tomography of insulating biological and geological materials using

Journal of Microscopy, Vol. 233, Pt 3 2009, pp. 372–383

Received 29 May 2008; accepted 17 October 2008

Tomography of insulating biological and geological materials usingfocused ion beam (FIB) sectioning and low-kV BSE imaging

D . A . M A T T H I J S D E W I N T E R ∗, C . T . W . M . S C H N E I J D E N B E R G ∗,M . N . L E B B I N K ∗, B . L I C H‡, A . J . V E R K L E I J ∗, M . R . D R U R Y†& B . M . H U M B E L ∗∗Electron Microscopy and Structure Analysis, Cellular Architecture and Dynamics, Faculty ofSciences, Utrecht University, NL-3584 CH Utrecht, The Netherlands

†Earth Sciences, Faculty of Geosciences, Utrecht University, NL-3584 CD Utrecht, The Netherlands

‡FEI Company, Achtseweg Noord 5, 5600 KA Eindhoven, The Netherlands

Key words. Ceramics, endothelial cell, focused ion beam, insulator, minerals,scanning electron microscopy, tomography.

Summary

Tomography in a focused ion beam (FIB) scanningelectron microscope (SEM) is a powerful method forthe characterization of three-dimensional micro- andnanostructures. Although this technique can be routinelyapplied to conducting materials, FIB–SEM tomography ofmany insulators, including biological, geological and ceramicsamples, is often more difficult because of charging effectsthat disturb the serial sectioning using the ion beam orthe imaging using the electron beam. Here, we show thatautomatic tomography of biological and geological samplescan be achieved by serial sectioning with a focused ion beamand block-face imaging using low-kV backscattered electrons.In addition, a new ion milling geometry is used that reduces theeffects of intensity gradients that are inherent in conventionalgeometry used for FIB–SEM tomography.

Introduction

A FIB–SEM microscope is a scanning electron microscope(SEM) combined with a focused ion beam (FIB) such thatboth beams coincide at their focal points. This combinationenables bulk samples to be locally sectioned by ion milling,producing new block faces, which are subsequently imagedat high resolution with the electron beam. One of the widelyused applications of FIB instruments, first developed on single-beam FIB microscopes and later applied in the FIB–SEM, iscommonly referred to as FIB tomography (Inkson et al., 2001;Kubis et al., 2004). This method involves serial sectioning

Correspondence to: Dr. M.R. Drury. Tel: +31-30-2535108; fax: +31-30-2537725;

e-mail: [email protected] or Dr. B.M. Humbel. Tel: +31-30-2533449;

fax: +31-30-2513655; e-mail: [email protected]

with the ion beam and imaging of each section, using the ionbeam in FIB microscopes or the electron beam in FIB–SEMmicroscopes. FIB-based tomography is a powerful method forstudying three-dimensional (3D) nano- and microstructuresof materials such as metals and semiconductors (Inksonet al., 2001; Kammer et al., 2005; Williams et al., 2005; Uchicet al., 2006, 2007; Holzapfel et al., 2007; Kato et al., 2007;McGrouther & Munroe, 2007).

Recently, several applications of FIB–SEM microscopyto biological and geological materials and ceramics havebeen reported (Schaffer et al., 2007; Desbois et al., 2008;Knott et al., 2008). In biology, the FIB–SEM microscope wasfirst used to analyse difficult-to-cut material, such as teeth(Giannuzzi et al., 1999; Nalla et al., 2005) and bone/dentalimplant interfaces (Giannuzzi et al., 2007). In addition toSEM imaging, thin lamellae were prepared in the FIB–SEMand analysed in a transmission electron microscope (TEM).Another approach was to use the ion beam to mill a cross-section at the place of interest and then to image the freshsurface (Drobne et al., 2005a, b, 2007, 2008; Greve et al.,2007) using secondary (SE) electrons, backscattered electrons(BSE) or ion beam-induced secondary electrons (ISE) (Drobneet al., 2005a). Although imaging with ISE avoids charging,the sample is milled away during imaging.

In earth and planetary sciences, the FIB–SEM is also usedfor making sections of difficult-to-prepare materials as wellas of rare and valuable samples. Heaney et al. (2001) werethe first to apply the FIB–SEM to prepare thin lamellae for TEMfrom a variety of mineral and meteorite samples. The FIB–SEMtechnique has been used to prepare TEM lamellae of naturalmicrodiamonds (Dobrzhinetskaya et al., 2003), phases fromhigh-pressure experiments (Dobrzhinetskaya et al., 2004;Irifune et al., 2005) and weathered surfaces of feldspar crystals(Lee et al., 2007). Other applications in earth sciences have

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society

T O M O G R A P H Y W I T H F I B – S E M 3 7 3

used FIB milling to make sections of melt inclusions (Bleineret al., 2006) and microfossils (Kempe et al., 2005).

In this study, we demonstrate the power of FIB–SEMtomography to find the place of interest within a bulk (mm-to-cm-sized) sample and analyse this region in three dimensions,at a comparable resolution to the TEM, in both biologicaland geological samples. In our instrument, an FEI Nova 600NanoLab, serial sections can have a thickness between 12 and500 nm, and volumes up to 4000μm3 can be sectioned in a fewhours, to tens of hours, using the automatic Slice and ViewTM

(Heymann et al., 2006) software (FEI Company, Eindhoven,The Netherlands). This technique is relatively straightforwardwhen applied to conducting materials. By contrast, FIB–SEMtomography of many insulators is more difficult because thefresh surfaces, produced by ion beam sectioning, can chargeeither during ion milling or during electron imaging. Thework on ceramics (Holzer et al., 2004) has shown that FIB–SEM tomography can be applied to some insulators such asBaTiO3, which have modest electrical conductivity at roomtemperature, using low-kV (5 kV) BSE imaging, and this is theapproach we have used.

To reduce charging of biological samples, extendedosmification, using the OTOTO technique, originallydeveloped by Tanaka & Mitsushima (1984) and Tanaka(1989), and surface coating (Drobne et al., 2005a,b, 2007)can be used. Extensive use of osmium, however, can have adestructive effect on the cell morphology (Maupin & Pollard,1978; Behrman, 1984). For many insulating geologicaland ceramic samples, no such treatments can be applied,although porous materials can be impregnated, if the porosityis connected, with a low-melting point alloy (Holzer et al.,2004). Here, imaging with the ion beam could remedycharging (Drobne et al., 2005a), but as milling continuesduring imaging, other approaches are needed.

The conventional geometry for FIB–SEM tomography,which produces sections perpendicular to the sample surface,

can cause problems such as a brightness gradient fromthe top to the bottom of the section, shadowing from thesidewalls, and a decrease of signal with depth (Fig. 1Aand B). The problem of shadowing can be avoided bymilling a sufficiently large, U-shaped trench that surroundsthe volume to be sectioned (Holzer et al., 2004). In thisstudy, we have developed ion milling conditions and electronimaging conditions suitable for FIB–SEM tomography ofinsulating biological and geological samples. We describea new sectioning geometry that overcomes the problems,such as those of the brightness gradient and the decrease ofsignal with depth, observed with the conventional FIB–SEMtomography geometry. Furthermore, we will consider aspectsof imaging of biological and geological samples, includingcontrast, resolution and charging.

Material and methods

Biological sample

Freshly isolated endothelial cells from human umbilical cord(HUVEC) were seeded on BD matrigelTM (BD Biosciences,San Jose, CA)-coated culture dishes and cultivated untilthey formed a dense layer called cobblestone. Then the cellswere fixed with 1.5% glutaraldehyde. While fixing, they arescraped with a rubber policeman and collected as a pellet ina centrifugation tube. After 1 h of glutaraldehyde fixation,the cells were post fixed with 2% osmium tetroxide and 1.5%ferrocyanide. Then they were dehydrated through a series ofalcohols and embedded in Epon (Fluka, Buchs, Switzerland)(Mollenhauer, 1964).

Using an ultramicrotome (Ultracut E; Reichert-Jung, nowLeica Microsystems, Vienna, Austria), the pellet was freedwith a glass knife and a small pyramid was prepared. Thesmall pyramid was cut off the resin block and mounted ona support stub of the FIB–SEM in a thick layer of conductive

Fig. 1. In the conventional slice and view geometry, the specimen is tilted towards the ion beam by 52◦. (A) Endothelial cells were imaged with BSE usingthe through-the-lens detector. The gradient in brightness from the top to the bottom of the section is clearly visible. (B) A sample of forsterite imaged withBSE using the ET detector. Note the shadow on the right-hand side of the milled section.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy, 233, 372–383

3 7 4 D . A . M A T T H I J S D E W I N T E R E T A L .

carbon. Special attention was paid to ensure that the carbonmade contact with the osmium-blackened pellet to improveconductivity. A 3-nm thick layer of platinum was depositedonto the sample, with a sputter coater (HQ280; Cressington,Cressington Scientific Instruments Ltd, Watford, UK).

Geological sample

The geological sample studied is a synthetic forsterite(Mg2SiO4) polycrystal containing 10% volume intergranularglass. The sample was held at high temperature above themelting temperature of the glass and then rapidly quenchedto avoid crystallization in the glass (ten Grotenhuis et al.,2005). The bulk sample was sectioned and mechanicallypolished using a diamond paste and colloidal silica. All phasesin the sample are insulators that charge strongly duringconventional electron imaging. These insulating specimenswere coated with carbon, gold or platinum to reduce chargingand enable conventional SEM imaging, X-ray microanalysis(EDX) or electron backscattered diffraction (EBSD).

Preparing for FIB–SEM tomography

Slice and ViewTM is an FEI software package of our FEI Nova600 NanoLab DualBeam instrument (FEI Company) thatallows automatic cutting of slices with the FIB and sequentialrecording of the freshly cut surface with the electron beam toproduce FIB–SEM tomograms. The principle of the preparationfor slice and view is similar for both biological and geologicalsamples, whereas the imaging conditions and the volume

studied by serial sectioning differ for each sample. The imagingconditions and volume studied mainly influence the totaltime of processing. In this section, the preparation for Epon-embedded cells is described and the specific parameter settingsare given.

Figure 2(A) shows an overview of the mounted Epon block,taken with the electron beam at 5 kV acceleration voltageand a beam current of 0.40 nA, in the SE imaging mode.At higher magnification, traces of the embedded cell layersbecame visible (Fig. 2B, arrow) and an area of densely packedcells could easily be located. The sample was tilted 52◦ toorientate the surface perpendicular to the Ga+ ion beam and aU-shaped trench was cut around the area of interest (Fig. 2C).The speed of the milling process depends on the sputter yieldof the material, the ratio between the incoming ions and thesputtered target atoms. The sputter yield is a material property.The milling time is a function of the sputter yield, ion currentand the milling area and depth. To mill the U-shaped trench,we used an ion current of 7–20 nA, at 30 keV accelerationvoltage, which took typically 15 min for an Epon sample. Highion beam currents were used in this ‘rough cutting’ stage sothat the sample could be prepared as quickly as possible, butwithout damaging the adjacent slice and view volume.

After the rough cutting, the central part of the U-shapedtrench was covered with an approximately 750-nm-thicklayer of platinum by ion beam-induced deposition (IBID)(Fig. 2D). The ion current was chosen to give a deposition timeof around 10–15 min. For a volume of 100 μm3 (10 × 10 ×1 μm), a current of 0.3 nA was sufficient.

Fig. 2. New slice and view geometry. An Epon block with embedded HUVECs is mounted on an SEM stub with conducting carbon. The surface wassputter-coated with a 2-nm platinum layer and images were taken in the SE mode. From (A) to (F), the sequence of preparation for FIB–SEM tomographyis depicted. An overview of the Epon block (A). At higher magnification, individual cells are visible (B, arrow). At the area of interest, a U-shaped trenchis milled at 52◦ stage tilt, with the specimen surface perpendicular to the ion beam (C), and a layer of platinum is deposited (D). Then the stage is set backat 0◦, and with an incident ion beam at 38◦, the surface is smoothened (D) before performing slice and view tomography (E, F).



Finally, for slice and view, the sample was tilted back to 0◦,and the block was milled with the ion beam pointing at anangle of 38◦ (Fig. 2E) to the sample surface. First, a coarsesurface was created with a moderate current of 0.5 nA–1nA, then the surface was milled with a reduced ion currentof 50–300 pA (Fig. 2E) to produce a smooth surface. Afterthis final milling, the cells were visible at low magnificationwith SE imaging (Fig. 2E, arrow) and the sample was ready forslice and view. The minimum slice thickness is approximately10 nm. Depending on the brittleness of the material, ioncurrents between 30 and 300 pA were used. Figure 2(F) showsthe sectioned volume after a set of slice and view images havebeen taken.

Electron imaging conditions

In an SEM, beam interactions with the specimen yield severaltypes of signals that can be used for imaging. The mostfrequently used are the SEs and the BSEs. Both the SE andthe BSE can be detected with the Everhart–Thornley detector(ETD). In SE mode, all electrons (SE + BSE) escaping fromthe surface are attracted by a positive bias (+250 V) on thedetector. In BSE mode, the low-energy SEs are repelled by anegative bias (–50 V). The FEI Nova 600 NanoLab microscopealso has a through-the-lens detector (TLD) for BSE, whichis used optimally with the immersion lens mode. The choicebetween SE and BSE mode depends on the type of material andthe microstructure under investigation. For both applications(biology and geology), different imaging modes and theirsettings are described separately, because of the fundamentaldifferences in material type and preparation methods.

3-D reconstruction

For the tomogram constructions, a series of slice and viewimages were combined into a single 3D volume using theAmira software package (Visage Imaging GmbH, Berlin,Germany) and converted into the .MRC file format using theIMOD package (Boulder Laboratory, University of Colorado,Boulder, Colorado, USA) (Kremer et al., 1996). These imageswere then aligned using the alignment operation in IMODset to a zero-angle difference between the images. For thegeological tomogram, the images were analysed and manuallymodelled in IMOD, whereas the biological tomogram wasmanually annotated in Amira. In both cases, the structures ofinterest were traced to provide a 3D representation.

Results and discussion

Electron imaging conditions: biological samples

For a successful application in cell biology, smallultrastructural details, such as membranes, need to beresolved. This requires a resolution of at least 5 nm, in both

the X and the Y directions, on an area of typically about10 × 10 μm2. Either SE or BSE can be used for imaging.SEs are predominately produced at the surface, and therefore,a high resolution can be achieved, mainly depending onthe spot size of the incident electron beam. The contrast,however, is caused by surface topology and is therefore verysusceptible to roughness and contamination. Furthermore,after ion sectioning, the surface of the material is rather smoothand limited topological contrast is produced, unless the sampleis porous or fractured. In addition, the low-energy SEs arestrongly influenced by charging, which is most prominentin geological samples, such as forsterite or silicate glasses(Fig. 1B).

BSEs have, in principle, a lower resolution but they revealatomic number contrast, which is better suited for imagingheavy-metal-stained structures in block faces. The biologicalsamples investigated were fixed with osmium tetroxide andferrocyanide, which results in high atomic number contrast.BSEs were usually detected with the TLD in the immersion lensmode.

The final resolution is dependent on the spot size ofthe incident electron beam and the excitation volume.To estimate the size of the electron, interaction volumesimulations have been done with CASINO (http://www.gel.usherbrooke.ca/casino), a software tool that tracks multiplescattering of electrons in multilayer materials. Importantoutputs are the radius and depth of the emitted BSEs andthe backscatter electron coefficient (BSEC). Figure 3A shows asimulation of 2 keV electrons in carbon and osmium.

The escape surface in carbon has a diameter from 8 to800 nm, with acceleration voltages from 1 to 10 keV (Fig. 3B).Hence, the optimum resolution would be achieved at 1 keV. Forosmium, the resolution at 1–5 keV is in the range of 2 nm, withlower resolution at 10 keV (Fig. 3B). The much higher BSECfor osmium compared with carbon, by a factor of about 20,suggests that the detected BSE are scattered mainly by osmiumand the low-resolution BSEs for carbon can be neglected. Wetherefore used 1–2 keV for imaging.

The contrast of BSE images is also dependent on theinclination of the specimen surface to the incident electronbeam. Figure 4 shows the BSEC of a 2 keV electron beam, atdifferent angles, for carbon and osmium. Although the BSECincreases, the difference between the BSECs for carbon andosmium reduces, which indicates that the atomic numbercontrast is less at higher tilts, but there is still strong atomicnumber contrast at 38◦ tilt (Fig. 4).

At 2 keV, and a pixel size of 2.5 nm, small structures suchas caveoli (arrow), budding vesicles from the Golgi apparatus(G) and the two nuclear membranes can be resolved (NM,Fig. 5). The resolution, however, is too low to resolve thebilayer appearance of, for example, the plasma membrane.From the results, it is deduced that we have an effectiveresolution in the X-direction of about 10 nm, as the bilayerof the plasma membrane is between 5 and 10 nm thick. The



Fig. 3. (A) Results from a Monte Carlo simulation using CASINO. The interaction volume was modelled for an incident electron beam, calculated for200 000 electrons, at 2 keV beam in carbon and osmium. The program allows defining some microscope parameters and the target material and thencalculates the trajectory of the primary electrons (blue). If a primary electron is directed out of the material (red), it becomes a backscattered electron(BSE). (B) The diameter of the BSE-emitting surface at various acceleration voltages was evaluated for a carbon and an osmium sample.

Fig. 4. The effect of specimen tilt on the BSE coefficients obtained from CASINO simulations for carbon and osmium. All simulations are conducted with200 000 electrons.



Fig. 5. (A) An endothelial cell was imaged at 2 keV with backscattered electrons using the TL detector in the immersion lens mode. The pixel size was2.5 nm, with an image size of 2048 × 1768 pixels. (B) Inverted contrast. The individual bilayer (approximately 5 nm) is not visible, but the doublemembranes surrounding the cell nucleus can be resolved (see insert). Even small vesicles, such as caveoli of about 200 nm in diameter (arrows), can bediscriminated. G, Golgi apparatus; NM, nuclear membranes; ER, endoplasmic reticulum.

image also clearly demonstrates the advantage of the newmilling geometry, as no shadowing and no intensity gradientoccur.

On metal-coated biological samples, with SEM equippedwith the state-of-the-art electron optics, a resolution of 1 nm is,in principle, achievable (Hermann et al., 1988; Joy & Pawley,1992; Walther et al., 1995). The resolution achieved, of 5–10 nm, using BSE imaging on uncoated, freshly ion-milledsurfaces (Fig. 5) is lower because of the larger interactionvolume of the primary electron beam in the sample and isconsistent with the model simulations (Fig. 3).

Electron imaging conditions: geological samples

For geological samples, the resolution criteria vary, dependingon the application. In some cases, a nanometre-scale(<100 nm) resolution, in X, Y and Z directions, is requiredto image small particles, pores or thin films along grainboundaries. In other applications, for instance, in rocks withlarge crystals, the resolution requirements are less demanding.The material we have studied in detail consists of insulating

crystals and intragranular glass. The glass charges morestrongly than the forsterite crystals. Because the samples werecarbon-coated, no charging problems occur at the surface.However, when the coating is milled away, charging occurred.The contrast between the grains and the melt was obtainedfrom topology (SEs) caused by mechanical polishing, atomicnumber contrast between the grains and the melt (BSEs)- andchannelling-induced contrast between grains (BSEs). (Figs. 6and 7) The glass has a higher BSEC than the crystals (Fig. 6);however, near the glass–crystal interface, this contrast wasreversed because of topography produced by mechanicalpolishing.

The images depicted in Figs. 6 and 7 were taken from thecarbon-coated, mechanically polished surface. During sliceand view, however, uncoated surfaces were created and strongcharging occurred in the melt pockets (Fig. 10D). To overcomethe charging problem during imaging, both the primaryelectron energy and the current were reduced as much aspossible. For the sample studied here, an acceleration voltageof 7 kV was needed to obtain sufficient contrast between theglass and the crystals in BSE images of the ion-milled sections.

Fig. 6. SE images of the mechanically polished, carbon-coated, forsterite glass samples obtained with the ET detector at (A) 5 keV, (B) 15 keV, SE, and (C)30 keV acceleration voltage.



Fig. 7. BSE image of the forsterite–glass sample imaged at 7 keV and 0◦

stage tilt using the ET detector. The glass (G) has higher backscatteredcoefficient than the forsterite (C).

Slice and view tomography

Biology. The real advantage of FIB–SEM is the largemagnification range. In biological as well as in materialsciences, the specimen may be rather complex. The mostinteresting area in such a complex system is often the placewhere something goes wrong, such as the development of anatherosclerotic plaque. In the SEM, large areas can be scannedand the area of interest can be located. Additionally, the FIBallows the third dimension to be explored. The cells in the largeresin block can easily be recognized at low resolution (Fig. 2B)and an area of high density can be located. There, tomogramscan be recorded by slice and view (Fig. 8; and Video Clip S1).Here, 108 sections with a thickness of 50 nm were cut andthe block face was recorded in the BSE mode with a pixel sizeof 3.1 nm, resulting in a total volume of 6.4 × 5.5 × 5.4 μm3

(Fig. 8A). In individual sections, the endoplasmic reticulum(orange) and the mitochondria (green) were modelled(Fig. 8B). The model of a sample volume, as large as 1.1 × 4.6× 5.4 μm3, shows the extended network of the endoplasmic

Fig. 8. BSE images of Epon-embedded HUVECs. The block face sections of 5.2 × 4.6 μm size were imaged with a pixel size of 3.1 nm. For easier comparisonto TEM images, the contrast of the images was inverted in (A). In individual sections, the endoplasmic reticulum (green) and the mitochondria (orange)were modelled (B, C). The reconstructed volume (1.2 × 4.6 × 5.4 μm3) reveals a very dense and complex network of the endoplasmic reticulum. Also,the mitochondria seem to be connected, forming a large complex (C). (See Video Clip S1 in Supporting Information).



reticulum. In contrast to most 2D images, this model showsthat in HUVEC, the mitochondria form an extended networkrather than small, oval-shaped organelles (Fig. 8C).

Compared with transmission electron tomography (X =Y, approximately 2 nm; Z, approximately 4–10 nm), theresolution of FIB tomography is about five times lower. It isin the range of 5–10 nm, and the Z resolution is dependenton the slice thickness and the acceleration voltage of theelectron beam, with a minimum of about 10 nm. Hence,small connections between organelles (Geuze et al., 2003;Trucco et al., 2004) or even molecular structures (Frangakis& Forster, 2004; Baumeister, 2005) cannot be resolved.By contrast, transmission electron tomography is limitedto a section thickness of about 500 nm. This limitationcan be overcome by using serial thick sections (Marsh,2007), but this approach is tedious. Slice and view FIB–SEM tomography, by contrast, can image large networks inlarge volumes at intermediate resolution. Furthermore, thelink from nanostructures to larger-scale features, up to thescale of millimetres and centimetres, can be made directlywith the FIB–SEM technique, making FIB–SEM tomographya very powerful technique that is complementary to TEMtomography. The ability to locate the place of interest in anFIB–SEM, assess the 3D microstructure, to a resolution of about10 nm, and then prepare TEM lamellae samples for higher-resolution studies makes the FIB–SEM a key instrument in thecomplete characterization of structures, over the full rangefrom nano to macro length scales.

Geology. The results of a successful automatic slice and viewanalysis of the forsterite–glass sample using BSE–TLD modeare shown in Figs. 9–11. Figure 9 shows the surface imageof a boundary between forsterite crystals with a lens-shapedglass inclusion along the boundary that appears to be isolatedfrom the triple junction melt pocket.

Serial sections though this grain boundary are shown inFig. 10. Ion beam milling conditions were 30 kV and 0.3nA. The sample was coated with a 1-μm protection layer ofplatinum before FIB–SEM tomography. In the tilted sample,the glass appears darker than the crystals in BSE images. Abright or dark line along the glass–crystal interface indicatesthat the ion milling produces some topography, consistentwith preferential milling of the glass phase. No orientationcontrast between the crystals was observed. Charging of theion-milled section is apparent in Fig. 10C and D, at pores alonghealed fractures and within the larger melt pocket visible inthe last section. An acceleration voltage of 7 kV was needed toobtain sufficient contrast between the glass and the crystals.In consequence, lower spatial resolution was obtained forthe geological sample compared with the biological samplesimaged at lower kV. In this experiment, a pixel size of 9.4 nmwas used. Charging of the sample was stronger in the TLD–BSE mode compared with the ETD–BSE mode, and in many

Fig. 9. BSE image of grain boundary in forsterite-melt sample. A lens-shaped glass inclusion occurs along the central grain boundary. Thisinclusion has been studied by FIB–SEM tomography. The scale barrepresents 10 μm.

cases, slice and view could not be conducted in TLD–BSE modebecause of excessive charging.

The slice and view sequence in Fig. 10 and the tomogramin Fig. 11 clearly show that the lens-shaped glass inclusionvisible at the surface is not connected to the glass tube thatruns along the junction between three crystals. The grainboundary appears to be melt-free; however, the presence ofthin glass films, or impurity segregations, less than 20-nmthick, along the grain boundary (Drury & Fitz Gerald, 1996;de Kloe et al., 2000; Hiraga et al., 2002) cannot be ruled out.Selected views of the tomogram show that the glass lens doesnot have a circular section in the plane of the grain boundary,suggesting an effect of crystal anisotropy on the inclusionshape. The true wetting, or dihedral angle, between the glassinclusions and the grain boundary can be directly measured



Fig. 10. A sequence of ion-milled sections through the forsterite-melt sample. This automatic slice and view tomography has been conducted along theboundary shown in Fig. 11 with melt inclusion at the surface. The serial sections show that the melt inclusion is lens-shaped at depth and is isolated fromthe triangular cross-sectioned melt inclusion that runs along the triple junction between three crystals. This triple junction tube is exposed on the initialsurface and at depth in the last section.

in the tomogram. A true angle of 36–38◦ is obtained fromFig. 11, which compares with the apparent angle measuredon the polished sample surface of 28◦ and an apparent angleof 22–24◦ measured on the milled sections.

The uncoated material investigated (Mg2SiO4) chargedbadly in the electron beam; yet, a moderate accelerationvoltage was needed to obtain sufficient contrast betweendifferent phases. It should be noted that not all geological andceramic materials charge as badly as the mineral and silicateglass studied in our research. Holzer et al. (2004) conductedFIB–SEM tomography on porous BaTiO3 using low-kV (5 kV)BSE–TLD imaging to reduce charging. In BaTiO3, which hasrelatively high electrical conductivity at room temperature,the strong contrast between the crystals and the pores allowedimaging at low kV. Schaffer et al. (2007) have shown that othertitanite phases, Mg2TiO4 and CaTiO3, can be investigatedby FIB–SEM tomography using a relatively high voltage of15 kV and nominal beam currents 6.2 nA. Both studies (Holzeret al., 2004; Schaffer et al., 2007) reported ion beam drift dueto charging and stage drift, which can result in variable slicethickness during ion milling. Holzer et al. (2004) applied anautomatic drift correction to reduce this problem; however,as discussed by Schaffer et al. (2007), this approach cannot

always be applied. In the Mg2SiO4 glass sample, we foundthat in extreme cases of charging, the slice is not milled awaycompletely, leaving lamellae of material that block the view ofsubsequent ion milling slices. In order to obtain a good resultfrom slice and view, the charge build-up during milling mustbe reduced. This can be achieved by scanning the area that isbeing milled with the electron beam. The negative charge ofthe electron beam will compensate the positive charge fromthe ions (Stokes et al., 2007).

Future developments

Cutting with the ion beam at 38◦ to the imaging electronbeam also has consequences on the dimensions of the cellularstructures. Round vesicles, as imaged with an electron beamperpendicular to the sectioned surface, are shortened in thedirection of the tilt when imaged at an angle of 38◦. Of course,distortion produced by tilting can be restored with imageprocessing, but a system that allows tilting of the sample,so that the sectioned surface is perpendicular to the electronbeam after milling, would result in higher-quality images, withequal resolution in the X and Y directions.



Fig. 11. Two views of tomogram showing the shape of glass inclusionsalong forsterite grain boundaries. (A) The three glass inclusions arecompletely isolated, with no connections along the grain boundary. (B)View in opposite direction to (A), showing the 3D shape of the triangularcross-sectioned glass tube. This tube is mainly bounded by straight crystalfacets.

It is clear that imaging sections at moderate angles tothe electron beam results in limited contrast and resolutioncompared with imaging at zero tilt. In addition, higher contrastallows reducing the beam current and beam-induced damageon the sample. Consequently, better FIB–SEM tomographydata may be obtained with our ion milling geometry (Fig. 2)by tilting the sample 38◦ after the milling step so that thesection is then normal to the electron beam. This sequenceof operation is not yet automated and may require the use offiducial markers to ensure that successive images are aligned.

When there are large differences in the BSEC within thematerial being studied, imaging at low kV is possible. However,when contrast is low, higher voltages are needed, and sometype of charge reduction strategy is required. A low kV floodgun is often used to reduce charging induced by ion milling,and a similar approach might also reduce charging duringelectron imaging.

Conclusions

FIB–SEM tomography has been successfully applied toinsulating biological and geological samples using acombination of low-kV BSE imaging to reduce chargingduring imaging, and a new ion milling geometry thatreduces the problems of shadowing and intensity gradientsinherent to the conventional FIB–SEM tomography geometry.The results show that FIB–SEM tomography is a powerfulmethod to investigate 3D structures, such as a connected,extended network of the mitochondria in HUVEC cells, andthe distribution of glassy phase along the grain boundariesin polycrystalline minerals and ceramics. Using low-kV BSEimaging, an effective X and Y resolution of about 10 nm can beobtained during automatic FIB–SEM tomography of osmium-fixed biological samples, whereas the Z resolution is dependenton the thickness of the sections and the acceleration voltageof the electron beam.

Acknowledgements

We are grateful to NWO Groot, FEI Company and UtrechtUniversity for generous funding to purchase the DualBeaminstrument. Further support was given by the EuropeanNetwork of Excellence, FP6, and the Dutch CyttronConsortium. The reviewers are thanked for their valuablecomments on the submitted manuscript. We would also liketo thank E. van Donselaar and K. Vocking for preparingthe HUVECs, Dr. S. ten Grotenhuis for providing the forsteritesample and Dr. G. Pennock for critical reading of themanuscript.

References

Baumeister, W. (2005) Mapping molecular landscapes inside cells. Biol.Chem. 385, 865–872.

Behrman, E.J. (1984) The chemistry of osmium tetroxide fixation. TheScience of Biological Specimen Preparation 1983 (ed. by J.P. Revel, T.Barnard & G.H. Haggis), pp. 1–5. SEM Inc., AMF O’Hare, IL.

Bleiner, D., Macri, M., Gasser, P., Sautter, V. & Maras, A. (2006) FIB,TEM and LA-ICPMS investigations on melt inclusions in Martianmeteorites—analytical capabilities and geochemical insights. Talanta68, 1623–1631.

Desbois, G., Urai, J.L., Burkhardt, C., Drury, M.R., Hayles, M. & Humbel,B.M. (2008) Cryogenic vitrification and 3D serial sectioning using highcryo FIB SEM technology for brine-filled grain boundaries in Halite: firstresults. Geofluids 8, 60–72.

Dobrzhinetskaya, L.F., Green, H.W., Weschler, M., Darus, M., Wang, Y.-C., Massonne, H.-J. & Stockhert, B. (2003) Focused ion beam technique



and transmission electron microscope studies of microdiamonds fromthe Saxonian Erzgebirge, Germany. Earth Planet. Sci. Lett. 210, 399–410.

Dobrzhinetskaya, L.F., Green, H.W.I.I., Renfro, A.P., Bozhilov, K.N.,Spengler, D. & van Roermund, H.L.M. (2004) Precipitation of pyroxenesand olivine from majoritic garnet: simulation of peridotite exhumationfrom great depth. Terra Nova 16, 325–330.

Drobne, D., Milani, M., Zrimec, A., Leser, V. & Berden Zrimec, M. (2005a)Electron and ion imaging of gland cells using the FIB/SEM system.J. Microsc. 219, 29–35.

Drobne, D., Milani, M., Zrimec, A., Zrimec, M.B., Tatti, F. & Draslar, K.(2005b) Focused ion beam/scanning electron microscopy studies ofPorcellio scaber (Isopoda, Crustacea) digestive gland epithelium cells.Scanning 27, 30–34.

Drobne, D., Milani, M., Leser, V. & Tatti, F. (2007) Surface damage inducedby FIB milling and imaging of biological samples is controllable. Microsc.Res. Tech. 70, 895–903.

Drobne, D., Milani, M., Leser, V., et al. (2008) Imaging of intracellularspherical lamellar structures and tissue gross morphology by a focusedion beam/scanning electron microscope (FIB/SEM). Ultramicroscopy108, 663–670.

Drury, M.R. & Fitz Gerald, J.D. (1996) Grain boundary melt films in anexperimentally deformed olivine-orthopyroxene rock: implications formelt distribution in upper mantle rocks. Geophys. Res. Lett. 23, 701–704.

Frangakis, A.S. & Forster, F. (2004) Computational exploration ofstructural information from cryo-electron tomograms. Curr. Opin.Struct. Biol. 14, 325–331.

Geuze, H.J., Murk, J.L., Stroobants, A.K., et al. (2003) Involvement ofthe endoplasmic reticulum in peroxisome formation. Mol. Biol. Cell 14,2900–2907.

Giannuzzi, L.A., Prenitzer, B.I., Drown-MacDonald, J.L., Shofner, T.L.,Brown, S.R., Irwin, R.B. & Stevie, F.A. (1999) Electron microscopysample preparation for the biological and physical sciences usingfocused ion beams. J. Process Anal. Chem. 4, 162–167.

Giannuzzi, L.A., Phifer, D., Giannuzzi, N.J. & Capuano, M.J. (2007)Two-dimensional and 3-dimensional analysis of bone/dental implantinterfaces with the use of focused ion beam and electron microscopy.J. Oral Maxillofac. Surg. 65, 737–747.

Greve, F., Frerker, S., Bittermann, A.G., Burkhardt, C., Hierlemann, A.& Hall, H. (2007) Molecular design and characterization of the neuro-microelectrode array interface. Biomaterials 28, 5246–5258.

ten Grotenhuis, S.M.D., Drury, M.R., Spiers, C.J. & Peach, C.J.(2005) Melt distributions in olivine rocks based on electricalconductivity measurements. J. Geophys. Res. 110, B12201,doi:12210.11029/12004JB003462,002005.

Heaney, P.J., Vicenzi, E.P., Giannuzzi, L.A. & Livi, K.J.T. (2001) Focusedion beam milling: a method of site-specific sample extraction formicroanalysis of earth materials. Am. Mineral. 86, 1094–1099.

Hermann, R., Pawley, J., Nagatani, T. & Muller, M. (1988) Double-axisrotary shadowing for high-resolution scanning electron microscopy.Scanning Microsc. 2, 1215–1230.

Heymann, J.A.W., Hayles, M., Gestmann, I., Giannuzzi, L.A., Lich, B. &Subramaniam, S. (2006) Site-specific 3D imaging of cells and tissueswith a dual beam microscope. J. Struct. Biol. 155, 63–73.

Hiraga, T., Anderson, I.M., Zimmerman, M.E., Mei, S. & Kohlstedt, D.L.(2002) Structure and chemistry of grain boundaries in deformedolivine + basalt and partially molten lherzolite aggregates: evidence

of melt-free grain boundaries. Contrib. Mineral. Petrol. 144, 163–175.

Holzapfel, C., Schaf, W., Marx, M., Vehoff, H. & Mucklich, F.(2007) Interaction of cracks with precipitates and grain boundaries:understanding crack growth mechanisms through focused ion beamtomography. Scr. Mater. 56, 697–700.

Holzer, L., Indutnyi, F., Gasser, P.H., Munch, B. & Wegmann, M. (2004)Three dimensional analysis of porous BaTiO3 ceramics using FIBnanotomography. J. Microsc. 216, 84–95.

Inkson, B.J., Mulvihill, M. & Mobus, G. (2001) 3D determination of grainshape in a FeAl-based nanocomposite by 3D FIB tomography. Scr.Mater. 45, 753–758.

Irifune, T., Isshiki, M. & Sakamoto, S. (2005) Transmission electronmicroscope observation of the high-pressure form of magnesiteretrieved from laser heated diamond anvil cell. Earth. Planet. Sci. Lett.239, 98–105.

Joy, D.C. & Pawley, J.B. (1992) High-resolution scanning electronmicroscopy. Ultramicroscopy 47, 80–100.

Kammer, D., Mendoza, R., Barnett, S.A. & Voorhees, P.W. (2005)The three-dimensional microstructure of materials: measurement andanalysis. Microsc. Microanal. 11(Suppl. 2), 72–73.

Kato, M., Ito, T., Aoyama, Y., Sawa, K., Kaneko, T., Kawase, N. & Jinnai,H. (2007) Three-dimensional structural analysis of a block copolymerby scanning electron microscopy combined with a focused ion beam.J. Polymer Sci.: Part B: Polymer Phys. 45, 677–683.

Kempe, A., Wirth, R., Altermann, W., Stark, R.W., Schopf, J.W. & Heckle,W.A. (2005) Focussed ion beam preparation and in situ nanoscopicstudy of Precambrian acritachs. Precambrian Res. 140, 36–54.

de Kloe, R., Drury, M.R. & van Roermund, H.L.M. (2000) Grain boundarymelt films and melt induced weakening in experimentally deformedolivine-orthopyroxene rocks. Phys. Chem. Miner. 27, 480–494.

Knott, G., Marchman, H., Wall, D. & Lich, B. (2008) Serial sectionscanning electron microscopy of adult brain tissue using focused ionbeam milling. J. Neurosci. 28, 2959–2964.

Kremer, J.R., Mastronarde, D.N. & McIntosh, J.R. (1996) Computervisualization of three-dimensional image data using IMOD. J. Struct.Biol. 116, 71–76.

Kubis, A.J., Shiflet, G.J., Dunn, D.N. & Hull, R. (2004) Focused ion-beamtomography. Metall. Mater. Trans. A 35A, 1935–1943.

Lee, M.R., Brown, D.J., Smith, C.L., Hodson, M.E., MacKenzie, M. &Hellmann, R. (2007) Characterization of mineral surfaces using FIBand TEM: a case study of naturally weathered alkali feldspars. Am.Mineral. 92, 1383–1394.

McGrouther, D. & Munroe, P.R. (2007) Imaging and analysis of 3-Dstructure using a dual beam FIB. Microsc. Res. Tech. 70, 186–194.

Marsh, B.J. (2007) Reconstructing mammalian membrane architectureby large area cellular tomography. Meth. Cell Biol. 79, 193–220.

Maupin, P. & Pollard, T.D. (1978) Actin filament destruction by osmiumtetroxide. J. Cell Biol. 77, 837–853.

Mollenhauer, H.H. (1964) Plastic embedding mixtures for use in electronmicroscopy. Stain Technol. 39, 111–114.

Nalla, R.K., Porter, A.E., Daraio, C., et al. (2005) Ultrastructuralexamination of dentin using focused ion-beam cross-sectioning andtransmission electron microscopy. Micron 36, 672–680.

Schaffer, M., Wagner, J., Schaffer, B., Schmied, M. & Mulders, H. (2007)Automated three dimensional X-ray analysis using a dual-beam FIB.Ultramicroscopy 107, 587–597.



Stokes, D.J., Vystavel, T. & Morrissey, F. (2007) Focused ion beam (FIB)milling of electrically insulating specimens using simultaneous primaryelectron and ion beam irradiation. J. Phys. D: Appl. Phys. 40, 874–877.

Tanaka, K. (1989) High-resolution scanning electron microscopy of thecell. Biol. Cell 65, 89–98.

Tanaka, K. & Mitsushima, A. (1984) A preparation method for observingintracellular structures by scanning electron microscopy. J. Microsc.133, 213–222.

Trucco, A., Polishchuk, R.S., Martella, O., et al. (2004) Secretorytraffic triggers the formation of tubular continuities across Golgi sub-compartments. Nat. Cell Biol. 6, 1071–1081.

Uchic, M.D., Groeber, M.A., Dimiduk, D.M. & Simmons, J.P. (2006)3D microstructural characterization of nickel superalloys via serialsectioning using a dual beam FIB-SEM. Scr. Mater. 55, 23–28.

Uchic, M.D., Holzer, L., Inkson, B.J., Principe, E.L. & Maunroe, P. (2007)Three-dimensional microstructural characterization using focused ionbeam tomography. MRS Bull. 32, 408–416.

Walther, P., Wehrli, E., Hermann, R. & Muller, M. (1995) Double-layer coating for high-resolution low-temperature scanning electronmicroscopy. J. Microsc. 179, 229–237.

Williams, R.E.A., Uchic, M., Dimiduk, D. & Fraser, H.L. (2005) Threedimensional reconstruction of alpha laths in alpha/beta titanium alloys

by serial sectioning with a FEI NOVA. Microsc. Microanal. 11(Suppl. 2),836–837.

Supporting Information

Additional Supporting Information may be found in the onlineversion of this article.

Video Clip S1. With the FIB a sequence of 108 sections with athickness of 50 nm were cut and the block face was recordedin the BSE mode with a pixel size of 3.1 nm resulting in a totalvolume of 6.4 × 5.5 × 5.4 μm3. In individual sections, theendoplasmic reticulum (orange) and the mitochondria (green)were modelled for one cell. The model of the sample volume,as large as 1.1 × 4.6 × 5.4 μm3, shows the extended networkof the endoplasmic reticulum.

Please note: Wiley-Blackwell is not responsible for the contentor functionality of any supporting materials supplied by theauthors. Any queries (other than missing material) should bedirected to the corresponding author for the article.


tomography of insulating biological and geological materials using

Documents