Ultramicroscopy 90 (2002) 207–213

Dual-color 4Pi-confocal with 3D-resolution in the 100 nm range

Hiroshi Kano, Stefan Jakobs, Matthias Nagorni, Stefan W. Hell* High Resolution Optical Microscopy Group, Max-Planck Institute for Biophysical Chemistry, Am Fassberg 11, 37070 Gottin. gen, Germany

Received 21 November 2000; received in revised form 2 July 2001

Abstract

We report the development of simultaneous two-color channel recordingin 4Pi-. A marked increase of spatial resolution over confocal microscopy becomes manifested in 4Pi-confocal three-dimensional (3D) data stacks of dual-labeled objects. The fundamentally improved resolution is verified both with densely labeled fluorescence beads as well as with membrane labeled fixed Escherichia coli. The synergistic combination of dual-color 4Pi-confocal recording with image restoration results in dual-color imaging with a 3D resolution in the 100 nm range. r 2002 Elsevier Science B.V. All rights reserved.

1. Introduction the typically 3.5 times sharper main maximum [1]. The side-lobes are removed mathematically by a By providinga fundamental improvement of linear filter, in which case a typical axial resolution axial sectioning, 4Pi-confocal microscopy [1] is a of 130–140 nm is achieved at a two-photon promisingdevelopment in three-dimensional (3D) excitation wavelength of 700–800 nm [3]. Alterna- far-field fluorescence microscopy. The 4Pi-confo- tively, the effective 3D- cal owes its superior sectioningcap- (PSF) of the 4Pi-confocal microscope is used ability to the coherent addition of the spherical altogether to restore the 4Pi-confocal image data wavefronts produced by two opposingobjective with a non-linear restoration algorithm [4]. Image lenses of high numerical aperture. In its simplest restoration not only removes the side-lobes, but version, also referred to as 4Pi-confocal micro- also leads to a further improvement of the spatial scopy of type A [2], the two objective lenses are resolution by about 50%, so that a spatial illuminated by coherent beams that are brought to resolution of the order of 100 nm is achieved in constructive interference in the common focal all directions [4]. Similar resolution has also been point. The use of two-photon excitation is reported by a related method, I5M, that is based particularly advantageous since it leads to a on a wide-field approach, featuringthe advantage reduction of the two side-lobes that accompany of parallel data acquisition [5]. However, this method is seriously challenged by a higher and *Correspondingauthor. Tel.: +49-551-201-1366; fax: +49- more complex side-lobe structure. 551-201-1085. This superior spatial resolution of 4Pi-confocal E-mail address: [email protected] (S.W. Hell). microscopy has been demonstrated in a series of

0304-3991/02/$ - see front matter r 2002 Elsevier Science B.V. All rights reserved. PII: S 0304-3991(01)00132-2 208 H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 applications, most notably in the imaging of the excitation was performed with a mode-locked actin [4] and the microtubule network [6] of fixed, Titanium Sapphire laser (Coherent, Inc.) operating glycerol mounted mouse fibroblast cells. More at the wavelength of 800 nm, whose beam was recently, 4Pi-confocal microscopy has been shown expanded and divided in two parts by a beam to be viable for imaging watery specimens, so that splitter. Each part illuminated the sample through for the first time the imaging of live specimens at one of the two oil objective lenses of 1.4 numerical this resolution was possible [7]. Dual-color detec- aperture (Leica 100X, Planapo), whereby the tion is an important extension to high resolution optical path lengths were matched with a precision imaging and particularly to 4Pi-confocal micro- of a few microns. The foci of the two lenses were scopy because the tacklingof many biological superimposed in space usinga piezo-stageoperat- problems requires precise knowledge about the ingin a closed loop mode. The emitted fluores- spatial distribution of differently labeled proteins, cence was collected by the objective lens on the organelles etc. Strictly speaking, co-localization of left-hand side and passed a dichroic mirror and the objects with different spectral properties, such as tube lens. To define the two color channels for color or fluorescence lifetime, has only little do detection, the fluorescence was separated by a long with the increase of resolution beyond classical wave pass dichroic mirror with an edge at 540 nm. barriers. But for a series of applications such as Each stream of fluorescence light was converged at fluorescence activated in situ hybridization (FISH) the core of an optical fiber and guided toward a or protein colocalization studies it is a viable single photon counting avalanche photo diode option to improve the spatial distribution infor- (Perkin–Elmer, Inc.). The detection system in mation [8]. conjunction with the finite openingof the fiber Successful dual channel detection faces technical provide confocality to the microscope with an challenges both in confocal fluorescence micro- openingcorresponded to 89% of the back scopy and even more so in its 4Pi-confocal projected Airy disk. counterpart, because the effect of longitudinal chromatic aberrations of the two detection chan- 2.2. Image processing nels must be eliminated [8]. Recently, we have shown that sub-resolution distance measurements Two kinds of data processingwere used for with a precision of 1 nm can be accomplished in a removingthe side-lobes in 4Pi-confocal micro- dual-color detection 4Pi-confocal axial image scan scopy. First, the 4Pi-confocal raw data were point- [9]. Here, we report for the first time 4Pi-confocal deconvolved, that is the lobes were removed by a imaging of extended dual-colored fluorescence linear filter applied in each axial line [3]. In order specimens, such as densely packed sub-resolution to establish the point-deconvolution by a linear fluorescence beads, as well as doubly labeled filter, only the relative location and height of the E. coli. We demonstrate a strikingincrease in two sidelobes have to be known. This 1D resolution over dual-color confocal fluorescence deconvolution is fast enough to be carried out microscopy. duringimageacquisition. We note that the effect of the point-deconvolution is restricted to the removal of the lobe artifacts alone and does not 2. Material and methods lead to an increase of resolution per se, because it does not widen the spatial frequency bandwidth of 2.1. 4Pi-confocal microscope the optical system. Secondly, we applied an iterative image restoration algorithm with a non- The employed two-photon excitation 4Pi-con- negativity constraint. We chose the Richardson– focal microscope of type A has been described in Lucy (RL) algorithm [10,11] which encompasses Ref. [9]; here we summarize the physical para- maximum likelihood estimation (MLE) for images meters of operations and expound on the method dominated by Poisson noise. This algorithm has with which we reduced chromatic aberrations. The been shown to be a good choice for images H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 209 recorded by a fluorescence microscope. The RL medium (4% Dapco, Serva, Heidelberg, Germany algorithm has been modified by Conchello and in 96% glycerol) and placed between two cover- McNally [12], so as to include Tikhonov regular- slips. ization as well. For our restorations, we used this modified algorithm [6]. Suitable regularization 2.5. Labeling of E. coli parameters were determined by simulated imaging of a known test object and subsequent restoration. E. coli K-12 DH5a were grown overnight to For a thorough illustration of the resolution steady state in LB-medium (10 g/l tryptone, 5 g/l improvement, we also recorded the data stack in yeast extract and 10 g/l NaCl) at 371C. Membrane the confocal mode, which can be accomplished by labellingwas accomplished by addingthe vital obstructingone of the highaperture lenses. stain FM 1–43 (Molecular Probes) to a final concentration of 0.1 mM from a 2.5 mM EtOH 2.3. Eliminating longitudinal chromatic aberrations stock solution to the bacterial culture. Staining was performed at room temperature for 1–2 h. Two-photon excitation enabled the excitation of Nucleoids were labeled by adding30 mm DAPI to both fluorophores in the same focal volume, so the cells 15 min prior to fixation. For fixation the that longitudinal chromatic aberration on the cells were incubated for 1 h in 3.5% paraformal- excitation side cannot occur. Chromatic aberra- dehyde. tion associated with the two detection color To mount the bacteria, glass cover slips were channels is equivalent to an axial offset of the first coated with poly-l-lysine (Sigma, St. Louis, detection PSF defined by the confocal spatial MO). The coverslips were briefly incubated with openings of the two fibers. We eliminated this a diluted suspension of 110 nm green spheres offset by imaging a monomolecular fluorescent (Molecular Probes). These spheres attached to polydiacetylene layer mounted on a cover slip by the glass and facilitated adjustment of the wave- the Langmuir–Blodgett technique [9,13]. Two- fronts to constructive interference and the control photon excitation of these thin layers results in a of the relative phase of the wavefronts during broad fluorescence coveringa wide rangeof the image acquisition. After a washing step the cover- visible spectrum. Therefore the fluorescence pro- slips were anew coated with poly-L-lysine. Finally duced by the same excitation spot could be the fixed labeled bacteria were re-suspended in recorded in both channels. The matchingof the Dapco mountingmedium and placed between two two detection PSFs was accomplished by translat- coverslips each covered by fluorescent beads. ingeach detection fiber openingalongthe optic axis, so that both detection PSFs matched in space with the common excitation spot. The chromatic 3. Results and discussion shift was quantified by scrutinizingthe axial responses to the monomolecular fluorescent layer 3.1. Imaging dense clusters of dual-color beads in the two color channels and found to be o5 nm. First, we investigated the performance of the 2.4. Dual-color bead sample two-color 4Pi microscope by imaging a test sample consistingof red, greenand non-fluorescent beads. Red- and green- fluorescent beads with a The fluorescence emission maxima of 605 and maximum emission of 605 and 515 nm, respec- 515 nm, respectively, allowed a straightforward tively, were densely packed and thoroughly mixed spectral separation of both colors. The non- with non-fluorescent 110 nm polystyrene spheres fluorescent beads served as invisible spacers (all from Molecular Probes, Inc. Eugene, OR; the between the fluorescent ones. We recorded 3D green beads used are referred to as ‘yellow–green’ data stacks of the same sample volume in the 4Pi- by the manufacturer). After a washingstep they and the confocal two-photon excitation mode. were dried on a coverslip, embedded in mounting Both image stacks consist of 24 parallel xz images 210 H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 with a pixelation of 64 96. The voxel sizes were color (green) can be deduced. The pronounced 36.1, 48.0, and 24.1 nm in x; y; and z; respectively. dip and clear separation of the two spectrally The acquisition time for a slice was 37 s in the 4Pi- identical beads in Fig. 1 indicates that the system is and the confocal mode. Fig. 1A shows the able to resolve beads at 100 nm distance, which is confocal (a–d), point-deconvolved 4Pi-confocal also confirmed when scrutinizingthe 3D data (e–h) and RL-restored images (i–l) of individual stack. The physically less challenging precise xz-optical sections of the bead sample image sta spatial discrimination of beads of different color ck. We observed no cross-talk of ‘red’ fluore- is also possible of course, as can be recognized scence into the ‘green’ channel but a 45% cross- from the two adjacent red and green beads in the talk from the green beads into the ‘red’ channel. upper left corner of Fig. 1A (i) and in the center After cross-talk elimination the red spheres part of Fig. 1A (l). could be unequivocally separated from the green Since this algorithm enhances resolution in 3D, beads. For the RL-restoration a regularization RL-restored images have an improved resolution parameter of l ¼ 4106 was chosen and the in the y-direction as well. This is manifested as a process was stopped after 300 iterations. The ‘disappearance’ or ‘reduction in intensity’ of beads shown xz-images are successive slices in the in some of the xz-sections of the RL-restored data. y-direction. For example, whereas in the confocal and 4Pi- Comparingthese slices impressively reveals the confocal data of Fig. 1A(d, h) two green and two difference in resolution of these imaging modes. red beads (marked by red circle) appear to be in The poorer axial resolution of 525710 nm in the same optical plane, the RL-restored image the confocal mode results in an elliptical reveals, that one of the green beads is localized in representation of the spheres (Fig. 1A, a–d). the next optical plane; see Fig. 1A(l) and The lateral full-width-at-half maximum (FWHM) Fig. 1C(d). of B200 nm of the 4Pi-confocal microscope is Obviously, confocal images can also be restored unchanged with respect to its confocal counter- by the same non-linear algorithm. Therefore the part, but in the axial direction the FWHM is question arises to which extent the resolution of reduced to 13575 nm. Whereas a number of confocal microscopy can be improved by restora- beads are so densely clustered that they tion alone. The answer is given in Fig. 1C, which cannot be distinguished by the confocal micro- directly compares the performance of all imaging scope, the axially narrower focus of the 4Pi modes on the dual-channel data. It is found that microscope resolves the beads markedly better while restoration leads to an improvement of (Fig. 1A, e–h). Not surprisingly, the 4Pi-confocal spatial resolution in confocal microscopy, Fig. 1C, images convey much more structural informa- (a and b), the improvement in axial resolution tion about the bead clusters. is not as significant as in the 4Pi-confocal The computationally more extensive image restoration by the iterative RL algorithm increases ——————————————————————————c the resolution of the dual-color 4Pi-confocal Fig. 1. (A) Confocal (a–d), point-deconvolved 4Pi-confocal images further, Fig. 1A, (i–l). Importantly, by (e–h) and RL-restored 4Pi (i–l) xz images of mixed red- and usingthis algorithmadjacent beads could be green-fluorescent beads. The signals have been detected in resolved, which could not be discriminated in the two channels, and, after cross-talk elimination, the images from both channels have been overlaid. Image size is confocal and only barely in the 4Pi-confocal 2.31 mm 2.31 mm; the confocal data (a–d) have been smoothed mode (see yellow arrows in Fig. 1A). The dramatic without affectingresolution. The voxel size is 36.1, 48.0, and gain in resolution is further outlined in Fig. 1B 24.1 nm in x; y and z; respectively; the bar represents 1 mm. (B) showingthe intensity line profiles alongthe Intensity line profiles alongthe optic axis in the regions optic axis in the regions indicated by the indicated by the yellow arrows in (A). (C) Comparison of RL-deconvolved confocal image (a), confocal image (b), arrows. From the intensity profile of the RL- linear deconvolved 4Pi (c) with RL-deconvolved 4Pi picture. restored 4Pi-confocal image a distance of 125 nm The images are taken in continuation from the same data between the centers of two beads of the same stack as in (A). H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 211 212 H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 microscope, Fig. 1C, (c and d). For example, in the els were easily separable. Crossover was 10% in restored confocal image data in Fig. 1C(a) a each direction and mathematically removed. number of beads are not separable alongthe optic The original xyz-image size was 3.08 1.54 axis, but clearly distinguishable in the non-restored 3.08 mm3 with a pixelation of 48.2, 96.1, and 4Pi-confocal data in Fig. 1C(c). However, the 24.1 nm in x; y; and z directions. The imaged restored 4Pi-confocal data testifies to a striking stacks consist of 16 parallel xz slices with a increase in spatial resolution, compared to all pixelation of 64 128. other imaging modes. Eight xz-images (2.7 1.8 mm2) taken from a stack coveringthe whole bacterium are shown in 3.2. Imaging of dual-color labeled bacteria Fig. 2(a–h) as recorded with the 4Pi-confocal microscope. A correspondingconfocal record- Next we recorded stacks of xz-images of ingwas taken for comparison, shown in Fig.2, fluorescently labeled E. coli (Fig. 2). For this (i–p). Whereas the 4Pi-confocal image nicely purpose the membranes and nucleoids of the reveals the membrane of the cell, the axially bacteria have been labeled in vivo with the dyes elongated focus of the confocal microscope FM 1–43 and DAPI, respectively. Prior to imaging strongly overemphasizes those parts of the mem- the bacteria were fixed with paraformaldehyde brane that are oriented parallel to the optic axis. and embedded in a glycerol-based medium. Due Moreover, due to the poor axial resolution it is to their different emission wavelengths (lDAPI barely able to recognize the membrane as a distinct B460 and lFM 1243B590 nm) the fluorescent lab feature surroundingthe cell. At the scale observed

Fig. 2. Single optical slices of fixed E. coli with FM 1–43 stained membranes and DAPI labeled nucleoid. The emissions of both fluorophores have been recorded separately in two detection channels. After cross-talk elimination both images have been overlaid. The dimensions of the displayed 3D stack are 2.7 0.8 1.8 mm3 in the respective xyz-directions. (a–h) Linear point-deconvolved 4Pi images; (i–p) corresponding sections taken in the confocal mode; (q–x) RL-deconvolved 4Pi images (shown are only the FM 1–43 signals); bar, 1 mm. H. Kano et al. / Ultramicroscopy 90 (2002) 207–213 213 by our the nucleoid of this bacterial detection and superior resolution are required, cell does not have an equally well-defined struc- such as investigations in the cell nucleus by ture, so that in this case, the 4Pi-confocal fluorescent in situ hybridization, or colocalization recordingis not more informative than the studies of proteins and organelles. In the future, it confocal data. will be highly interesting to explore the potential of Finally, after non-linear image restoration the this method for these applications. FM 1–43 marked membranes can be seen with impressive clarity. The gain of resolution obtained in the 4Pi-confocal restored data over those of the Acknowledgements confocal is best recognized by directly contrasting Fig. 2(q–x) with Fig. 2(i–p). We note, however, We would like to thank David Britt for careful that this mathematical procedure emphasized readingof the manuscript. One of us (HK) would some minor artefacts that can be seen as faint like to thank the Japanese Society for the linear structures runningparallel to the mem- Promotion of Science for a postdoctoral fellow- branes. These artefacts most likely are remnants of ship. This work was supported by the Deutsche the sidelobes that are not perfectly eliminated due Forschungsgemeinschaft (DFG) by a grant to to imperfections in phase adjustment or phase SWH (He-1977). jitter. The RL-deconvolution did not improve the image quality of the DAPI stained nucleoid, since its structures are still too small to be resolved with References the current resolution, the membrane, however, is imaged with unprecedented clarity. [1] S.W. Hell, E.H.K. Stelzer, Opt. Commun. 93 (1992) 277. [2] S. Hell, E.H.K. Stelzer, J. Opt. Soc. Am. A9 (1992) 2159. [3] P.E. Hanninen,. S.W. Hell, J. Salo, E. Soini, C. Cremer, 4. Conclusion Appl. Phys. Lett. 66 (1995) 1698. [4] S.W. Hell, M. Schrader, H.T.M. van der Voort, J. Microsc. 185 (1997) 1. In summary, our results demonstrate that the [5] M.G.L. Gustafsson, D.A. Agard, J.W. Sedat, J. Microsc. fundamentally improved axial resolution of 4Pi- 195 (1999) 10. confocal microscopy is also attained in dual-color [6] M. Nagorni, S.W. Hell, J. Struct. Biol. 123 (1998) 236. imaging. In combination with non-linear compu- [7] K. Bahlmann, S. Jakobs, S.W. Hell, Ultramicroscopy 87 tational image restoration the method provides 3D (2001) 155. [8] H. Bornfleth, K. Satzler, R. Eils, C. Cremer, J. Microsc. resolution in the 100 nm range. As a result, we 189 (1998) 118. were able to resolve details in doubly labeled [9] M. Schmidt, M. Nagorni, S.W. Hell, Rev. Sci. Instrum. 71 bacteria that are clearly not accessible by standard (2000) 2742. confocal (two-photon) microscopy. Dual-color [10] W.H. Richardson, J. Opt. Soc. Am. 62 (1972) 55. detection significantly enlarges the scope of 4Pi- [11] L.B. Lucy, Astron. J. 79 (1974) 745. [12] J.-A. Conchello, J.G. McNally, SPIE Proc. 2655 (1996) confocal microscopy. It will most likely turn out to 199. be of great relevance to a series of important [13] M. Schrader, U.G. Hofmann, S.W. Hell, J. Microsc. 191 biological applications in which both multi-color (1998) 135.