correlative microscopy: bridging the gap between ﬂuorescence...

Available online at www.sciencedirect.comJournal of

www.elsevier.com/locate/yjsbi

Journal of Structural Biology 160 (2007) 135–145

StructuralBiology

Correlative microscopy: Bridging the gap between fluorescencelight microscopy and cryo-electron tomography

Anna Sartori *, Rudolf Gatz, Florian Beck, Alexander Rigort, Wolfgang Baumeister,Juergen M. Plitzko

Max Planck Institute of Biochemistry, Department of Molecular Structural Biology, Am Klopferspitz 18, 82152 Martinsried, Germany

Received 14 May 2007; received in revised form 19 July 2007; accepted 25 July 2007Available online 16 August 2007

Abstract

Cryo-electron tomography of frozen-hydrated biological samples offers a means of studying large and complex cellular structures inthree-dimensions and with nanometer-scale resolution. The low contrast of unstained biological material embedded in amorphous iceand the need to minimise the exposure of these radiation-sensitive samples to the electron beam result in a poor signal-to-noise ratio.This poses problems not only in the visualisation and interpretation of such tomograms, it is also a problem in surveying the sampleand in finding regions which contain the features of interest and which are suitable for recording tomograms. To address this problem,we have developed a correlative fluorescence light microscopy-electron microscopy approach, which guides the search for the structuresof interest and allows electron microscopy to zoom in on them. With our approach, the total dose spent on locating regions of interest isnegligible. A newly designed cryo-holder allows imaging of fluorescently labelled samples after vitrification. The absolute coordinates ofstructures identified and located by cryo-light microscopy are transferred to the electron microscope via a Matlab-based user interface.We have successfully tested the experimental setup and the whole procedure with two types of adherent fluorescently labelled cells, aneuronal cell line and keratinocytes, both grown directly on EM grids.� 2007 Elsevier Inc. All rights reserved.

Keywords: Correlative microscopy; Fluorescence microscopy; Cryo-electron microscopy; Cryo-electron tomography

1. Introduction

Cryo-electron tomography (cryo-ET) allows to performstructural studies of large pleiomorphic objects, such asorganelles or cells with a resolution of a few nanometers(Medalia et al., 2002; Kurner et al., 2005). It combinesthe advantages of three dimensional (3D) imaging with aclose-to-life preservation of biological materials by keepingthem embedded in vitreous ice throughout the entire imag-ing process and by avoiding potentially harmful pre-treat-ments, such as chemical fixation, dehydration and stainingwith heavy metals (Afzelius and Maunsbach, 2004; Giep-mans et al., 2005; Grabenbauer et al., 2005). A distinct dis-advantage of cryo-ET, however, and of cryo-electron

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

doi:10.1016/j.jsb.2007.07.011

* Corresponding author. Fax: +49 89 85782641.E-mail address: [email protected] (A. Sartori).

microscopy (cryo-EM) in general, is the poor signal-to-noise ratio (SNR) of the data which is due to the low intrin-sic contrast of biological material embedded in ice on theone hand and the need to minimise the exposure of thesample to electron irradiation in order to avoid radiationdamage on the other hand. As a consequence, it is notori-ously difficult to detect and identify features of interest inthe tomograms and post-processing methods, such as a‘denoising’ of the tomograms, offer only a partial remedy.Rather sophisticated image analysis techniques must beemployed in the interpretation of cryo-electron tomogramsto take advantage of the wealth of information containedin them. Another problem arising from the low contrastof cryo-EM images is the initial survey of the grids andthe task of locating under the aforementioned conditionsregions of interest. One has to keep in mind that a typicaleukaryotic cell covers an area of approximately �100 lm2

mailto:[email protected]

Fig. 1. Schematic of the whole cryo-holder assembly. (A) Section of thecryo-holder in operational mode on the stage of the optical microscopeand (B) view in perspective of the cryo-holder. Insets in (A) and (B) showmagnified views of the central part, pointing at the vitrified sampleposition, working distance (WD, in this case 7.7–8.3 mm) and insulationrequirements (scale bar: 24 mm).

136 A. Sartori et al. / Journal of Structural Biology 160 (2007) 135–145

which is large compared to the area covered by a singletomogram (�1 lm2). Therefore, there is a need to developprocedures guiding the search for specific features and tolocate areas worth recording a tomogram. Ideally, thiswould be accomplished without any exposure to the elec-tron beam prior to recording the tomogram; if pre-irradia-tion cannot be avoided altogether, one must at least aim atminimising it.

In this communication we describe a correlative lightmicroscopy (LM)-EM method which greatly facilitatesthe search for regions of interest on EM grids. Specific fea-tures highlighted by fluorescent labels are identified andlocated by fluorescent light microscopy (FLM) on the fro-zen-hydrated samples at modest magnifications, their coor-dinates are then transferred to the EM such that they canbe addressed with negligible pre-irradiation. In additionto software for the automated correlation of the FLMand EM images, we had to develop a device enabling usto examine grids with the frozen-hydrated sample by lightmicroscopy. Unlike existing cryo-holders, that have beenused for studying the effects of freezing and thawing in bio-logical materials (Echling, 1992; Belzer and Southard,1998; Cosman et al., 1989; Irimia and Karlsson, 2005;Namperumal and Coger, 1998), the holder described hereensures maintenance of the vitreous state of samples bykeeping them at liquid nitrogen (LN2) temperature andshielding it from potential ice contamination. The imple-mentation and tests of the correlative approach were per-formed with fluorescently labelled adherent cell culturesgrown on EM grids, but essentially the same approachcan also be used with cryo-sections deposited on the grids.In addition to information regarding the location of fea-tures of interest, the LM examination of the grids in aphase contrast mode provides valuable information aboutlocal variations in ice thickness, ice crystal contaminationor other defects which could affect data quality.

2. Materials and methods

2.1. Cryo-holder for cryo-FLM: design and experimental

setup

To enable the imaging of a vitrified hydrated and fluo-rescently labelled biological specimen on EM grids withLM we designed and constructed a cryo-holder adaptedto the motorised stage of an inverted epifluorescentmicroscope.

A schematic overview of the cryo-holder is given inFig. 1. The cryo-holder comprises two main parts: an iso-lation box adapted to the motorised stage of the opticalmicroscope and a large metal block designed as a holderfor the vitrified grid. The metal holder has to be madeout of a highly thermal conductive material, like copperor brass, to ensure thermal stability. In our case, the com-plete metal assembly is made from brass, which has a spe-cific heat capacity comparable to that of copper but can bemore easily machined. This brass-metal holder consists of a

sealed LN2 reservoir (Fig. 1A(a)) equipped with two out-lets, one for the LN2 refill (Fig. 1A(b)) and the secondone for cold N2 vapour release (Fig. 1A(c)). An insertwithin the reservoir is designed to place EM-grid storageboxes for the transfer of samples between LM and EM(Fig. 1A(d)). The grid is positioned in the central part ofthe metal holder and is clamped with a 1 mm thick brassring (Fig. 1A(f)) with a central hole of 2.5 mm diameter.To further isolate the vitrified grid from the ambient atmo-sphere during the investigation, the hole of the ring is cov-ered with a conventional glass coverslide of 0.17 mmthickness (7 mm in diameter).

The metal part of the holder is positioned in a plasticbox, effectively insulated from the ambient atmosphere bya thermoplastic layer similar to conventional Styrofoam(Pomalux Acetal, Westlake Plastic Company, Lenni, Penn-sylvania, USA) (Fig. 1A(e)). The whole box is covered with

A. Sartori et al. / Journal of Structural Biology 160 (2007) 135–145 137

a lid with a central opening for the insertion of the con-denser. In order to protect the vitrified sample from detri-mental thermal gradients along the light path, a ‘dry-air’slab-acting as a isolation layer-is placed between the gridand the air objective. It is made of a plastic ring (6 mmthick, 1.5 cm outer diameter, 0.8 cm inner diameter), sealedwith two conventional glass coverslides (0.17 mm in thick-ness) enclosing dry-air (see enlarged inset in Fig. 1A).While the upper glass slide is in direct contact with the coldmetal holder (0.5 mm below the vitrified grid) shielded bythe ‘dry-air’ insulation, the outside surface of the lowerglass slide is exposed to the normal working environment,e.g. room temperature and 60–80% of humidity. In order toprevent condensation and the formation of frost, it is con-stantly flushed with nitrogen gas at room temperature andit is protected from the humid ambient atmosphere via aninsulating layer placed around the air objective. The insula-tion of the frozen sample implies the use of long working-distance (LWD) air objectives.

The design of the cryo-box in its dimensions and featuresallows an easy transfer of frozen-hydrated specimens on EMgrids. For mounting grids into the FLM cryo-holder, themetal part of the holder has to be first cooled down toLN2-temperature. A grid-storage container is transferredto the metal block and one grid at a time can be placed atthe designated imaging position. After closing the box withthe protection cover, the complete assembly is then trans-ferred to the stage of the light microscope for an initialscreening. The vitrified grid is kept within a large metal blockin equilibrium with a LN2 reservoir in a cold N2 atmosphereand imaged with no LN2 present in the optical path. The LN2

reservoir filled with LN2 keeps the temperature stable forapproximately 30 min and can then be manually refilled,extending the investigation time up to 2 h.

2.2. Grid scanning automation

The systematic search for areas of interest and the sub-sequent correlation with the transmission electron micro-

Fig. 2. Grid scan software (‘tom-grid-scan’) within the TOM toolbox. (A) Seelectron micrographs and the field of view is 1.96 mm2 at a final magnification

scope requires automated screening and image acquisitionboth at the light and at the EM level. The automation ofscreening processes at the light microscopy level is wellestablished and several software solutions are available,some of which from the manufacturers (e.g. Olympus,Zeiss, etc.). We scan the sample with FLM at low magnifi-cation (10·) before freezing and with cryo-FLM after freez-ing at higher magnification (20–40·), by using the scanningand z-stack modules of the light microscope acquisitionsoftware (see section 2.4). Z-stacks of images with 7, 5and 3 lm increment for 10, 20 and 40· magnification,respectively, are acquired at every imaging position. Incryo-FLM the scanning and z-stacks of the grid areobtained with a LWD 20· objective with 0.4 numericalaperture (N.A.), while higher resolution z-stacks areacquired in selected areas of interest with a LWD 40·objective (0.55 N.A.).

Once the frozen grid has been scanned with cryo-FLM,it is transferred to the transmission electron microscope inan EM cryo-holder and automatically scanned at low mag-nification (typically 300·, see Fig. 2). By using softwareroutines that we developed within the scientific computingplatform Matlab (The MathWorks, Natick, USA) largeareas (�2 mm2) of vitrified EM grids can be scanned andthe coordinates of numerous locations can be stored andretrieved at the EM level. These routines are part of theTOM toolbox software package which is dedicated to theacquisition and analysis of tomographic tilt-series (Nickellet al., 2005). The main procedure called ‘tom-scan-grid’(Nickell et al., 2007) controls the stage and the imageacquisition to generate an overview scan of the EM grid.Suitable positions or even areas identified in the overviewmap can be stored consecutively and if necessary a secondscan can be acquired at a higher magnification aroundthese positions or within specified areas.

The determination of the positions of the fluorescentlylabelled structures on the grid at the LM level and theirretrieval in the EM is facilitated by the use of commerciallyavailable EM indexed grids, so-called ‘finder’ grids, marked

tup interface and (B) a scanned image. The image represents �200 singleof �300· (scale bar: 200 lm).


with numbers and letters. The use of a regularly-patternedholey carbon coating, e.g. Quantifoil coating with holes of2 lm diameter (see Section 2.3), further facilitates the co-localisation procedure. There are basically two proceduresfor relocating in the EM the positions identified by cryo-FLM. Larger structures can be easily retrieved manuallyby a visual comparison of the cryo-FLM scan with theoverview scan at the EM level. For smaller structures threereference points can be chosen, e.g. the corners of an iden-tified square, and with a coordinate transformation themarked positions on the cryo-FLM scan can be retrievedin the EM map.

2.3. Cell cultures on EM grids and fluorescent labelling

Two types of adherent eukaryotic cells-NG108-15 neu-roblastoma/glioma hybrid cell line and keratinocytes-werecultured in vitro on gold EM indexed grids. The gold gridsNH2A (200 mesh, 127 lm pitch) and NHF15A (135 mesh,188 lm pitch) are indexed with letters and numbers andhave a standard outer diameter of 3.05 mm (Plano, Wetz-lar, Germany). They are coated with a 10 nm layer of per-forated amorphous carbon (Quantifoil, Jena, Germany).These grids were sterilised in 70% EtOH for 5 min. In orderto enhance the attachment and growth of the cells on thecarbon coated grid, the carbon layer was coated for 1 hwith 5 mg/cm2 fibronectin (Tebu-bio, Offenbach, Ger-many) in the case of keratinocytes or with laminin(Sigma–Aldrich, Munich, Germany) in the case ofNG108-15 cells. The NG108-15 neuroblastoma/gliomahybrid cells (courtesy of Prof. J. Kas, University of Leipzig,Germany) transfected with eGFP-Actin (vector from Prof.B. Wehrle-Haller, Centre Medical Universitaire, Geneva,Switzerland) were cultured in DMEM medium with theaddition of 10% FCS, 5000 U/ml Penicillin and 100 lg/ml Streptomycin antibiotics and 500 lg/ml GeneticinG418 (all chemicals are from Sigma). They were platedon laminin coated indexed grids and cultured for up to48 h to allow neurite development. Normal human epider-mal keratinocytes from neonatal foreskin (Cambrex BioScience, Verviers, Belgium) were grown at 37 �C and 5%CO2 in complete keratinocyte basal medium (KGM)(Cambrex Bio Science). Cells were harvested for sub-cul-turing using 0.025% trypsin and 0.01% EDTA in HBSS(Cambrex Bio Science) only between passages 2–4. To ini-tiate keratinocytes migration, 50 nM EGF was added tothe culture medium. Keratinocytes were plated on fibronec-tin coated indexed grids and grown for 24 h in KBM med-ium with 50 nM EGF.

Live labelling of the whole cell body of NG108-15 cellswas done by using the membrane permeable calcium dyeFura-2 acetoxymethylester ester (Fura-2 AM, Biotrend,Cologne, Germany, 340–380/510 nm excitation/emission),which can be introduced into the cells via incubation. Onceinside the cell, these esters are cleaved by intracellular ester-ases to yield cell-impermeant blue fluorescent indicators.NG108-15 cells—grown on the laminin coated indexed

EM grids—were loaded with Fura-2 by incubation with5 lM Fura-2 and 0.1% Pluronic F127 at 37 �C and 5%CO2 in their medium for 40 min followed by a 5 min washin medium at 37 �C. For live immunolabelling of b1 inte-grins, keratinocyte cultures we used monoclonal antibodiesagainst human b1 integrin (CD29 Clone JB1A, Chemicon,Temecula, USA). After blocking with 1% BSA in KGM for15 min at 37 �C and brief washing in PBS at 37 �C and 5%CO2, the cells were incubated for 20–25 min with primaryantibodies diluted 1:100 in 1% BSA in KGM at 37 �Cand 5% CO2. After a brief washing with PBS, the cells wereincubated with the secondary Alexa 488-conjugated goatanti-mouse antibodies (Invitrogen, Karlsruhe, Germany),diluted 1:100–1:200 in 1% BSA in KGM for 20–25 min at37 �C and 5% CO2 and subsequently briefly washed beforeimaging.

2.4. Optical microscopy: Image acquisition and analysis at

room temperature and under cryo-conditions

All light microscopy investigations were performedusing a Zeiss Axiovert 200 M inverted epifluorescencemicroscope (Carl Zeiss, Oberkochen, Germany). Themicroscope is fully motorised (stage DC 120 · 100 withmotorised stage control MCU 28, Marzhauser, Wetzlar,Germany) and equipped with a halogen lamp (100 HAL,12 V, 100 Watt) for transmitted light microscopy and anepifluorescence system with a HBO 103 Mercury Bulb.The DAPI (EX BP 365/EM LP 297) and the FITC (EXBP 450–490/EM BP 515–565) filtersets were used to detectFura-2 and Alexa488 labelled molecules, respectively.Transmitted and fluorescence light images were acquiredwith a cooled CCD black and white camera (AxioCamHRm, 1.4 Megapixels, Zeiss), with Zeiss AxioVision 4.5software, controlling the image recording and the micro-scope stage. The grid scans were acquired using the Mosaixand the z-stack modules of the Axiovision software. Z-stacks of images (with 7, 5 and 3 lm increment for 10, 20and 40· magnification, respectively) are acquired at everyimaging position to account for the partial bending ofthe grid due to manual handlings and to ensure correctfocusing on each portion of the labelled frozen sample.The room temperature images for the scan of the fluores-cently labelled samples on the EM grids were acquiredusing a Zeiss 10· Plan-Neofluar Ph1 air objective with0.3 N.A., while the EM grids were positioned in glass-bot-tom dishes (MatTek corporation, Ashland, USA) in thecells’ medium. Under cryo-conditions, the images wereacquired with a Zeiss 10· Plan-Neofluar with 0.3 N.A., aZeiss 20· LD Achroplan Ph2 with 0.4 N.A. and an Olym-pus (Olympus Europa GmbH, Hamburg, Germany) 40·SLCPlanFL Ph2 with 0.55 N.A. and 7.7–8.3 mm WD.

2.5. Cryo-EM and cryo-ET and image processing

Images and tomographic tilt series were recorded using aTecnai F30 Polara transmission electron microscope (FEI

Fig. 3. Scheme for the correlative microscopy cycle at cryo temperaturesapplied to NG108-15 neuronal cells. (A) Light microscopy: growth of cellcultures on EM marker grids, fluorescent labelling with Fura-2, phasecontrast and fluorescent imaging at 37 �C. The yellow square points at thecorrelative area, a labelled cell neurite. (B) Cryo-light microscopy:embedding of the sample in vitreous ice, cryo-phase contrast and cryo-fluorescent imaging (at �196 �C) in the cryo-holder device and identifi-cation of the labelled neurite. (C) Cryo-electron microscopy: transfer ofvitrified grid and recovery of the coordinates of the area of interest in theEM. Low (3500·) and high (27500·) magnification micrographs of aneurite region. (D) Cryo-electron tomography of the targeted area. The3D tomographic reconstruction clearly reveals the cytoskeletal organisa-tion with the complex interplay between microtubules and a dense actinnetwork. These features are barely visible in the corresponding 2D EMimage (scale bars: (A) and (B) 50 lm; (C) lower image 1 lm; (C) upperimage and (D) lower image 350 nm; (D) upper image 100 nm).


Company, Eindhoven, The Netherlands) equipped with afield emission gun and operated at an acceleration voltageof 300 kV. The frozen grids were transferred into cartridgesof the liquid nitrogen cooled multispecimen-holder, andinserted into the microscope. To minimise the electron doseapplied to the ice-embedded specimens, data were recordedunder low-dose conditions using automated data acquisi-tion software (TOM toolbox and Explore3D by FEI).The total dose accumulated during the tilt series was keptat 670 electrons per A2. The microscope is equipped witha Gatan post-column energy filter (GIF 2002) operated inthe zero-energy-loss mode with a slit width of 20 eV. Toaccount for the increase of specimen thickness at high-tiltangles, the exposure time was multiplied by a factor of 1/cosh (where h = tilt angle). The recording device was a2048 · 2048-pixel CCD camera (Gatan). The pixel size inunbinned images was 0.8 nm at a final magnification of�37000·. Images were recorded at 16 lm defocus forenhanced contrast. All 2D projection images of a tilt serieswere aligned with respect to a common origin by using10 nm colloidal gold particles as fiducial markers. 3Dreconstructions were calculated by weighted back-projec-tion. All image processing was done using MatLab7 (TheMath Works Inc., Natick, MA) using the TOM toolbox(Nickell et al., 2005).

3. Results

3.1. Cryo-FLM: performance and resolution

The key element of the cryo-FLM technique is a deviceallowing the imaging the frozen cells with transmitted andfluorescent light while keeping them in a vitrified state.Once the position of the labelled structures of interest havebeen recovered in the EM, one can zoom into them athigher magnification, check the ice quality and select thecorrelative areas suitable for cryo-EM. Finally, it is possi-ble to acquire tomograms in order to obtain 3D imagesof structures identified by fluorescent labelling (seeFig. 3). However, the preservation of the vitrified sampleat LN2 temperature and the strict insulation requirementsimpose a major constraint on the experimental setup andon the attainable resolution. First, cryo-FLM can not beperformed in combination with high resolution oil or waterimmersion objectives—with high N.A.—since at LN2 tem-perature the immersion media would freeze and it thereforerequires the use of LWD air objectives. Our current setupincludes LWD air objectives with magnifications up to40· (N.A. 60.55), featuring a maximal lateral resolutionR � 500 nm, with R � 0.61 k/N.A., where k is the wave-length of the incident light. In practice, the lateral resolu-tion in cryo-FLM is reduced to �1 lm by the opticalaberrations generated by the presence of two glass slidesin the optical path and by the high scattering of the lightin the frozen medium, resulting in blurring effects andbackground noise. In order to validate our estimate, weshow in Fig. 4 images of calibrated fluorescent beads of

1 lm size obtained at room temperature (dried beads ona coverslip) and at cryo temperatures (beads embedded invitreous ice) with a 40x air long working distance objective(0.55 N.A.). We also include the corresponding cryo-phasecontrast image as a reference (Fig. 4C). The estimatedthickness of the ice in the areas around the beads is�1 lm. Both the fluorescent image at room temperature(Fig. 4A) and the one under cryo-conditions (Fig. 4B) wereacquired at 60 ms exposure time. The lateral resolution ofthe 1 lm beads (estimated from the width at mid-heightof the average density profiles through the beads inFig. 4D) is �1.1 lm both under cryo-conditions and atroom temperature. However, under cryo-conditions weobserve a spreading (blurring) of the signal as a result ofthe optical aberrations. These optical aberrations are gen-erated by the presence of two glass slides in the optical pathon the one hand and to the higher scattering of the light invitreous ice as compared to standard imaging conditionson the other hand. The latter is due to ice inhomogeneities,resulting for instance from partial vitrification of thick

Fig. 4. Comparison between images of calibrated fluorescent beads of 1 l size obtained (A) at room temperature (dried beads on a coverslip) and (B)under cryo-conditions (beads embedded in vitreous ice) with a 40· air long working distance objective (0.55 N.A.), at 60 ms exposure time. The contrast inthe fluorescence images is inverted for a clearer visualisation of the beads and of the optical aberrations. (C) Corresponding cryo-phase contrast image. (D)Average density profiles of the beads, normalised by the data standard deviation. The lateral resolution of the fluorescent signal (estimated from the widthat mid-height of the density profiles indicated by the arrows) is �1.1 lm both under cryo-conditions (black line) and at room temperature (grey line). Theoptical aberrations induced by the presence of two glass slides in the optical path and by the high scattering of the light in the frozen medium result in aweak spreading (blurring) of the signal (scale bars: 20 lm).


regions, giving rise to a variable refractive index in the fro-zen sample.

A major advantage of FLM investigations at cryo-tem-peratures is the reduction of photobleaching (the loss of

Fig. 5. Correlative cycle applied to keratinocyte cells migrating on EM marAlexa488). Co-localisation of fluorescently labelled features (back retractingfluorescent scan of grid at 37 �C. (C) Low magnification (300·) scan of the vCryo-phase contrast and (E) cryo-fluorescent 20· scan of grid after plunge-frsurface, can be easily identified with cryo-phase contrast imaging. (F) 3micrograph of retracting fibres in the correlative area (indicated by the whitarrangement of actin filaments in bundles is clearly visible. The red arrow(F) 100 nm; (G) 500 nm).

fluorescence upon irradiation) by a factor of 10–50 times,depending on the excitation light and on the nature ofthe fluorescent probe; this observation is in accordancewith reports by Zondervan et al. (2004). Thus longer expo-

ker grid, immunolabelled against b1-integrins (antibodies conjugated tofibres) in FLM, cryo-FLM and cryo-EM. (A) Phase contrast and (B)

itrified grid and retrieval of the labelled area of interest in cryo-EM. (D)eezing. Large cell bodies, that determine the main morphology of the iceD reconstruction (x–y slice along z-axis) and (G) corresponding 2De arrows) at the rear of the migrating keratinocyte. The regular parallels indicate the contamination by thick ice (scale bars: (A)–(E) 80 lm;

Fig. 6. Assessment of the quality and intactness of the vitrified sample after plunge-freezing with cryo-FLM. (A) and (B) Cryo-transmitted light imageand a cryo-fluorescence image of a portion of a grid with vitrified NG108-15 cells (live labelled with Fura-2). (C) and (D) Cryo-phase contrast of thesame portion of the grid and corresponding color-coded image, obtained by normalising the pixel values ‘i of image (C) with the average pixel value ‘o

of the transmitted light through an empty field of the grid (area 1). The fluorescent SNR in areas where 30% 6 ‘i/‘o 6 70% (ice thickness 61 lm) ishigh (areas 2 and 3), while in regions with thick ice where ‘i/‘o 6 30% and ‘i/‘o P 70% the signal is blurred (area 4). The contamination of the grid bythick ice is clearly detectable in the cryo-transmitted light image (A), areas 5 and 6, or in the cryo-phase contrast image (C), areas 5 and 6. (scale bars:(A)–(D) 50 lm).


sure times are applicable with a negligible decay of the fluo-rescent signal, leading to a detectable fluorescent SNR.

The imaging of the frozen sample with transmitted light(e.g. with phase contrast) does provide little if any morpho-logical information on the cells (see Fig. 5D, where the celledges can be barely identified) but it is required for thedetermination of the position of the labelled structures onthe indexed grid. Furthermore, phase contrast imagingallows to assess the quality and intactness of the vitrifiedsample after plunge-freezing. The contamination by thickice (see red arrows, Fig. 5D and E, and, area 5 and 6,Fig. 6) and the disruptions of the carbon film (Fig. 5Dand Fig. 6, area 1) can be easily visualised. Additionally,an estimate of the ice thickness can be obtained by compar-ing the light transmitted through the ice (‘i = greyscalepixel values, see Fig. 6C) and the one transmitted throughempty areas on the grid (‘o; e.g. area 1, Fig. 6C). In regionswhere 30% 6 ‘i/‘o 6 70% the ice thickness can be estimatedto be smaller than 1 lm (see color-coded image Fig. 6D)

and is therefore potentially suitable for subsequent investi-gations with cryo-EM. In these areas, the fluorescent SNRis high (areas 2 and 3, Fig. 6) and comparable to room tem-perature observations, while in areas with thick ice (‘i/‘o 6 30% and ‘i/‘o P 70%) the signal is blurred (area 4,Fig. 6).

Further information about the preservation of the sam-ple after blotting can be obtained by comparing the fluores-cent scans of the grid obtained before and after freezing. Avisual comparison highlights dislocations, disruptions orremoval of parts of the fluorescently labelled sample.

The lateral uncertainty in the co-localisation of an areaof interest in cryo-FLM and cryo-EM can be estimated as

D �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiD2

FLM þ D2EM

q� 2:3 lm, where DFLM � 1 lm and

DEM � 2 lm are the localisation errors in cryo-FLM andin cryo-EM. DFLM accounts for the lateral uncertainty inthe position determination on the cryo-FLM images;DEM includes the error in the determination of the position

Fig. 7. Correlative cycle applied to NG108-15 neuronal cell line, labelled with Fura-2, membrane permeant calcium indicator highlighting neurites. (A)Low magnification fluorescent scan of grid at 37 �C and (B) at cryo-temperatures in the cryo-holder device and identification of a suitable area for cryo-ET. (C) Corresponding low magnification (300·) scan of a vitrified grid in the EM. (D) Low magnification (9000·) cryo-EM micrograph of the targetedarea on the neurite. (E) High magnification (27500·) 2D micrograph and (F) corresponding 3D-reconstruction (x–y slice along z-axis) of a neuriteextension. The white insets indicate the correlative area, a fluorescently labelled neurite; its details are shown in the central fluorescent images. The whitearrows point at the correlative area (scale bars: (a)–(c) 200 lm; inset in (A), (B) and (C) 50 lm; (D) 2 lm; (E) and (F) 100 nm).


of the corners of a square on the cryo-EM scan due to thethick ice covering the square edges and to the possible localbending of the grid. In practice, by using Quantifoil car-bon-coated grids, one can take advantage of the regulartopology of the holes distribution and increase the accu-racy of the co-localisation.

3.2. Application to eukaryotic cells

The feasibility of our correlative approach was testedwith the NG108-15 neuroblastoma/glioma hybrid cells(see Fig. 3 and Fig. 7) and on primary keratinocyte culturesto study cytoskeletal organisation in neurites and inretracting fibres (see Fig. 5 and Fig. 8), respectively.NG108-15 cells and primary keratinocytes are ideal sys-tems to investigate cytoskeletal organisation due to theirunique morphology with long and flat cellular extensions.The cells were cultured directly on EM grids as describedin Section 2.3.

The model neuronal cell line NG108-15, which is usedto study innervation of striated myotubes, is known topossess long neurites forming pre-synaptic-like neuronal

varicosities upon differentiation (Chen et al., 2001; Betzet al., 2006). We fluorescently labelled the cells with amembrane-permeant calcium indicator (Fura-2 AM Ester)to highlight and target intact neurites. Fig. 3 shows acomplete correlative cycle applied to labelled NG108-15cells: a fluorescently labelled neurite was localised andidentified both in the fluorescent (Fig. 3A) and in thecryo-fluorescent scan (Fig. 3B), and finally a tomogramwas recorded of a fragment of the neurite. The tomo-graphic slice clearly reveals the cytoskeletal organisationwith the complex interplay between microtubules andthe dense actin network (Fig. 3D). These features arebarely distinguishable in the corresponding 2D EM image(Fig. 3C). In Fig. 7, we show another example of correl-ative cycle for NG108-15 cells. The low magnificationFLM scan of the grid in Fig. 7A and the cryo-FLM scanin Fig. 7B can be directly compared with the low magni-fication cryo-EM scan of Fig.7C. As it can be clearly seen,in thick areas (>1 lm) of the sample—showing little or notransmission in cryo-EM—or in areas containing thick icethe fluorescent signal is strongly blurred (see for instancethe top of Fig. 7B).

Fig. 8. (A) and (B) Cryo-ET of selective areas (C) and (E) on keratinocytes obtained by correlative microscopy. (D) Cryo-fluorescent image (b1-integrinlive immunolabelling). The 3D reconstructions within the areas indicated by the yellow squares reveal differences in the organisation of the cytoskeleton atthe front ((C), an intermediate filament network is clearly visible) and rear region ((E), a filopodium extension shows a thick actin bundle originating froma dense cortical actin network) of the migrating cell (scale bars: (A) and (B) 2 lm; (C) and (E) 100 nm; (D) 50 lm).


In the case of keratinocyte cultures, retracting fiberswere highlighted by performing live immunolabellingagainst b1 integrins with antibodies conjugated to a fluo-rescent dye (Alexa488, see section 2.3) (Rigort et al.,2004) in order to target the cytoskeletal organisation atperipheral areas of the cell and in retracting fibres, wherethese adhesion proteins are known to be abundant (seeFig. 5B and e and Fig. 8D). Migrating keratinocytes sharea characteristic polarity with broad and elongated cyto-plasmic protrusions at the cell front, called lamellipodia,and thin cellular extensions, the retracting fibres, at the cellrear. The cells are connected to the underlying substrate bycell adhesions which are discrete transmembrane regions inwhich cytoskeletal components, either microfilaments orintermediate filaments are tightly linked via integrins (e.g.

b1 integrins) to proteins of the extracellular matrix(Fig. 5 and Fig. 8). In Fig. 5 we show the co-localisationof fluorescently labelled retracting fibres in FLM (Fig. 5Aand B), cryo-FLM (Fig. 5D and E) and cryo-EM(Fig. 5C and G) and their higher 3D structural investiga-tion by cryo-ET on the correlative area (Fig. 5F). Thehighly regular parallel arrangement of actin filaments inbundles within the retracting fibres is clearly revealed inthe tomographic slice in Fig. 5F. Note that in this examplelarge cell bodies that determine the main morphology of

the ice surface can be easily identified in cryo-phase con-trast images (Fig. 5D). In the example of Fig. 8, by apply-ing our correlative approach at cryo-temperatures, wesucceeded in targeting and imaging with 3D cryo-ET thedifferent cytoskeletal organisation at the front (Fig. 8Aand its inset) and the rear end (Fig. 8B and its inset) of amigrating keratinocyte cell.

An advanced version of our correlative approach atcryo-temperatures, featuring an improved coordinateretrieval procedure, has been applied to primary matureneuronal cultures (15 days division) and has enabled usto target and to structurally characterise active synapsesand pools of pre-synaptic vesicles in neurons preserved intheir intact state (Lucic et al., 2007).

4. Discussion

In this communication we describe a new correlativemicroscopy approach, that aims at improving the localisa-tion and identification of fluorescently labelled features ofinterest on frozen hydrated biological samples, prior totheir in-depth investigation by cryo-EM. We have built acryo-holder that keeps the vitrified sample at LN2 tempera-ture and protects it from contamination and detrimentalthermal gradients which enables us to image the sample


by transmission and fluorescence light microscopy. A majoradvantage of this new correlative approach is the auto-mated data acquisition and transfer of coordinates betweenthe light and the transmission electron microscope, usingMatLab-based software. The vitrified sample is scannedboth with cryo-FLM and with cryo-EM and the identifiedfluorescently labelled structures of interest can be addressedwith a lateral resolution of �1 lm at the FLM level andwith co-localisation precision of �2 lm in the electronmicroscope. This guided search minimises radiation dam-age to the sample and avoids some of the major inconve-niences related to cryo-EM and cryo-ET investigations:the ’’blind’’ search for features of interest within the lowcontrast of biological materials in amorphous ice and thepositive identification of the imaged structures. Besides,observing the sample with transmitted light after vitrifica-tion allows to assess the degree of contamination by thickice and to estimate whether the ice thickness is suitablefor cryo-EM investigations (i.e. 61 lm).

We aim at improving the resolution of our system, onthe one hand, by making the current cryo-holder designsuitable for air objectives with higher magnification (63·)and N.A. (0.75)—and thus higher resolution—and, onthe other hand, by combining cryo-FLM with techniquesthat reduce the background noise. The SNR of the imagescould be enhanced by replacing conventional fluorescencemicroscopy with laser scanning confocal microscopy(LSCM) and by post-processing the images using deconvo-lution methods (Larson, 2002). In LSCM, the sample isscanned point by point on each focal plane, so that the sig-nal coming from the ‘‘out-of-focus’’ areas of the sampledoes not contribute to the generation of background noisein the images. This technique is thus particularly suited forvitrified samples which are often bent due to manual han-dling of the delicate grids. Moreover, by imaging undercryo-conditions, the recurrent LSCM problem of fast fluo-rescent bleaching caused by point-scanning the sample withthe laser beam is mitigated.

The attainable resolution in cryo-FLM sets a lowerbound to the co-localisation accuracy of fluorescentlylabelled features in cryo-FLM and cryo-EM. This limitcould be overcome by using labels that are detectable bothwith fluorescence and with electron microscopy-like quan-tum dots and fluorescent colloidal gold conjugates. Theseprobes are perfectly suited for correlative microscopy andmolecular co-localisation. Quantum dots nanocrystals arecurrently increasingly used for live cell imaging (Michaletet al., 2005; Bruchez et al., 1998; Parak et al., 2005), dueto their narrow, size-tunable emission spectra, high quan-tum yield and resistance to photobleaching (Hines andGuyot-Sionnest, 1996) and have opened up new possibili-ties of correlating light and electron microscopy (Nismanet al., 2004; Grabenbauer et al., 2005; Giepmans et al.,2005). But it remains a major challenge to introduce theminto living cells (Duan and Nie, 2007) and, at the sametime, target the molecules of interest with adequatespecificity.

Acknowledgments

Timo Betz for useful discussions about the NG108-15cell line. This work was supported by the SFB 563 andby the EU within the Network of Excellence NoE-3DEM.

References

Afzelius, B.A., Maunsbach, A.B., 2004. Biological ultrastructuralresearch; the first 50 years. Tissue Cell 36, 83–94.

Belzer, E.O., Southard, J.H., 1998. Principles of solid organ transplan-tation by cold storage. Transplantation 45, 673–676.

Betz, T., Lim, D., Kas, J.A., 2006. Neuronal growth: a bistable stochasticprocess. Phys. Rev. Lett. 96, 098–103. Available online.

Bruchez, M., Moronne, M., Gin, P., Weiss, S., Alivisatos, A.P., 1998.Semiconductor nanocrystals as fluorescent biological labels. Science281, 2013–2016.

Chen, X.-L., Zhong, Z.-G., Yokohama, S., Bark, C., Meister, B.,Berggren, P.-O., Roder, J., Higashida, H., Jeronim, A., 2001.Overexpression of rat neuronal calcium sensor-1 in rodent NG108-15cells enhances synapse formation and transmission. J. Physiol. 532,649–659.

Cosman, M.D., Toner, M., Kandel, J., Cravalho, E.G., 1989. Anintegrated cryomicroscopy system. Cryo-Lett. 10, 17–38.

Duan, H., Nie, S., 2007. Cell-penetrating quantum dots based onmultivalent and endosome-disrupting surface coating. J. Am. Chem.Soc. 129, 3333–3338.

Echling, P., 1992. Low Temperature Microscopy and Analysis. PlenumPress.

Giepmans, B.N.G., Deerinck, T.J., Smarr, B.L., Jones, Y.Z., Ellisman,M.H., 2005. Correlated light and electron microscopy imaging ofmultiple endogenous proteins using quantum dots. Nat. Meth. 2, 743–749.

Grabenbauer, M., Geerts, W.J.C., Fernandez-Rodriguez, J., Hoenger, A.,Koster, A.J., Nilsson, T., 2005. Correlative microscopy and electrontomography of GFP through photooxidation. Nat. Meth. 2, 857–862.

Hines, M.A., Guyot-Sionnest, P., 1996. Synthesis and characterization ofstrongly luminescing ZnS-capped CdSe nanocrystals. J. Phys. Chem.100, 468–471.

Irimia, D., Karlsson, J.O.M., 2005. Kinetics of intracelluar ice formationin one-dimensional arrays of interacting biological cells. Biophys. J. 88,647–660.

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

Larson, J.M., 2002. 2d and 3d deconvolution of confocal fluorescentimages by maximum likelihood estimation. Proc. SPIE 2641, 86–94.

Lucic, V., Kossel, A.H., Yang, T., Bonhoeffer, T., Baumeister, W.,Sartori, A., 2007. Multiscale imaging of neurons grown in cultures:from light microscopy to cryo-electron tomography. J. Struct. Biol.160, 146–156.

Medalia, O., Weber, I., Nicastro, A.S., Gerisch, G., Baumeister, W., 2002.Macromolecular architecture in eukaryotic cells visualized by cryo-electron tomography. Science 298, 1209–1213.

Michalet, X., Pinaud, F.F., Bentolila, L.A., Tsay, J.M., Doose, S., Li, J.J.,Sundaresan, G., Wu, A.M., Gambhir, S.S., Weiss, S., 2005. Quantumdots for live cells, in vivo imaging, and diagnostics. Science 307, 538–544.

Namperumal, R., Coger, R., 1998. A new cryostage design for cryomi-croscopy. J. Microscopy 192, 202–211.

Nickell, S., Beck, F., Korinek, A., Mihalache, O., Baumeister, W., Plitzko,J.M., 2007. Automated cryoelectron microscopy of ‘‘single particles’’applied to the 26s proteasome. FEBS Lett. 581, 2751–2756.

Nickell, S., Forster, F., Linaroudis, A., Net, W.D., Beck, F., Hegerl, R.,Baumeister, W., Plitzko, J.M., 2005. Tom software toolbox: acquisitionand analysis for electron tomography. J. Struct. Biol. 149, 227–234.


Nisman, R., Dellaire, G., Ren, Y., Li, R., Bazett-Jones, D.P., 2004.Application of quantum dots as probes for correlative fluorescence,conventional and energy-filtered electron microscopy. J. Histochem.Cystochem. 52, 13–18.

Parak, W.J., Pellegrino, T., Plank, C., 2005. Labelling of cells withquantum dots. Nanotechnology 16, R9–R25.

Rigort, A., Grunewald, J., Herzog, V., Kirfel, G., 2004. Release of integrinmacroaggregates as a mechanism of rear detachment during keratino-cyte migration. Eur. J. Cell Biol. 83, 725–733.

Zondervan, R., Kulzer, F., Kolchenko, M.A., Orrit, M., 2004. Photoble-aching of rhodamine 6 g in poly(vinyl alcohol) at the ensemble andsingle-molecule levels. J. Phys. Chem. A 108, 1657–1665.

correlative microscopy: bridging the gap between ﬂuorescence...

Documents