scales as drN2. For a drN�30 nm zone plate at

l�2´4 nm, the depth of focus becomes only 2dl �3´7 mm.

This would suggest the adoption of 3D imaging techniques

used in optical microscopy such as the recording of a

through-focus image set followed by 3D deconvolution

(Agard & Sedat, 1983; Fay et al., 1989). However, these

approaches exploit the fact that ¯uorescence microscopy is

an intrinsically incoherent imaging process, and dyes

¯uoresce only according to the intensity of the illumination.

In X-ray microscopy, most specimens exhibit both absorp-

tion and phase contrast (Schmahl et al., 1994); in-focus

incoherent bright®eld images map only the absorption

contrast of the specimen, but out-of-focus planes are imaged

with a defocus-dependent mix of absorption and phase

contrast (furthermore the image intensity is not strictly

proportional to the phase). We believe that the solution lies

in the recording of holograms at each rotation angle, and

processing the data according to the methods of diffraction

tomography (Devaney, 1986; Maleki & Devaney, 1994), and

preliminary experiments have been carried out towards this

goal.

Conclusion

We have performed tomography experiments with a cryo

scanning transmission X-ray microscope, and a three-

dimensional reconstruction of a mouse 3T3 ®broblast

with 100 nm transverse resolution and 250 nm long-

itudinal resolution was obtained. The use of cryo methods

allows the full data set to be recorded without observable

specimen changes from radiation damage. The specimen

thickness greatly exceeds that which can be studied in

electron tomography, so that whole eukaryotic cells can be

studied. Efforts are being made to combine tomography

with the elemental and chemical state mapping capabilities

of the X-ray microscope to identify features of sample in

three dimensions. Further re®nements are being made

to improve the ef®ciency and user-friendliness of the

instrument for tomographic data acquisition.

Acknowledgements

We thank Raymond Fliller, Konstantin Kaznacheyev, Janos

Kirz, Steve Spector, Don Tennant, Jan Warnking, Matthias

Weigel and Sue Wirick for their contributions to cryo STXM,

and Azeddine Ibrahimi for his help with cell culture. We

especially thank Barry Winn for his involvement in

additional tomography experiments not reported here, and

for his help with Fig. 2. We gratefully acknowledge support

from the Of®ce of Biological and Environmental Research,

US DoE under contract DE-FG02-89ER60858, the National

Science Foundation under grants DBI-9605045 and ECS-

9510499, and the Alexander von Humboldt-Foundation

(Feodor-Lynen Fellowship, JM). This work was carried out at

the National Synchrotron Light Source at Brookhaven

National Laboratory, which is supported by the US

Department of Energy.

References

Ade, H., Zhang, X., Cameron, S., Costello, C., Kirz, J. & Williams, S.

(1992) Chemical contrast in X-ray microscopy and spatially

resolved XANES spectroscopy of organic specimens. Science, 258,

972±975.

Agard, D. & Sedat, J. (1983) Three-dimensional architecture of a

polytene nucleus. Nature, 302, 676±681.

Bennett, P.M., Foster, G.F., Buckley, C.J. & Burge, R.E. (1993) The

effect of soft X-radiation on myo®brils. J. Microsc. 172, 109±119.

Bigler, E., Polack, F. & Lowenthal, S. (1983) Quantitative mapping

of atomic species by X-ray absorption spectroscopy and contact

microradiography. Nucl. Instrum. Meth. Phys. Res. 208, 387±

392.

Born, M. & Wolf, E. (1980) Principles of Optics, 6th edn. Pergamon

Press, Oxford.

Bracewell, R.N. & Riddle, A.C. (1967) Inversion of fan-beam scans

in radio astronomy. Astrophys. J. 150, 427±434.

Buckley, C.J., Khaleque, N., Bellamy, S.J., Robins, M. & Zhang, X.

(1997) Mapping the organic and inorganic components of tissue

using NEXAFS. J. Physique, IV 7 (C2 Part 1), 83±90.

Carrington, W.A., Lynch, R.M., Moore, E.D.W., Isenberg, G.,

Fogarty, K.E. & Fay, F.S. (1995) Superresolution three-dimen-

sional images of ¯uorescence in cells with minimal light

exposure. Science, 268, 1483±1487.

Chapman, H.N., Fu, J., Jacobsen, C. & Williams, S. (1996) Dark-

®eld X-ray microscopy of immunogold-labeled cells. J. Microsc.

Soc. Am. 2 (2), 53±62.



Coetzee, J. & van der Merwe, C.F. (1984) Extraction of substances

during glutaraldehyde ®xation of plant cells. J. Microsc. 135,

147±158.

Devaney, A.J. (1986) Reconstructive tomography with diffracting

wave®elds. Inverse Problems, 2, 161±183.

Dierksen, K., Typke, D., Gegerl, R., Koster, A.J. & Baumeister, W.

(1992) Towards automatic electron tomography. Ultramicro-

scopy, 40, 71±87.

Dubochet, J., Adrian, M., Chang, J.J., Homo, J.-C., Lepault, J.,

McDowell, A.W. & Schultz, P. (1988) Cryo-electron microscopy of

vitri®ed specimens. Q. Rev. Biophys. 21, 129±228.

Echlin, P. (1992) Low-Temperature Microscopy and Analysis. Plenum,

New York.

EngstroÈm, A. (1946) Quantitative micro- and histochemical

elementary analysis by Roentgen absorption spectrography.

Acta Radiologica (Supplement), 63, 1±106.

Fay, F.S., Carrington, W. & Fogarty, K.E. (1989) Three-dimensional

molecular distribution in single cells using the digital imaging

microscope. J. Microsc. 153, 133±149.

Goodman, J.W. (1968) An Introduction to Fourier Optics. McGraw-

Hill, San Francisco.

Grimm, R., Singh, H., Rachel, R., Typke, D., Zillig, W. & Baumeister,

W. (1998) Electron tomography of ice-embedded prokaryotic

cells. Biophys. J. 74, 1031±1042.

Gustafsson, M.G.L., Agard, D.A. & Sedat, J.W. (1996) 15M: 3D

wide®eld light microscopy with better than 100 nm axial

resolution. J. Microsc. 195, 10±16.

Haddad, W.S., McNulty, I., Trebes, J.E., Anderson, E.H., Levesque,

R.A. & Yang, L. (1994) Ultra high resolution X-ray tomography.

Science, 266, 1213±1215.

Hegerl, R. & Hoppe, W. (1976) In¯uence of electron noise on three-

dimensional image reconstruction. Z. Naturforsch. 31a, 1717±

1721.

Hell, S.W., Schrader, M. & van der Voort, H.T.M. (1997) Far-®eld

¯uorescence microscopy with three dimensional resolution in

the 100-nm range. J. Microsc. 187, 1±7.

Henke, B.L., Gullikson, E.M. & Davis, J.C. (1993) X±ray interac-

tions: photoabsorption, scattering, transmission and re¯ection at

E�50±30 000 eV, Z�1±92. Atomic Data Nuclear Data Tables,

54, 181±342.

Howells, M.R., Jacobsen, C.J. & Lindaas, S. (1994) Possibility for

one-shot tomography using a high-gain free-electron laser.

Proceedings of the Workshop on Scienti®c Applications of Coherent

X-Rays, number NTIS CONF-940250, Washington, D.C.,

February 1994 (ed. by J. Arthur, G. Materlik and H. Winick).

National Technical Information Service. SLAC-SSRL 0066,

Stanford, California.

Hubbell, J.H., Gimm, H.A. & YÈ verbù, I. (1980) Pair, triplet and total

atomic cross sections (and mass attenuation coef®cients) for

1 MeV±100 GeV photons in elements Z�1±100. J. Phys. Chem.

Ref. Data, 9, 1023±1147.

Jacobsen, C., Medenwaldt, R. & Williams, S. (1998) A perspective

on biological x-ray and electron microscopy. X-ray Microscopy

and Spectromicroscopy (ed. by J. Thieme, G. Schmahl, E. Umbach

and D. Rudolph), pp. II-93±102. Springer-Verlag, Berlin.

Jacobsen, C., Lindaas, S., Williams, S. & Zhang, X. (1993) Scanning

luminescence X-ray microscopy: imaging ¯uorescence dyes at

suboptical resolution. J. Microsc. 172, 121±129.

Kaczmarz, S. (1937) Angenaherte Au¯osung von systemen

linearer Gleichungen. Bull. Acad. Polon. Sci. Lett. A, 1937,

355±357.

Kirz, J., Jacobsen, C. & Howells, M. (1995) Soft X-ray microscopes

and their biological applications. Q. Rev. Biophys. 28 (1), 33±

130. Also available as Lawrence Berkeley Laboratory report

LBL±36371.

Koster, A.J., Chen, H., Sedat, J.W. & Agard, D.A. (1992) Automated

microscopy for electron tomography. Ultramicroscopy, 46, 207±

227.

Lehr, J. (1997) 3D X-ray microscopy: tomographic imaging of

mineral sheaths of bacteria Leptothrix ochracea with the

GoÈttingen X-ray microscope at BESSY. Optik, 104(4), 166±170.

Maleki, M.H. & Devaney, A.J. (1994) Noniterative reconstruction of

complex-valued objects from two intensity measurements. Opt.

Eng. 33 (10), 3243±3253.

Maser, J., Jacobsen, C., Kirz, J., Osanna, A., Spector, S., Wang, S. &

Warnking, J. (1998) Development of a cryo scanning X-ray

microscope at the NSLS. X-ray Microscopy and Spectromicroscopy

(ed. by J. Thieme, G. Schmahl, E. Umbach and D. Rudolph), pp. I-

35±44. Springer-Verlag, Berlin.

Maser, J., Osanna, A., Wang, Y., Jacobsen, C., Kirz, J., Spector, S.,

Winn, B. & Tennant, D. (2000) Soft X-ray microscopy with a

cryo STXM: I. Instrumentation, imaging and spectroscopy. J.

Microsc. 197, 68±79.

McEwen, B.F., Downing, K.H. & Glaeser, R.M. (1995) The relevance

of dose-fractionation in tomography of radiation-sensitive speci-

mens. Ultramicroscopy, 60, 357±373.

Michette, A.G. (1986) Optical Systems for Soft X-rays. Plenum, New

York.

Moronne, M.M. (1999) Development of X-ray excitable lumines-

cent probes for scanning X-ray microscopy. Ultramicroscopy, 77,

23±36.

NeuhaÈusler, U., Jacobsen, C., Schulze, D., Stott, D. & Abend, S.

(1999) A specimen chamber for soft X-ray spectromicroscopy on

aqueous and liquid samples. J. Synchrotron Radiat. in press.

O'Toole, E., Wray, G., Kremer, J. & McIntosh, J.R. (1993) High

voltage cryomicroscopy of human blood platelets. J. Struct. Biol.

110, 55±66.

Sayre, D., Kirz, J., Feder, R., Kim, D.M. & Spiller, E. (1977) Potential

operating region for ultrasoft X-ray microscopy of biological

specimens. Science, 196, 1339±1340.

Schmahl, G., Rudolph, D., Schneider, G., Guttmann, P. & Niemann,

B. (1994) Phase contrast X-ray microscopy studies. Optik, 97,

181±182.

Schneider, G. (1998) Cryo X-ray microscopy with high spatial

resolution in amplitude and phase contrast. Ultramicroscopy, 75,

85±104.

Schneider, G., Schliebe, T. & Aschoff, H. (1995) Cross-linked polymers

for nanofabrication of high-resolution zone plates in nickel and

germanium. J. Vac. Sci. Technol., B, 13(6), 2809±2812.

Schneider, G. & Niemann, B. (1998) Cryo x-ray microscopy

experiments with the x-ray microscope at BESSY. X-ray Microscopy

and Spectromicroscopy (ed. by J. Thieme, G. Schmahl, E. Umbach

and D. Rudolph), pp. I-25±34. Springer-Verlag, Berlin.

Schneider, G., Niemann, B., Guttmann, P., Rudolph, D. & Schmahl,

G. (1995) Cryo X-ray microscopy. Synchrotron Radiation News, 8

(3), 19±28.

92 Y. WANG ET AL.


Shepp, L.A. & Logan, B.F. (1974) The Fourier reconstruction of a

head section. IEEE Trans. Nucl. Sci. NS-21, 21±43.

Spector, S., Jacobsen, C., Tennant, D. (1997) Process optimization

for production of sub-20 nm soft X-ray zone plates. J. Vac. Sci.

Technol., B, 15 (6), 2872±2876.

Tanabe, K. (1971) Projection method for solving a singular system.

Numer. Math. 17, 203±214.

Taylor, K. & Glaeser, R. (1974) Electron diffraction of frozen,

hydrated protein crystals. Science, 106, 1036±1037.

Williams, S., Zhang, X., Jacobsen, C., Kirz, J., Lindaas, S., van'tHof,

J. & Lamm, S.S. (1993) Measurements of wet metaphase

chromosomes in the scanning transmission X-ray microscope.

J. Microsc. 170, 155±165.

Winn, B., Ade, H., Buckley, C., Howells, M., Hulbert, S., Jacobsen,

C., Kirz, J., McNulty, I., Miao, J., Oversluizen, T., Pogorelsky, I. &

Wirick, S. (1996) X1A: second generation undulator beamlines

serving soft X-ray spectromicroscopy experiments at the NSLS.

Rev. Sci. Instrum. 67 (9), 1±4, Paper A31.

Zhang, X., Balhorn, R., Mazrimas, J. & Kirz, J. (1996) Mapping and

measuring DNA to protein ratios in mammalian sperm head by

XANES imaging. J. Struct. Biol. 116, 335±344.



Documents

Soft X-ray microscopy with a cryo scanning transmission X