Imaging single living cells with a scanning near-®eld infraredmicroscope based on a free electron laser

M.K. Hong a, A.G. Jeung b, N.V. Dokholyan a, T.I. Smith b, H.A. Schwettman b,P. Huie c, S. Erramilli a,*

a Physics Department and the Center for Photonics, Boston University, MA, USAb Hansen Experimental Physics Laboratory, Stanford University, Stanford, USAc Department of Pathology, Stanford University Medical School, Stanford, USA

Abstract

We report the ®rst sub-wavelength mid-infrared images under water, and describe an application to obtaining images

of single living cells in water using the Stanford Free Electron Laser (FEL). Spatial resolution is enhanced at the peak of

infrared absorption of water. Images were obtained of single motile ®broblasts with the FEL wavelength tuned to

absorption peaks of both protein and lipid molecules. Analysis of the unexpectedly strong absorption due to lipid

molecules in motile ®broblasts suggests that the concentration of lipid molecules in lamellopodia is consistent with

membrane ¯ow. Ó 1998 Elsevier Science B.V. All rights reserved.

PACS: 87.64t; 41.60Cr; 87.64Je; 87.15Mi

1. Introduction

Infrared spectroscopy is one of the most sensi-tive techniques for the analysis of biological sys-tems. Many biomolecules like lipids, proteins andnucleic acids have normal modes that are infraredactive [1]. These normal modes are so character-istic that the region of the spectrum between 4 and12 lm is often called the ``®ngerprint'' region (seeFig. 1). A combination of infrared spectroscopy

and microscopy provides a new tool for non-de-structive characterization of samples, and for lo-calization of speci®c molecules within a sample[2,3]. The ability to localize speci®c molecules ormolecular groups is useful in a wide range ofproblems in biological physics. The absorptioncross-sections associated with many vibrationalmodes are su�ciently high (between 10ÿ19±10ÿ16

cm2) that it is possible to identify the presence ofspeci®c molecules at the sub-picogram level with-out having to use ¯uorescent stains or radioactivelabels. Recognition of this sensitivity has led to thecommercial development of Fourier transforminfrared spectrometers (FTIR) attached to infra-red microscopes as a powerful method for micro-analysis in biomedical applications. These

Nuclear Instruments and Methods in Physics Research B 144 (1998) 246±255

* Corresponding author. Address: 590 Commonwealth Av-

enue, Boston, MA 02215, USA. Tel.: 617 353 1271; fax: 617 353

9393; e-mail: shyam@bu.edu.

0168-583X/98/$ ± see front matter Ó 1998 Elsevier Science B.V. All rights reserved.

PII: S 0 1 6 8 - 5 8 3 X ( 9 8 ) 0 0 3 1 1 - 5

microscopes use Cassegrainian objectives, whichhave reduced numerical apertures (generally 0.6 orless) compared to refractive objectives, but havethe advantage that the focal length is the samethroughout the infrared region of the spectrum.Spectra are usually collected from a small (�25lm ´ 25 lm ´ 5 lm) region of a dehydratedsample placed on an infrared microscope stage.The relatively small numerical apertures, com-bined with the low brightness of the broadbandGlobar source (�1200 K) used limits the spatialresolution to roughly 25 lm, well above the dif-

fraction limit. With the recent development of aninfrared beamline at the synchrotron light sourceat Brookhaven, Carr et al. have demonstrated thatit is possible to reach the di�raction limit in in-frared microscopy [4]. Images may be acquired byraster scanning a sample beneath the focussedbroadband infrared beam collecting a spectrum ateach pixel [5]. At a rate of several minutes for eachpixel, acquisition of a 100 ´ 100 pixel image of asample is time consuming. Infrared microspec-trometers coupled to infrared focal plane arraydetectors allow for parallel acquisition of spectrafrom every pixel in the image [6]. Because of theoptics and the low brightness of the Globarsources used, the spatial resolution of this methodhas not reached the di�raction limit, even on de-hydrated samples.

When water is added, the strong absorptionthroughout the infrared region of the spectrum(see Fig. 2) makes both spectrum and image ac-quisition in the infrared region very di�cult. Theattenuation length for liquid water in this wave-length range is so short (�0.8±10 lm) that in orderto be able to obtain images, the imaging opticshave to be positioned extremely close to the sur-face of the sample. Conventional far ®eld imagingtechniques using optical elements like lenses aredi�cult with such short path lengths. The di�-culties are acute enough that, until this year therehave been no published infrared images underwater in this spectral range.

Fig. 2. Infrared spectrum of liquid water. The relative transmission (arb. units) is shown as a function of frequency. The strong ab-

sorption near 1600 cmÿ1 coincides with the amide band region.

Fig. 1. An infrared spectrum of a cross-section of artheroscle-

rotic tissue, showing absorption bands due to the presence of

ester-linked acyl chains in lipid, amide bands in proteins, and

cholesterol [3]. The absorption lines serve as a ``®ngerprint'' for

localizing molecules.

M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255 247

A solution to the problem of short path lengthsis to position the source of light extremely close tothe sample in order to reduce the e�ects of ab-sorption in the surrounding medium, as illustratedin Fig. 3 [7]. Infrared radiation from the StanfordFree Electron Laser (FEL) is coupled to one endof a chalcogenide ®ber. The other end of the ®beris tapered to form a sub-wavelength aperture. Thesample is placed on a CaF2 window in the near-®eld of infrared radiation transmitted by the ta-pered ®ber [8,9]. The transmitted signal is detectedusing infrared optics positioned below the sub-strate [10]. The sample is scanned using piezo-electric transducers, and images are obtained onepixel at a time. The technique is called ScanningNear-®eld Infrared Microscopy (SNIM), an ex-tension to the infrared region of scanning near-®eld optical microscopy (SNOM [11] or NSOM[12]) under water [13]. Here we show that theproblem of underwater infrared imaging can beovercome with scanning probe techniques.

2. Methods

One feature of the near-®eld technique is thatthe tapered end of the ®ber is less than the cut-o�wavelength. SNIM probes have very low trans-mission factors, in the range 10ÿ4±10ÿ6. Attempt-ing to overcome the small transmission factor byusing high laser power, such as that provided bythe Stanford FEL, is not possible for imaging inair or vacuum because of thermal damage to theend of the ®ber. In conventional SNOM, in thevisible region of the spectrum, the laser power isusually kept below about 10 mW in order to avoiddestruction of the gold coating on the tapered endof the ®ber. Immersion of the tapered ®ber inliquid water allows us to use the full 2 W power ofthe Stanford FEL without destroying the tip. Forroutine imaging studies, and for initial alignment,a small portion (�10%) of the FEL beam waspicked o� using a beam splitter and used forSNIM. The SNIM could thus be operated in

Fig. 3. (Left) Schematic picture of the essential idea for obtaining underwater infrared images. (Right) Normalized transmitted signal

at an FEL wavelength of 6.05 lm, through a tapered SNIM probe as a function of the separation between the probe and the surface of

the substrate, immersed in liquid water and D2O. The range of separation over which a signal can be detected is about 6 lm in H2O

compared to about 30 lm in D2O, consistent with the absorption cross-sections of the two liquids at this wavelength.

248 M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255

parasitic mode for extended periods. For under-water imaging of single living cells, the beamsplitter was replaced by a mirror, and the fullpower of the FEL beam was available for near-®eld imaging. The laser power emerging from thetip into the water was estimated to be about 2 lW.The temperature rise, measured by using a ¯uo-rescent marker, is less than 2 K [14]. The use ofhigh laser power allows for faster data acquisition,limited only by the response of the positioningstage (15 Hz with feedback, 10 kHz withoutfeedback) with a signal-to-noise ratio of about 3%.Because of the pulse structure of the FEL sourceused, the sample stage used for scanning has to besynchronized [9] to the Stanford FEL pulses,which occur in macropulses that last 3 ms at a 20Hz repetition rate. Each macropulse consists of �1ps duration micropulses, repeating at 11.8 MHz[15]. A comparison of various sources of infraredradiation for SNIM is shown in Fig. 4.

As the distance between the tip and sample isdecreased, the signal increases exponentially(Fig. 3). Furthermore, when the signal was mea-sured through both H2O and D2O at an FELwavelength of 6.05 lm one can see a signi®cantdi�erence in the decay lengths due to the di�erentattenuation lengths for the two liquids. Spatialresolution was determined by depositing a thin(�80 nm) ®lm of Chromium on a calcium ¯uoride

substrate. The thickness of the metal ®lm wasmeasured with an Atomic Force Microscope [16].Fig. 5 shows a sequence of images obtained atvarying heights. The spatial resolution in near-®eldmicroscopy degrades as the tip-sample distance isincreased, with the best spatial resolution achievedunder water being k/3. When uncoated tips or tipswith damaged coating are used, spatial resolutionis also degraded because the signal collected isemitted not just from the aperture at the end, butfrom points higher up on the probe. The presenceof an absorbing medium like water prevents signalleaking from higher up the tip from reaching thedetector. It is therefore possible to obtain highresolution infrared images from tapered probesthat are not coated at all. This is particularly usefulfor biological systems where the metal on the tipcan a�ect living cells. Fig. 5 shows that a resolu-tion of 2 lm was obtained at a wavelength of 7.47lm. Fig. 5 also shows that spatial resolution is lostwhen water is removed, but is restored when wateris reintroduced. One interesting consequence ofusing uncoated tips is that the spatial resolution isexpected to be highest at the wavelengths that arestrongly absorbed by the surrounding medium.Fig. 6 shows an underwater image at two di�erentwavelengths of a pattern of gold lithographed ontoa silicon substrate. An image was ®rst obtained at6.25 lm, close to an absorption peak in liquid

Fig. 4. Comparison of various infrared sources for SNIM, using measured values of transmission losses in the tapered SNIM probes.

M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255 249

Fig

.5

.(T

op

):S

equ

ence

of

imag

eso

fa

ned

ge

form

edb

ya

120

nm

go

ldla

yer

on

asi

lico

nsu

bst

rate

,im

aged

intr

an

smis

sio

nin

wate

r,at

vary

ing

pro

be

hei

gh

tsz.

(Bo

tto

m

Lef

t):

Un

der

wa

ter

ima

ge

at

aw

av

elen

gth

of

7.4

7lm

,o

fan

80

nm

chro

miu

med

ge

on

aC

aF

2su

bst

rate

.(B

ott

om

Cen

ter)

:A

naly

sis

of

the

tran

smit

ted

sign

al

acr

oss

the

edg

e(h

ori

zon

tal

ax

isis

inp

ixel

s)y

ield

sa

spati

al

reso

luti

on

of

2lm

.(B

ott

om

Rig

ht)

:U

nd

erw

ate

rim

age

of

an

arr

ay

of

go

ldd

ots

on

the

surf

ace

of

sili

con

.D

uri

ng

the

sca

n,

wa

ter

wa

sre

mo

ved

,re

sult

ing

inlo

sso

fsp

ati

al

reso

luti

on

(cen

tral

wh

ite

patc

h).

Rei

ntr

od

uct

ion

of

wate

rre

sto

red

the

con

trast

an

dsp

ati

al

reso

luti

on

.

250 M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255

water (Fig. 2). The FEL wavelength was thenchanged to 5.14 lm, away from the peak and asecond image was obtained without changing thesample or SNIM setup. It can be seen that thespatial resolution in the second image, taken at ashorter wavelength, is worse than in the ®rst.

3. Infrared images of single living cells

Infrared images were obtained of single living®broblasts. Fibroblasts are cells that are involvedin a variety of repair processes in the body, such aswound healing. These cells secrete ®brinogen andcan further change into other cells that specialize inmaking collagen, or into adipocytes for storing fat.Fibroblasts serve as models for studying cell mo-tility [17]. Fig. 7 shows a phase contrast opticalimage of human ®broblasts. The cells have thinnedout and spread across the substrate, generating

very thin (�1 lm thick) extensions called ®lopodiaor lamellopodia. Cell motion occurs in the direc-tion of the extensions, as shown schematically inFig. 8. The use of a tapered probe to obtain imagesallows us to localize molecules to speci®c featuresof the cell, such as the lamellopodium. Infraredimages were obtained by tuning the FEL to boththe amide absorption bands, and to the lipid ab-sorption bands. Fig. 9 shows infrared images ofhuman ®broblasts, with contrast provided by theintrinsic amide absorption bands characteristic ofproteins. (Also shown for comparison are imagesof human hybridoma cells imaged at two wave-lengths around 6 lm). Images were obtained of the

Fig. 7. Phase contrast microscope images of human ®broblasts

[9].

Fig. 8. Schematic diagram of a crawling cell, with the lamello-

podium leading the cell body containing the nucleus. The dis-

tance between the cell surface and the probe, as well as the

thickness of the lamellopodium, have been exaggerated for

clarity.

Fig. 6. Underwater image of a metal pattern taken at two di�erent wavelengths. (Left) Image acquired at a wavelength of 6.25 lm.

(Right) Image acquired of the same sample with the same probe and probe-to-sample height, at a wavelength of 5.14 lm. Note that the

spatial resolution at the shorter wavelength is actually worse than the resolution at the absorption peak at water.

M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255 251

lamellopodia with the FEL tuned to 5.76 lm,where the absorption is dominated by lipid mole-cules. What is surprising is that strong contrast isobserved. The images in Fig. 9 were acquired withthe shear force mechanism turned o�, so that theheight of the tip above the substrate was main-tained. A ratio of the signal obtained through thelamellopodium with the signal obtained throughbu�er provides an estimate of the absorption of(5.6 � 0.5) ´ 10ÿ2 OD at this wavelength. The ab-sorption cross-section due to the C±O bond hasbeen measured in standard lipid quantitationstudies and is about 3 ´ 10ÿ19 cm2. Not all lipidmolecules exhibit this absorption band, and someproteins with carboxylic acid groups in the sidechains can also have some absorption cross-sec-tion, but these secondary contributions are gener-ally negligible. This band has been used forquantitating fatty acids, trigylycerides and ester-linked phospholipids. One small point is that theabsorption in this band is an estimate of the totalconcentration of acyl tails in lipid molecules. Mostphospholipids have two or more tails. The com-position of lipid molecules in the cell membrane

exhibits a complex pro®le, with combinations ofphospholipids, fatty acids and triglycerides. Untilmore is known about the detailed biochemistry andlipid composition in living cells, the estimates oflipid concentration based on absorption studiesmay vary by a factor of 2. Even with such a largepotential source of systematic error, the observedabsorption is between 250 and 500 times what onewould expect from a single membrane bilayer.

3.1. Potential artifacts

Perhaps the most important potential artifact isinherent in using the near-®eld technique forquantitative spectroscopy. The extinction coe�-cients measured using far-®eld FTIR techniquescannot be used in order to assess the concentrationof biomolecules, especially if there is signi®cantcoupling of longitudinal modes. In order to testwhether the unusual coupling inherent in near-®eldtechniques can lead to a several hundred-fold en-hancement in the absorption cross-section [18],sections of atherosclerotic tissue were taken using acryomicrotome, and the absorption was measured

Fig. 9. (Top): Images of senescent ®broblasts at di�erent wavelengths, and di�erent magni®cations. Images were acquired at 6.06 lm

(amide I), and the image at the right was obtained at 6.25 lm. (Bottom Left): Images of crawling ®broblasts at a wavelength of 6.26 lm

and at 6.45 lm (amide II), and a high resolution scan of the lamellopodium at the bottom, with the FEL tuned to the lipid peak at 5.76

lm, showing surprisingly high contrast. (Bottom Right): Image of a single human hybridoma cell at 6.06 lm, with the FEL tuned to

the amide I absorption band, and to 6.25 lm, showing the variation in contrast in a di�erent cell system.

252 M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255

using both near-®eld microscopy and FTIR studiesof lipid deposits. No evidence of such an unusualenhancement of the cross-section was observed,either in the dry state or in the presence of water.

4. Cell motility

There are several di�erent models of cell mo-tility. The so-called ``membrane ¯ow'' hypothesiswas ®rst postulated to explain the phenomenon ofcapping [19]. In this model, the cell membraneundergoes a cycle of polarized endocytosis, inwhich small pieces of the cell membrane in thetrailing edge of the cell break away to form smallvesicles. The resulting vesicles from this topologi-cal transformation are transported to the leadingedge of the motile cell, thought to be the site ofexocytosis. The alternative ``rake'' model [20],which is more widely accepted, suggests that po-lymerization of actin at the leading edge, and de-polymerization at the trailing edge, lead to a``pull'' on the membrane surface. The membrane istransported primarily along the cell surface, with-out any change in the topology of the cell. Actin isknown to undergo polarized cycles of polymer-ization. Fluorescent labels are often used in at-tempts to distinguish between these models, incombination with speci®c mutations or the e�ectsof drugs. There is interest in developing methodsthat do not require perturbing stains or labels ca-pable of testing predictions that the two modelshave provided. One di�erence between the twomodels is that in the ®rst case, the lamellopodiumcontains vesicles, whereas in the second model,relatively few vesicles are present in the cytoplasm.Because the presence of excess lipid molecules as-sociated with these vesicles is expected to lead toincreased absorption at wavelengths characteristicof the normal modes of lipid molecules, infraredmicrospectroscopy is particularly suited to testingthese two models.

The absorbance change is given by 2.303Nrh,where r is the absorption cross-section, h thesample thickness and N the number of absorbersper unit volume. Let A0 be the absorbance changedue to a single bilayer of lipid. If the actual ob-served extinction is A1, the ratio j�A1/A0 provides

a measure of the number of absorbing lipid mole-cules per unit volume, and as a test for the twoleading models. Standard studies with glyceroltrioleate have suggested that cross-section (per lipidtail) at 5.75 lm is around r� 3 ´ 10ÿ19 cm2, givingan A0 of about 1 ´ 10ÿ4. Lack of knowledge of thedetailed lipid composition in the lamellopodiacontributes to the uncertainty. The best estimatesof the observed absorbance change range betweenan upper limit of about 500 times that from a singlebilayer and the lowest limit of about 250.

For a lamellopodium in Fig. 8 in the absence ofany internal vesicles, organelles or ripples on thecell surface, the absorption due to the cell mem-brane (and associated smooth inner membrane) isexpected to be of the order of 4A0. Even withsigni®cant uncertainties, it appears that any modelthat does not require a large amount of lipid insidethe lamellopodium predicts an absorbance that isseveral orders of magnitude too small.

Turning to models which do predict a largeamount of lipid inside the cell extension, when thelamellopodium is ®lled with vesicles, the ratio isexpected to increase because of the concentration cof vesicles, to j � 1� 1

2cSlh; where the thickness of

the lamellopdium is h � 1 lm, the surface area ofthe vesicle is Sl (�10ÿ11 cm2 for spherical vesicleswith �10 nm diameter). An enhancement in j by afactor of about 250 is possible if the concentrationc is about 1017 vesicles per cm3. This rather highconcentration corresponds to a value that is onlyabout an order of magnitude less than the lipidconcentration in neat lipid. This is surprising, andsuggests that the vesicles are not far from beingclose-packed inside the lamellopodium. Furthermorphological studies would help to clarify thegeometric shape of the vesicles, but at least en-hancement of this magnitude is possible in thisclass of models.

Several enhancements of the two main modelscan be considered. One idea suggests that the la-mellopodium is ®lled with endoplasmic reticulumthat is folded back on itself many times. Mem-brane transport occurs via vesicles that fuse withthe folded endoplasmic reticulum. This model isconsistent with the absorbance change observedand also requires transport within the cell, and is avariant of the second membrane ¯ow model. Our

M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255 253

data at present cannot distinguish between theendoplasmic reticulum and the cell membrane.

A variant of the ``rake'' model was suggestedrecently by Evans [21], and independently by Do-kholyan [22]. In this model, the cell surface ishighly rippled in the lamellopodium. One wouldexpect an increase in lipid absorption while stillcon®ning the lipid molecules to the cell surface ±there is then no need for vesicle formation or fortransformations in cell topology. For a scanningtip of diameter L, and a sinusoidal ripple onthe membrane surface of amplitude w andwavelength kr, detailed calculations [22] showthat j � ���p 1� a2�=b E�jbÿ p=2j n sinÿ1�a= ���p 1�a2��; where a � 2pw=kr and b � 2pw=L. Numericalestimates using the elliptic integral suggest thatb � 300, a � 300 is needed for an enhancement inj of about 250, which corresponds to an extremelyrough surface indeed. Such a rough surface wouldhave to be reconciled in some way with the pres-ence of the polymerized actin network in the la-mellopodium. Other surface morphologies, usingrough surfaces that are not simply plane wave si-nusoidal ripples, also suggest that small amplituderipples cannot account for the observed enhance-ment. Models based on such a rough surface pre-dict that the lateral di�usion rate of lipid molecules(measured by ¯uorescence studies) would have tobe reduced greatly, by about a factor of 3. Such anenhancement has not been observed, although thequestion of the presence of ripples on the cellsurface has not been settled.

The two alternative models were ®rst proposedto explain the interesting phenomenon of capping[23]. Cap formation is restricted to motile cells, andmodels for cell capping have inspired the two dif-ferent views of cell motility. Viewed as special casesof a much more general observation of polarizedprocesses in single cells (such as early develop-mental processes in some oocytes), the potential forinfrared spectroscopy to be able to study suchprocesses appears to be of general value.

5. Conclusions

The extension of near-®eld techniques to un-derwater infrared imaging creates many problems,

but also o�ers some advantages. We ®nd that thespatial resolution achievable is actually enhancedby the presence of liquid water, with improvedresolution being obtained at the peak of the ab-sorption of the surrounding liquid. Because of thevery low throughput of the tapered near-®eldprobes, the use of increased laser power is desir-able. The presence of water provides protectionfrom thermal damage to the tapered ®bers used,and images could be obtained at signi®cantlyhigher laser power (�2 W) than has ever beendemonstrated in near-®eld microscopy. InfraredFELs provide a unique source of tunable high-power radiation for this application.

The new underwater imaging technique wasused to address a question of great interest in cellbiology, concerning the mechanism of cell motion.Images were acquired of single living motile hu-man ®broblasts with the FEL tuned to absorptionbands characteristic of proteins, as well as ab-sorption bands characteristic of lipid molecules.Our results suggest that the absorption due to lipidmolecules in the lamellopodium is higher than thatexpected from the above surface membranetransport model of cell motility. These results maylend support to an idea that lipid molecules areinvolved in substantial topological or geometricmembrane modi®cations associated with cell mo-tility.

The new technique of underwater infrared mi-croscopy has been shown to be promising. A re-cent report using the technique of aperturelessimaging suggests that a spatial resolution of k/600may be achieved in near-®eld infrared microscopy[24]. The possibility of obtaining underwater in-frared images with such enhanced resolutionwould mean that it is feasible to think of in situvibrational spectroscopy at the macromolecularlevel.

Acknowledgements

Work supported in part by the National Sci-ence Foundation (DMR) and the O�ce of NavalResearch (Contract No. N00014-94-1-1024). NVDacknowledges support from the NSF (CH9728854) and the National Institutes for Health

254 M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255

(HG00989). S.E. acknowledges the receipt of aduPont Young Professor award. We thank S.J.Hong for advice on image analysis; D. Palankerfor advice and help in AFM and ¯uorescence mi-croscopy; Y.N. Jun, D. Fung, C. Smith, E. Evansfor discussions; and the FEL center sta� for theirsupport and encouragement.

References

[1] F.S. Parker, Applications of Infrared, Raman, and Res-

onance Raman Spectroscopy in Biochemistry, Plenum,

New York, 1983.

[2] D.R. Kodali, D.M. Small, J. Powell, K. Krishnan,

Applied Spectroscopy 45 (1991) 1310.

[3] A.G. Jeung, Ph.D. Thesis, Stanford University, 1998.

[4] G.L. Carr, G.P. Williams, J.A. Re�ner, Review of

Scienti®c Instruments 66 (1995) 1643.

[5] N. Jamin, P. Dumas, J. Moncuit, W.H. Fridman, J.-L.

Teillaud, G.L. Carr, G.P. Williams, Proc. SPIE 3153

(1997) 131.

[6] L.H. Kidder, I.W. Levin, E.J. Heilweil, Optics Letters 22

(1997) 742.

[7] M.K. Hong, A.G. Jeung, T.I. Smith, H.A. Schwettman, P.

Huie, S. Erramilli, Third Free Electron Laser User's

Workshop, Rome, 1996.

[8] M.K. Hong, A.G. Jeung, S. Erramilli, First Free Electron

Laser User's Workshop, Stanford, 1994.

[9] M.K. Hong, A.G. Jeung, S. Erramilli, Second Free

Electron Laser User's Workshop, New York, 1995.

[10] M.K. Hong, S. Erramilli, P. Huie, G. James, A. Jeung,

Proc. SPIE 2863 (1996) 54.

[11] D.W. Pohl, W. Denk, M. Lanz, Applied Physics Letters 44

(1992) 651.

[12] E. Betzig, P.L. Finn, J.S. Weiner, Applied Physics Letters

60 (1992) 2484.

[13] P.J. Moyer, S.B. Kammer, Applied Physics Letters 68

(1996) 3380.

[14] D. Palanker, in preparation.

[15] H.A. Schwettman, T.I. Smith, R.L. Swent, Nucl. Instr.

and Meth. A 341 (1994) ABS19.

[16] Y.N. Jun, D. Palanker, in preparation.

[17] R. Bellairs, A. Curtis, G. Dunn (Eds.), Cell Behaviour,

Cambridge University Press, Cambridge, MA, 1982.

[18] A.G. Jeung, P. Huie, S. Erramilli, M.K. Hong, T.I. Smith,

Proc. SPIE 3153 (1997) 115.

[19] M.S. Bretscher, Science 224 (1984) 681.

[20] M.S. Bretscher, Cell 87 (1996) 601.

[21] E. Evans, private communication.

[22] N.V. Dokholyan, Boston University Technical Report,

1998.

[23] R.B. Taylor, W.P.H. Du�us, M.C. Ra�, S. de Patris,

Nature New Biol. 233 (1971) 225.

[24] A. Lahrech, R. Bachelot, P. Gleyzes, A.C. Boccara, Optics

Letters 21 (1996) 1315.

M.K. Hong et al. / Nucl. Instr. and Meth. in Phys. Res. B 144 (1998) 246±255 255