VOLUME 88, NUMBER 16 P H Y S I C A L R E V I E W L E T T E R S 22 APRIL 2002 Focal Spots of Size l 23 Open Up Far-Field Florescence Microscopy at 33 nm Axial Resolution Marcus Dyba and Stefan W. Hell* High Resolution Optical Microscopy Group, Max-Planck-Institute for Biophysical Chemistry, 37070 Göttingen, Germany (Received 19 September 2001; published 4 April 2002) We report spots of excited molecules of 33 nm width created with focused light of l 760 nm wavelength and conventional optics along the optic axis. This is accomplished by exciting the molecules with a femtosecond pulse and subsequent depletion of their excited state with red-shifted, picosecond- pulsed, counterpropagating, coherent light fields. The l 23 ratio constitutes what is to our knowledge the sharpest spatial definition attained with freely propagating electromagnetic radiation. The sub-diffraction spots enable for the first time far-field fluorescence microscopy with resolution at the tens of nanometer scale, as demonstrated in images of membranes of bacillus megaterium. DOI: 10.1103/PhysRevLett.88.163901 PACS numbers: 42.30.­d, 41.90.+e, 42.25.­p, 42.79.­e In 1873 Ernst Abbe discovered that the smallest focal cence spot, i.e., an effective point-spread-function (PSF) spot of a lens is limited by diffraction to about l 2n, with of the microscope, that is below the diffraction limit [6]. l denoting the vacuum wavelength of light and n the re- In this work we demonstrate fluorescent spots of 33­ fractive index [1]. Abbe's discovery put an end to the im- 46 nm extent produced with l 745 760 nm and con- provement in the resolution of far-field light microscopy ventional lenses. Moreover, we employ these spots to and established a prominent physical problem, known as deliver for the first time far-field microscopy images with the "diffraction barrier." Despite the enormous progress resolution on the order of tens of nanometer. brought about by electron, scanning probe, and near-field Our result became possible by synergistically imple- optical microscopy, the limited resolution has remained menting elements of the two unrelated concepts of STED an obstacle in many areas of science. Cell biology, for and 4Pi microscopy. The fluorescent sample is placed example, depends on focused light for noninvasively prob- in the common focus of two opposing lenses, but exci- ing the cellular interior at the sub-mm scale. Similarly, in tation and detection are performed through a single lens, lithography and optical data storage efforts are being made L1, only (Fig. 1). For this purpose a train of 250 fs to produce smaller spots in a noncontact mode. pulses of 554 nm wavelength are directed via mirror M1, As with any great physical challenge, the diffraction bar- beam-splitter BS, and the dichroic mirror DC2. The lenses, rier preoccupied a number of scientists. In 1952 Toraldo di which are alternatively pairs of water or oil immersion Francia proposed an intriguing concept to produce smaller lenses, feature the numerical aperture, 1.2 and 1.4, respec- focal spots [2]. In a theoretical study he showed that with tively, thereby establishing a tight excitation intensity PSF finely tuned pupil filters the light intensity can be diverted hexc r (Fig. 2a). The fluorescence is imaged onto a con- away from the focal point to leave a tiny central spot. focal point detector, described by a detection PSF hdet r . Unfortunately, the onset of giant side lobes rendered the We used the styryl dyes RH414 (Molecular Probes, Eugene method impractical. A more straightforward approach has OR) and Pyridine 2 (Lambda Physik, Göttingen, Germany) been to reduce l, as in x-ray-microscopy. Indeed a reso- emitting in the 600­760 nm range. lution approaching 30 nm has been reported with this tech- Immediately after the excitation, a pulse of l 745 nique [3]. However, abandoning visible light in favor of 760 nm and t 13 ps duration, denoted by STED pulse, l 2 4 nm synchrotron radiation not only adds com- enters the focal region. These photons primarily act on plexity but is incompatible with live cell studies [3]. An- the excited state S1, inducing stimulated emission down other approach is to enlarge the aperture as realized in to a vibrational sublevel of the ground state Svib 0 (Fig. 1b). 4Pi microscopy [4]: by the coherent addition of the two Subpicosecond vibrational decay empties Svib 0 , so that wave fronts of opposing lenses the resolution along the repumping into S1 is largely ineffective. By the time the optic axis has been improved from 500 to 100 nm. STED pulse has vanished, the population of the S1 is However, these concepts are still diffraction limited. A N t t, hSTED N0 exp 2kflt 2 shSTED , where N0 concept which really breaks the diffraction limit in the is the population just after excitation, kfl 1 ns 21 ø important imaging by fluorescence is stimulated emission t21 is the radiative decay rate, s 10216 cm2 the depletion (STED) microscopy [4,5]. Its rationale is to cross section for stimulated emission, and hSTED r suppress the spontaneous emission at the periphery of the is the PSF of the STED pulse in photons per area per pulse. R diffraction-limited fluorescence spot of a scanning confo- Hence, fluorescence is reduced by h h R STED dt 3 cal microscope by stimulated emission. The suppression N t, hSTED dt N t, 0 exp 2shSTED . (The con- occurs in such a way that fluorescence is allowed at the sideration of a finite vibrational lifetime requires the focal point, but not in its proximity. The result is a fluores- numerical solution of a set of differential equations, in 163901-1 0031-9007 02 88(16) 163901(4)$20.00 © 2002 The American Physical Society 163901-1 VOLUME 88, NUMBER 16 P H Y S I C A L R E V I E W L E T T E R S 22 APRIL 2002 FIG. 2. Calculated intensity point-spread functions for: (a) ex- citation, (b) stimulated emission, and (c) the effective PSF, i.e., fluorescent spot of the STED-4Pi-microscope. Panel (d) depicts the measured fluorescence as a function of the stimulating pho- ton density. Panel (e) is the calculated response V z to an infi- nitely thin fluorescence plane. (c) and (e) reveal a fundamental improvement of resolution in z. FIG. 1. STED-4Pi microscope. (a) Fluorescence excitation and detection occur via lens L To reduce the spot size, hSTED r must vanish at the 1, whereas stimulated emission is generated by the light field of counterpropagating, aberrated focal point r 0 but be high elsewhere. A narrow mini- wave fronts of L1 and L2. Imaging is accomplished by scan- mum of l 4n full-width-half-maximum (FWHM) is ning the sample through the sub-diffraction-sized spot of the achieved with a standing wave. A planar standing wave, two lenses. The inserted sketches depict the aberration induced however, exhibits many minima in which fluorescence by the phase plate on the counterpropagating STED-beam wave fronts. (b) Fluorophore energy levels. would be still present. To create a hSTED r with a single minimum, we exploit the high focusing angle and sym- metry of the 4Pi microscope. An aberration prior to BS blurs potential side minima, but still renders hSTED 0 0. which case a slight deviation from the exponential law is Hence, we introduced an aberration C q mpu q 2 found [4].) In Fig. 2d we show the measured h hSTED Pa in the STED beam, whereby 0 # q # a denotes for the dyes used, with hSTED denoting hSTED r aver- the semiaperture angle and u the Heaviside step func- aged over the Airy disk. For hSTED . 1016 cm22 STED tion. For the oil and water lenses we elected moil is the predominant process with Pyridine 2 and fluores- 0.4, Poil 0.37 and mwater 0.56, Pwater 0.57, re- cence is reduced down to 5.5%. The depletion of the spectively, C q was realized by a MgF2 coated glass plate membrane-incorporated dye RH414 is less efficient, but [6], as depicted in Fig. 1. Destructive phase at r 0 was still pronounced. adjusted by a piezo acting on BS. Following the theory by Richards and Wolf [7] we obtain for the focal intensity: 0 1 n2c´ 1 0 0 h 0t B C STED r j E @ 0 21 0 A E 2 ¯hv 1 r 1 E2 r j2, with E2 r, z, f 1 r, 2z, 2f , 0 0 21 Z a 2iE p E 0nf 1 r, z, f dq df0 cosq sinq exp i C q 1 k s 2 f (1) l 0 0 1 cos2 f0 2 f cosq 1 sin2 f0 2 f 3 B @ sin f0 2 f cos f0 2 f cosq 2 1 C A , 2 cos f0 2 f sinq where E1,2 denotes the electric field amplitude in cylin- any phase deviation from a spherical wave front. Fig- drical coordinates, E0 is the pulse-averaged amplitude ure 2b depicts the numerically calculated hSTED r, z, 0 at the lens entrance pupil, c the speed of light, ´0 the featuring the desired central minimum. hexc r of Fig. 2a permittivity of free space, f is the focal length, s the path and hdet r are gained by setting C q 0 and calculat- from the point f, u, f on the converging wave front to ing j E1 r j2 for the excitation and fluorescence wave- r, z, f , and k 2pn l. C q an aberration denoting length, respectively. For excitation and stimulated emission 163901-2 163901-2 VOLUME 88, NUMBER 16 P H Y S I C A L R E V I E W L E T T E R S 22 APRIL 2002 we assumed x polarization (f 0); the detection was unpolarized (f p 4). Apart from a constant C, the effective PSF and spot size of our STED-4Pi-microscope is now heff r Chexc r hdet r h hSTED r Chexc r hdet r exp 2shSTED r . (2) The exponential suppression by hSTED r squeezes the spot below the diffraction limit. An unbound increase of hSTED r would lead to resolution at the molecular scale. The value achievable with our set of conditions, however, is finite and assessed by implementing the measured h hSTED of Fig. 2d into hSTED r of Fig. 2b. The evaluation of heff r for max hSTED 4 3 1017 cm22 yields a spot that is fundamentally reduced along the optic axis (Fig. 2c). The associated axial response RR V z heff r, z, f r dr df features a FWHM of 36 nm (Fig. 2e). To verify this prediction, we carried out a series of mea- surements. V z was acquired by imaging an axial fluores- cence half space, such as the edge of a Pyridine 2 solution and calculating its derivative. Figure. 3a shows V z for max hSTED 7.3 3 1017 cm22, l 760 nm, and oil immersion lenses. The FWHM was 33 6 2 nm, in agree- ment with prediction. Being relevant to biological imag- ing, we also investigated water immersion. In this case we precipitated a thin (,10 nm) dye layer on the cover slip out of a dilute aqueous solution of Pyridine 2, so as to es- tablish the z response directly. The V z of the STED-4Pi- microscope, recorded at max hSTED 3.8 3 1017 cm22 FIG. 3. Z resolution as quantified by the axial response V z and l 745 nm, FWHM of 46 6 5 nm, was narrower for (a) oil and (b) water immersion lenses. Note the experimental than its confocal counterpart by 17.5 6 2 (Fig. 3b). confirmation of the prediction in Fig. 2(e). Panel (c) compares At the highest power level, the residual stimulating confocal and STED-4Pi xz images of thin fluorescent layers intensity at r 0 led to a 55% reduction of V z 0 of at 560 nm distance, separated by a dilute watery fluorescence solution: the z profiles are shown in (d). The layers are not the STED-4Pi signal as compared to its confocal counter- distinguishable in the confocal, but clearly in the STED-4Pi part. Figure 3b also reveals that V z response exhibited microscope. side maxima of 18 24% at z 6lSTED 2n, originating from an incomplete depletion at the side minima of membrane were more strongly represented. The main hSTED r . Potential remedies are an improved C q and result, however, was a vastly improved axial resolution. filtering by linear deconvolution. Figure 3c shows xz The image also exhibited side-lobe effects, anticipated images of two ultrathin fluorescent layers on cover slips from the measured V z . One reason for the lobes was separated by 560 nm. In contrast to the confocal recording the incomplete suppression in the side minima; another the STED-4Pi-microscope separates them clearly. was the less favorable h hSTED for the membrane- In a biological imaging application, we labeled the incorporated RH414 molecules, as displayed in Fig. 2d. membrane of bacillus megaterium with RH414. The Fortunately, the effect of the lobes can be dealt with dye molecules are incorporated with the transition dipole a single-step, linear Tikhonov-filter [9] extracted from oriented primarily perpendicular to the membrane [8]. an experimental V z featuring the same lobe height as Figure 4a is a standard confocal xz image, revealing the bacterial membrane. The resulting image is shown that the confocal mode overemphasized the membrane in Fig. 4c. The profiles through the membranes exhibit regions oriented in z, partly due to the excitation field a 30 nm FWHM and a distinct separation of opposite being parallel to the transition dipoles, but also to the membrane regions. confocal spot being elongated by a factor of 4 in z. In The max hSTED 5.1 3 1017 cm22 at l 745 nm the STED-4Pi microscope, the situation was inverted. was achieved by focusing 8.78 mW of average power, a According to the calculation of Fig. 2c, heff was now 4 level well within the range of what is typically used in times narrower in z. The STED-4Pi-counterpart image nonlinear microscopy. It is tempting to target molecu- of Fig. 4b confirms this since the transverse parts of the lar resolution by further increasing max hSTED , but this 163901-3 163901-3 VOLUME 88, NUMBER 16 P H Y S I C A L R E V I E W L E T T E R S 22 APRIL 2002 focal volume as defined by the FWHM could be squeezed 17 3 32 153-fold. We note that the xz image in Fig. 4 would not have been possible with near-field optics, because 3D sectioning requires freely propagating waves. We also note that the axial resolution benchmark reported here is on the order of the transverse resolution in x-ray microscopy [3], which, however, requires a 250-fold smaller wavelength. The strategy by which we break the diffraction barrier is photoinduced saturated depletion of the excited state right in the proximity of a fully unaffected region. This strat- egy could be applied to any photostable three-level sys- tem with a transient state into which the excited molecules are shelved. Therefore possible alternatives to stimulated emission are depletion of the ground state [4] or photoin- duced cis-trans isomerization. Our results may also have implications for microlithography and data storage, for which methods are being sought that open up the nanome- ter scale with visible light. Having defined a 30 nm spot of excited molecules, one could apply a third light field ini- tiating a photochemical reaction with the same resolution. In summary, we have demonstrated focal spots of ex- cited molecules with 33 nm spatial extent along the op- tic axis in full-width-half-maximum. Corresponding to 1 23 of the applied wavelength, this is fundamentally be- yond the diffraction limit and constitutes to our knowledge the smallest spatial definition hitherto attained with freely propagating electromagnetic radiation. We have imple- mented these spots in light microscopy and, using conven- tionally focused visible and near-infrared light, obtained the first far-field images featuring a resolution of tens of FIG. 4. XZ images of membrane-labeled bacteria: (a) confo- nanometer. cal, (b) STED-4Pi, (c) STED-4Pi after linear filtering. (b) and We thank S. Jakobs and T. Klar for valuable discus- (c) mark the first far-field light microscopy images with spatial sions. This work was supported by the DFG through Grant resolution of 30 40 nm, corresponding to l 23. No. He-1977. endeavor will be challenged by photobleaching and by the residual hSTED 0 0. The latter could possibly be met by implementing active optical elements. Importantly, *Email address: shell@gwdg.de the STED wavelength and not its excitation partner es- [1] E. Abbe, Arch. Mikrosk. Anat. 9, 413 (1873). tablish the achievable resolution. Two consequences arise [2] G. Toraldo di Francia, Nuovo Cimento Suppl. 9, 426 (1952). from this fact. First, with a molecular system operating at [3] W. Meyer-Ilse, D. Hamamoto, A. Nair, S. A. Lelievre, l 380 nm one should be able to achieve 17 nm. Sec- G. Denbeaux, L. Johnson, A. L. Pearson, D. Yager, ond, the details of the excitation spot are not important; in M. A. Legros, and C. A. Larabell, J. Microsc. 201, 395 fact, our system was optimized only for l 745 760 nm, (2000). which simplifies realization. [4] S. W. Hell, in Topics in Fluorescence Spectroscopy, edited The STED-4Pi microscope improves the resolution in by J. R. Lakowicz (Plenum Press, New York, 1997), the z direction. Evaluation of the intensity profile of a Vol. 5, p. 361. z-oriented bacterial membrane revealed a transverse reso- [5] S. W. Hell and J. Wichmann, Opt. Lett. 19, 780 (1994). lution of 250 6 10 nm (FWHM), both for the 4Pi-STED [6] T. A. Klar, S. Jakobs, M. Dyba, A. Egner, and S. W. Hell, and the regular confocal setup. This is in agreement with Proc. Natl. Acad. Sci. U.S.A. 97, 8206 (2000). the FWHM of 234 nm predicted for the utilized water im- [7] B. Richards and E. Wolf, Proc. R. Soc. London A 253, 358 (1959). mersion lens. To gain a factor of 3 or more in the trans- [8] L. Moreaux, O. Sandre, S. Charpak, M. Blanchard-Desce, verse direction one can apply an additional STED pulse and J. Mertz, Biophys. J. 80, 1568 (2001). with a doughnut-shaped hSTED r . Consecutive action of [9] A. N. Tikhonov and V. Y. Arsenin, Solutions of Ill-Posed both would sculpt the spot laterally and axially so that the Problems (Wiley, New York, 1977). 163901-4 163901-4