- Email: [email protected]
Nonlinear optical methods for cellular imaging and localization
Methods xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Methods journal homepage: www.elsevier.com/locate/ymeth
Nonlinear optical methods for cellular imaging and localization A. McVey a, J. Crain a,b,⇑ a b
School of Physics, The University of Edinburgh, EH9 3JZ Scotland, United Kingdom National Physical Laboratory Teddington, TW11 0LW, United Kingdom
a r t i c l e
i n f o
Article history: Received 20 December 2013 Revised 28 February 2014 Accepted 2 March 2014 Available online xxxx Keywords: Imaging Nonlinear optics Biophysics Fluorescence Raman spectroscopy
a b s t r a c t Of all the ways in which complex materials (including many biological systems) can be explored, imaging is perhaps the most powerful because delivering high information content directly. This is particular relevant in aspects of cellular localization where the physical proximity of molecules is crucial in biochemical processes. A great deal of effort in imaging has been spent on enabling chemically selective imaging so that only specific features are revealed. This is almost always achieved by adding fluorescent chemical labels to specific molecules. Under appropriate illumination conditions only the molecules (via their labels) will be visible. The technique is simple and elegant but does suffer from fundamental limitations: (1) the fluorescent labels may fade when illuminated (a phenomenon called photobleaching) thereby constantly decreasing signal contrast over the course of image acquisition. To combat photobleaching one must reduce observation times or apply unfavourably low excitation levels all of which reduce the information content of images; (2) the fluorescent species may be deactivated by various environmental factors (the general term is fluorescence quenching); (3) the presence of fluorescent labels may introduce unexpected complications or may interfere with processes of interest (4) Some molecules of interest cannot be labelled. In these circumstances we require a fundamentally different strategy. One of the most promising alternative is based on a technique called Coherent Anti-Stokes Raman scattering (CARS). CARS is a fundamentally more complex process than is fluorescence and the experimental procedures and optical systems required to deliver high quality CARS images are intricate. However, the rewards are correspondingly very high: CARS probes the chemically distinct vibrations of the constituent molecules in a complex system and is therefore also chemically selective as are fluorescence-based methods. Moreover,the potentially severe problems of fluorescence bleaching and quenching are circumvented and high-resolution three dimensional images can be obtained on completely unlabelled specimens. We review here aspects of CARS and Multiphoton fluorescence techniques to cellular localization and measurement. Ó 2014 Published by Elsevier Inc.
1. Motivation Despite their chemical specificity, high resolution and optical sectioning capability, fluorescence-based imaging techniques all share one fundamental limitation. Namely, that the molecular constituents of interest must be stained by incorporating ’’foreign’’ labels to create image contrast. In the most benign cases, these labels act as robust but passive dyes which have no effect other than to render visible the molecules to which they are attached. In some circumstances, however, the dyes are neither robust nor passive. Specifically, prolonged exposure to the excitation beam often leads to permanent damage to the dye molecules. This ⇑ Corresponding author. School of Physics, The University of Edinburgh, EH9 3JZ Scotland, United Kingdom E-mail address: [email protected] (J. Crain).
photobleaching degrades image contrast and forces compromises between illumination intensity, image quality and acquisition speed. Image quality may also suffer from fluorescence quenching whereby non-radiative electronic relaxation processes become significant and suppress fluorescence yield. Beyond these practical issues connected to obvious loss of signal, other problems are more subtle. Namely, free radical reaction products generated during photobleaching may be toxic for biological specimens. Moreover, fluorescent labelling itself can cause non-negligible perturbations to chemical or biophysical pathways and to cellular functions the severity of which cannot be assessed a priori. Despite intensive worldwide efforts to design more robust fluorescent labels with tailored properties and the use of further chemical additives and stabilizers such as anti-fading agents, new imaging modalities that do not rely on fluorescence at all offer a very elegant alternative for applications across the physical and life sciences. Multiphoton
http://dx.doi.org/10.1016/j.ymeth.2014.03.002 1046-2023/Ó 2014 Published by Elsevier Inc.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
2
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
methods partially mitigate these risks and we consider aspects of the method below. An alternative strategy involves the inherent and specific molecular vibrations of molecules in complex systems as the basis for imaging contrast in unstained samples. In principle, either infrared (IR) or Raman spectroscopy is potentially suitable as the basis for vibrational microscopies but both have serious disadvantages in practical applications. Specifically, the long excitation wavelengths required for IR imaging severely degrade spatial resolution. Also, strong water absorption complicates applications in aqueous systems including biological cells. The combination of infrared spectroscopy with scanning nearfield optical microscopy (IR-SNOM) [1,2] or tip-enhanced Raman spectroscopy (TERS) [3] can provide signal enhancement and spatial resolution below the diffraction limit. Raman scattering avoids these limitations through the use of visible excitation sources but the intrinsically small scattering cross section makes detection very difficult (especially in the presence of background autofluorescence) and usually precludes application in 3D microscopy of biological systems. In what follows we will show that a multiphoton variant of spontaneous Raman scattering is emerging as a powerful tool for 3D microscopy of complex systems and that it may offer significant advantages over incoherent laser-induced fluorescence and spontaneous Raman techniques for certain applications. 1.1. Physical principles . 1.2. Two-photon fluorescence Insertion of fluorophores into samples allows samples otherwise indistinct to light microscopy to be examined with far greater resolution and accuracy. By exciting an electron within the flourophore from the ground state to a higher electronic state the emission of a photon at a Stokes shifted (red-shifted) frequency can be observed. Single-photon fluorescence is induced by a single photon causing the excitation and as such the excitation is proportional to the intensity (I) of the incident light as every photon has an equal probability of causing emission. It is therefore necessary for the photon to be of high-energy in order to match the required excitation gap (typically UV or visible light). Two-photon fluorescence (TPF) occurs when two photons are absorbed simultaneously (within 10(18) seconds of each other) [4] by an electron inducing a excitation from the ground state to a higher order electronic state. Any two photons which sum to match the energy gap between the ground and excited state are sufficient but in practice they are usually of the same wavelength. As energy is inversely proportional to wavelength, the required wavelengths are twice as long as those necessary for single photon absorption. This increase in wavelength from UV–vis to nearinfrared significantly reduces the photodamage caused by incident photons within the sample and allows for deeper penetration depths for live tissue imaging [5]. Fig. 1 shows the energy level scheme for two photon excitation. Since TPF involves the combination of two photons, the fluorescence emission is proportional to the square of the incident intensity. This phenomenon means that in order to generate TPF a high density of photons must be present in the sample effectively confining TPF to a small focal volume within the sample. This is extremely advantageous for imaging live samples as only the focal volume is subject to photodamage at any one time. Unfortunately the high density of photons (1020–1030 photons per square cm) [4] required to generate TPF means that peak laser powers must be higher than those needed for single photon
Fig. 1. Energy level diagrams for CARS (left) and two-photon fluorescence (right). CARS is an excitation to electronically non-resonant energy levels (light grey) by three incident photons, frequency xp and xS respectively to create a fourth at a blue-shifted frequency, xas . This limits the energy absorption by the sample. Twophoton fluorescence is an electronically resonant excitation to a higher electronic energy level (black) by two incident photons (usually of the same frequency, xS ). By combining two photons photodamage is significantly reduced within the sample and the high localization of the excitation within the sample allows optical sectioning.
fluorescence increasing the potential for photobleaching within the focal volume. This limits TPF to pulsed laser sources (between 100 femtoseconds and 1 picosecond) which allow peak laser powers (and thus peak density of photons) to remain high while keeping the overall power incident on the sample relatively low. TPF shares many of the same incident light conditions required to generate a signal with CARS (and indeed other nonlinear microscopy techniques) and as such lends itself well to multimodality within a CARS microscope. 1.3. Coherent anti-Stokes Raman scattering Coherent anti-Stokes Raman scattering (CARS) is the nonlinear optical analogue of spontaneous Raman scattering. Unlike two photon processes, it is based on an electronically non-resonant interaction so that energy absorption by the sample is minimized thereby rendering the process largely non-invasive and free from bleaching. To understand the CARS process and its place in the diverse family of nonlinear optical phenomena we begin with the conventional expansion of the induced polarization ~ P in powers of electric field strength:
~ P ¼ vð1Þ E1 þ vð2Þ E1 E2 þ vð3Þ E1 E2 E3
ð1Þ
The v expansion coefficients are the tensor susceptibilities of order n. In linear optics the polarization ~ P is linearly dependent on the electric field strength E1 and the susceptibility vð1Þ is related to the dielectric constant of the material. This leading term accounts for all aspects of linear optical phenomena including refraction, Rayleigh scattering, absorption (when vð1Þ is complex) and birefringence (when the explicit tensor properties are included). Processes arising from the second-order response include second harmonic generation, hyper-Rayleigh scattering and sum/difference frequency generation. This lowest order nonlinear susceptibility is zero for all materials with inversion symmetry. The third order susceptibility is responsible for four-wave mixing in which three optical fields interact to produce a fourth one and at least some of its tensor components are non-zero for any symmetry. Physically, the four-wave mixing process can be understood by recognizing first that a single input field causes an ðnÞ
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
oscillating polarization in a dielectric medium. This is re-radiated (and phase-shifted) as Rayleigh scattering (linear optics). Addition of a second optical field also drives oscillations of the dielectric and the interference of the two leads to harmonics in the polarization at sum and difference frequencies. If a third field is introduced it beats with the sum and difference frequencies generating the fourth field. The coupling between the waves occurs only when energy and momentum are both conserved. Coherent Anti-Stokes Raman scattering (first demonstrated in 1965) [6] is a special case of this four-wave mixing process involving two pump and one Stokes input fields (with frequencies xp and xS , respectively). The energy level scheme and phase matching conditions describing these processes are shown in Fig. 1. The pump and Stokes fields coherently stimulate oscillations at the difference frequency xp xS . A third incident photon (typically also from the pump beam) is then scattered leading to emission at xCARS ¼ 2xp xS as prescribed by energy conservation. The emission is therefore blue-shifted relative to the incident beams which helps detection above one-photon-induced autofluorescence. The emission direction is determined by the momentum conservation or phase matching condition which requires jDkj n ¼ ð~ kCARS ð2~ kp ~ ks ÞÞ n p where n is a measure of the interaction length. For the pump and Stokes input fields the expression for vð3Þ has the general structure (See for example Lotem et al.) [7]
vð3Þ ¼
SR
XR ðxp xS Þ iCR
þ
Stp
xtp xp iCtp
ð3Þ þ vnonres:
ð2Þ
Term 1 accounts for Raman scattering events where XR corresponds to a Raman mode frequency (see Figure), SR is the Raman cross-section and CR is the corresponding linewidth. Term 2 arises from twophoton electronic resonances in which the corresponding cross-section and linewidth are given by Ctp and Stp , respectively. The third term is a background non-resonant susceptibility. It is approximately constant due to its nearly flat frequency response. From this expression it is clear that vibrationally-resonant enhancement of vð3Þ and the corresponding CARS signal occurs when the frequency difference xp xS is equal to a that of a vibrational Raman transition XR in the specimen. The intensity of the CARS emission depends on the square of the number of scatterers in the focal volume and also non-linearly on the excitation field intensities (Ip and IS , respectively) according to ICARS / jvð3Þ j2 I2p IS and can be several orders of magnitude larger than the spontaneous Raman signal. The sensitivity of the CARS technique is limited by the background arising from the vibrationally non-resonant electronic contributions to vð3Þ . The polarization properties of CARS emission have been explored in detail by Oudar et al. [8] The important result for imaging applications is that if the orientation of linearly polarized pump and Stokes beams is set so that they are inclined at an angle relative to each other then the polarization properties of the resonant and non-resonant parts of the third-order polarization will be different. This difference can be exploited to reduce the non-resonant contribution using polarization-resolved detection as we will see later. Finally we note that the nonlinear CARS signal is only generated at the focal point where the intensities of the optical fields are maximal. Therefore, as with multi-photon fluorescence imaging, CARS is also inherently confocal and can, in principle, be used for 3D optical sectioning. 1.4. Technical developments The basic physical principle of the CARS process outlined above was first exploited for imaging applications in Duncan’s CARS microscope [9]. This first system successfully demonstrated the principle of CARS microscopy and was capable of imaging in wavenumber
3
ranges >2000 cm1 using visible radiation from picosecond dye lasers. However the electronic contributions to vð3Þ degraded sensitivity. The use of visible light sources exacerbates this problem by exciting two-photon resonances (term 2 in vð3Þ ) thereby increasing the background. It was therefore evident from the earliest days of CARS microscopy that efficient methods for reduction of the nonresonant background were necessary if it was to become a powerful and versatile imaging method [10]. However, these did not come quickly. It was not until almost twenty years later when advances in ultrafast light sources enabled Zumbusch et al. to use tightly focussed near infrared (femtosecond) pulses to improve sensitivity by avoiding two-photon electronic resonances [11]. Further improvements have been realized by narrowing spectral bandwidths and polarization-resolved detection [12] and possibly the use of chirped laser pulses [13] and coherent pulse shaping techniques [14]. The latter method aims for maximal background suppression by exploiting the different spectral responses of the resonant and non-resonant terms of the susceptibility tensor. A further technical step in CARS imaging of heterogeneous samples comprising objects in solution (including intracellular constituents) came from the realization that the CARS radiation profile depends on the physical size of the scatterer. This is a consequence of the fact that for objects that are small compared to the pump excitation wavelength (i.e., those with small interaction lengths n) the phase matching condition is relaxed and can be met for all ~ kCARS . The CARS radiation profile is therefore effectively symmetric in the forward and backward directions for copropagating excitation beams. This is not the case for larger objects where the phase matching condition is more restrictive and the CARS emission is biased towards the forward scattering direction. This property can also be exploited to improve image quality in certain cases. Specifically, Forward-detected CARS (F-CARS) microscopy emerges as most suitable for imaging objects of a size comparable to or larger than the excitation wavelength. For smaller objects, the F-CARS signal becomes dominated by the non-resonant solvent background. In this case, however, the backpropagating resonant component can be detected in so-called Epi-geometry thereby rejecting the forward-biased background. Epi-CARS (E-CARS) [15] therefore provides a sensitive means of imaging objects having an axial length much smaller than the excitation wavelength. A full discussion of the radiation pattern and its dependence on sample geometry (shape as well as size) is given by Cheng [16]. For CARS to be capable of monitoring dynamical processes in living cells or other systems increasing acquisition speed is critically important. Initially, CARS microscopy images were constructed by mechanical scanning of the sample. The resulting data collection speed was prohibitively slow. This too has largely been overcome by scanning the two collinearly combined laser beams over a static specimen [17]. Acquisition times are reduced by two orders of magnitude. Scanning of the two beams of different wavelength does however cause potential aberration that limits the region of overlap of the two focal spots and thereby the effective field of view. The use of CARS imaging in real biological environments [18] is still at an early stage with relatively few but very active research groups engaged in both technique development and applications. So far effort has focussed on strong CARS signals such as those observed from lipids [19–21]. These are rich in aliphatic CH2 modes which have a prominent stretch feature around 2,800 cm1. Despite their postulated importance in membrane biology, direct observation of lipid rafts in cells remains a challenge: Potma and Xie have used CARS to see separate lipid domains within a single lipid bilayer [22]. For their initial demonstration, they use a model system of giant unilamellar vesicles made of a 50–50 mixture of two lipids, 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC). At room temperature, these lipids form two distinct phases. The two lipids show only
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
4
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
minor differences in the C–H stretching vibrations, which provide only a slight contrast for the image. To improve the contrast, Xie and Potma use deuterated DSPC and regular DOPC. The greater difference between the C–D and the C–H stretching vibrations makes it easy to distinguish the two lipids when they segregate into separate domains. A further increase in speed has been enabled by so-called multiplex CARS imaging [23–26]. Here the same physical principle applies as in conventional single-frequency CARS. The difference is that Stokes light source is not scanned in frequency. Instead a broadband Stokes source is used to allow simultaneous resonant excitation over a wide range of Raman frequencies XR . The broadband source can be generated from ultrafast light pulses passed through a tapered optical fiber. CARS microscopy thus has the potential of becoming a major intrinsically chemo-specific optical imaging technology with the following specific advantages: Vibrational contrast is intrinsic to complex samples. Fluorescent probes are not necessary problems with photobleaching, cytotoxicity or other undesirable effects are circumvented. Strong and directional signals are obtainable at only modest average incident laser power tolerable by biological samples. CARS being a non-linear process the signal provides inherent confocal 3D optical sectioning. CARS emission is typically higher in frequency than autofluorescence and hence is detectable against an strong background. In the case where near-infrared excitation beams are used, there is reduced scattering thereby allowing deeper penetration. 1.5. CARS microscopy - methods and beam delivery We describe here a CARS system consisting of two Coherent Mira titanium-sapphire lasers (Mira 900-P and 900-D) pumped
by a Coherent Verdi V-10 diode-pumped solid-state laser. Our apparatus uses the 785 nm output of a Ti:Sapphire oscillator (200 mW, 76 MHz, 150 fs pulses) fed into a tapered nonlinear fiber, 9 mm long with a 2 mm diameter waist to create a continuum that spans the wavelength range (500–1100 nm). A spectrally narrow portion of the same oscillator output is used for the pump and probe light. This allows the flexibility of tuning necessary for creating the correct wavelength spacing for producing Raman shifts of the magnitude of 2000–3000 cm1. The two lasers are overlapped spatially using the optical setup shown in Fig. 2, a combination of mirror M3 and a beamsplitter (BS). A pulse locked loop system (Coherent Synchrolock) is used to maintain overlap between two laser pulses temporally at pulse lengths of 2 ps and repetition rate of around 80 MHz. Unfortunately, as the two laser beams derive from different laser cavities any slight alteration to the cavity length of one of the lasers will cause the two pulses to walk away from each other. For a CARS signal to be generated the two laser pulses must be precisely overlapped both spatially and temporally at the sample. The Synchrolock contains a phase shifter that allows this to be achieved by adding an arbitrary phase ð0 360 Þ to one of the pulse trains. To ensure the beams are overlapped a portion of the two incident beams is passed through two photodiodes a slow GaAsP photodiode (G1116 Hamamatsu) and a fast silicon P-I-N chip (818-BB-02, Newport). The fast photodiode allows the user to resolve individual pulses meaning the beams can be brought approximately into phase. The response time of this photodiode is not enough to completely overlap the pulse trains. It is therefore necessary to use the slow photodiode, which with a rise time of 4000 ns integrates the signal over several hundred pulses, to fine-tune the phase correlation until the trains are completely aligned within the sample. Neutral Density (ND) filters are used to control the pulse power from each laser beam. ND1 and ND2 primarily control the ratio of the two beams matches the 2 : 1 ratio of pump to Stokes beams as outlined in xCARS ¼ 2xp xS . ND3 is used to tune the power
Fig. 2. Schematic setup of our CARS/Multiphoton microscope. Two Coherent Mira lasers are overlapped spatially using mirror M 3 and beamsplitter (BS). The pulses of the lasers are overlapped temporally using the Phase locked loops (PLLs) and fine-tuned using a GaAsP photodiode and Silicon P-I-N chip. The beams are passed through an aperture (diameter W 0 ) and Faraday Isolator before being expanded to 5 mm diameter by lenses L1 and L2 then to fill the back focal plane of the microscope objective by Lscan and Ltube . Galvonometer scanning mirrors Mx and M y allow fast raster scanning across the sample plane with acquisition times of a few seconds for a 5122 pixel image. Signal wavelengths are separated from the incident by longpass dichroic mirrors (DM) and further filtered using bandpass filters before being detected using photomultiplier tubes (PMT). Brightfield images are obtained using a CCD camera illuminated either with a halogen light or an infrared LED.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
5
Fig. 3. Left. Image of polystyrene beads of diameter 1 lm imaged using the CARS setup. Beads are clearly distinguished both from the background and from beads directly next to each other with a pixel per micron resolution in the lateral direction of as great as 90:65 4:00. The generated CARS signal is vastly stronger than the non-resonant background effects and the signal-to-noise ratio (SNR) is as great as 3 : 1. Right. Image of a polystyrene bead of diameter 2 lm imaged using the CARS setup. The resolution of the system is greatly enhanced compared to previous images obtained using the system.
Fig. 5. The growth of two ecoli MG1655 microcolonies on M9 agarose as imaged using two-photon microscopy. The ecoli contain a GFP expressing plasmid (pcH60) which is excited using light of wavelength 896 nm causing two-photon fluorescence of the GFP protein. Signal is acquired in PMT2 allowing the possibility of simultaneously imaging a colony using CARS and multiphoton microscopy. Image intervals are at 30 min and are chosen to be well within the reproduction rate of the bacteria. The sample was imaged from between 17.5 and 25.5 h after the sample was placed and sealed onto the M9 agarose. The scale bars are 5 lm.
Fig. 4. Z stack of MG1655 microcolony 25.5 h after being placed and sealed onto M9 agarose. Images are acquired using epi-detection of two-photon excited GPF plasmid pcH60 within the E. coli. The stacks are produced by imaging from the coverslip into the agarose using an inverted method. Each slice is separated by 0.2 lm in the z direction. The scale bars are 5 lm.
incident on the sample to limit effects of photobleaching from secondary processes (e.g. Two-photon fluorescence).
The overlapped beams are passed first through an aperture of diameter w0 and then a Faraday Isolator in order to limit effects of back-reflections on the stability of the laser cavities. To ensure the maximum CARS signal is obtained the beams are expanded to fill the back aperture of the microscope objective. This is done using two sets of lenses. Lenses L1 and L2 perform an initial beam expansion from 0.8 to 5 mm. The beams are then passed through the x and y scanning mirrors, Mx and My mounted on scanners (6215HSM60, Cambridge Technology). These mirrors and the controlling servoamplifier circuits (677215HHJ, Cambridge Technology) allow for fast raster scanning through the sample plane. The beams are then expanded to fill the back aperture of the microscope objective by lenses Lscan and Ltube . The CARS microscope has been designed to allow maximum flexibility while maintaining rigidity and simplicity of use. The user has the option of utilising either a commercial microscope (NIKON Eclipse TE300) or a custom-built system to achieve the desired images. The custom-built microscope allows simultaneous acquisition from two photomultiplier tubes (R2896, Hamamatsu), one setup for forward detection, the other for epi-detection. This allows multimodal acquisitions in real time. The desired CARS signal is separated from the incident wavelengths by long wave pass dichroic mirrors (683dcxr Chroma) and further filtered using two band pass filters (HQ590/20, Chroma). Bright field images are recorded using
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
6
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
a CCD camera illuminated by either a halogen light or infrared LED depending on the sensitivity of the sample to external light. Either microscope can be fitted with a piezo objective scanner (Physik Instrumente) below the microscope objective to allow optical sectioning of the sample up to 100 lm. The detected signal is displayed using a custom-built program adapted from Nonlinear Optical Microscopy Scanning and Imaging System (NOMSIS) provided by Anthony Lee of Texas A& M University. 1.6. Methods of image optimization CARS images of bulk dodecane (a liquid hydrocarbon with a large proportion of C-H aliphatic bonds) are initially imaged. Bulk measurements are taken to allow characterisation of the area within the sample plane where CARS signals are generated. Laser wavelengths of 896 and 714 nm are used for the master and slave lasers respectively to give a Raman shift of 2845 cm1. CARS signals are optimised using polystyrene beads of fixed diameter (2 and 1 lm respectively) dried onto a glass coverslip. This system geometry directly mirrors the geometry used for live imaging of samples but provides a stable and reproducable medium from which to test the signal characteristics. Fig. 3 shows polystyrene beads of 1 lm (left) and 2 lm (right) diameter as imaged using the above settings. Lateral resolution is shown to be as great as 90.65 ± 4.00 pixels per micron which is of a quality greater than that required to image at a sub-cellular level (cellular level is defined here as being of the order of 1 lm) and signalto-noise ratio (SNR) is greater than 3:1. This is in excess of
diffraction-limited conventional confocal Raman microscopy techniques [27]. 1.7. Live cell imaging Using the microscope system described above we describe imaging of an Escherichia coli colony: E. coli samples of strain K-12 MG1655 containing a GFP expressing plasmid (pcH60) are grown from overnight stationary phase to exponential phase in LB and tetracycline at a concentration of 1 ll per ml. Samples are then diluted in PBS to concentrations of the magnitude of 1 109 cells per ml. 1.7.1. Microscope slide Prepared samples are imaged using a custom-made microscope slide. The microscope slide is fitted with a GeneFrame (Thermo Scientific) which is filled with a low fluorescing media, M9 agarose. 5 ll of E. coli solution is pipetted onto the surface of the M9 agarose and allowed to dry before a coverslip is secured in place. The sealed microscope slide is inverted and placed onto the custom-build microscope and a suitable bacterium is selected for imaging using the brightfield microscope. 1.7.2. Axial resolution Stacks of images of E. coli (MG1655) were recorded using excitation of GFP plasmid pcH60 inserted within the bacteria directly. The sample is scanned initially at low pass rate and short dwell time per pixel (2562 pixel image at 20 ls dwell per pixel) to locate
Fig. 6. (A) Forward CARS image of fibroblast cells using the C-H aliphatic bond excitation at 2845 cm1 during their infection with mouse cytomegalovirus genetically modified to express GFP with the immediate early promoter 3. Two-photon fluorescence images of the corresponding area can be made simultaneously using the epimounted PMT (B). This allows direct comparison of the two modalities and significantly enhances understanding. Superimposing the CARS and two-photon images (D) allows the user to distinguish directly between lipid droplets and inclusion droplets (both showing black on the two-photon image but lipids showing bright red in the CARS image). The brightfield image of the sample area (C) is made using the infrared diode.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
the bacterium on the PMT output. The pixel dwell is then increased for acquistion of high resolution images (5122 pixels) and scanned at intervals of 0.2 lm in the axial direction. Fig. 4 shows a stack of a microcolony of E. coli imaged 25.5 h after being placed on the sample. Laser power incident on the sample is approximately 10:0 0:1 mW. The pixel dwell is 30 ls and each image plane is imaged 3 times (to reduce pixel-to-pixel variance) resulting in a sample plane acquisition time of 24.01 ± 0.01 s. Individual E. coli are clearly distinguishable from each other throughout the stack.
1.7.3. Temporal measurements Stacks of images of the E. coli microcolonies are taken every 30 min to allow for suitable tracking of division of bacterium within the colony. Fig. 5 is a compilation of the image plane centred on the bacterial colony from each stack. Images are separated temporally by 30 min and are from 17.5 h after sealing the microscope slide to 25.5 h. Clear division of the imaged bacteria is seen for both microcolonies at suitable lateral resolution to allow accurate tracking of individual bacteria within the microcolony. Images are obtained with a dwell per pixel of 30 ls and are scanned three times to reduce the pixel to pixel noise. Power incident on the sample is approximately 10.0 ± 0.1 mW and the sample shows no photobleaching during the process of imaging the microcolony over 48 h.
1.7.4. Multimodal imaging of live samples High resolution CARS and two-photon fluorescence microscopies can be combined to image the same sample at the same time. Fig. 6 [28] shows the combination of high resolution CARS (A) and two-photon fluorescence (B) microscopy of the same area of fibroblast cells during infection with a mouse cytomegalovirus modified to express GFP with the immediate early promoter. Superimposing the two techniques (D) it is possible to distinguish between the high concentration lipid droplets (bright red in A and D but black in B) and inclusion droplets (dark in A, B and D). The bright field image of the corresponding region (C) made using the infrared diode as fibroblast cells are susceptable to even small intensities of light is included for comparison. CARS images are taken using dwell per pixels of 40 lm and are averaged over 4 scans of the sample plane resulting in an image acquisition time of about 11 s. The power incident on the sample is estimated to be 28 mW.
7
References [1] D.T. Schaafsma, R. Mossadegh, J.S. Sanghera, I.D. Aggarwal, J.M. Gilligan, N.H. Tolk, M. Luce, R. Generosi, P. Perfetti, A. Cricenti, G. Margaritondo, Ultramicroscopy 77 (1-2) (1999) 77–81. [2] D. Vobornik, G. Margaritondo, J.S. Sanghera, P. Thielen, I.D. Aggarwal, B. Ivanov, N.H. Tolk, V. Manni, S. Grimaldi, A. Lisi, S. Rieti, D.W. Piston, R. Generosi, M. Luce, P. Perfetti, A. Cricenti, J. Alloys Compd. 401 (1–2) (2005) 80–85. [3] Bruno Pettinger, Bin Ren, Gennaro Picardi, Rolf Schuster, Gerhard Ertl, Phys. Rev. Lett. 92 (9) (2004) 096101. [4] Martin Oheim, Darren J. Michael, Matthias Geisbauer, Dorte Madsen, Robert H. Chow, Adv. Drug Deliv. Rev. 58 (7) (2006) 788–808. [5] C. Buehler, C.Y. Dong, B.R. Masters, P.T.C. So, IEEE Eng. Med. Biol. Magazine 18 (5) (1999) 23–30. [6] P.D. Maker, R.W. Terhune, Phys. Rev. 137 (3A) (1965) A801–A818. [7] Haim Lotem, R.T. Lynch, N. Bloembergen, Phys. Rev. A 14 (5) (1976) 1748– 1755. [8] JeanLouis Oudar, Robert W. Smith, Y.R. Shen, Appl. Phys. Lett. 34 (11) (1979) 758–760. [9] M.D. Duncan, J. Reintjes, T.J. Manuccia, Opt. Lett. 7 (8) (1982) 350–352. [10] Eric O. Potma, David J. Jones, J.-X. Cheng, X.S. Xie, Jun Ye, Opt. Lett. 27 (13) (2002) 1168–1170. [11] Andreas Zumbusch, Gary R. Holtom, X. Sunney Xie, Phys. Rev. Lett. 82 (20) (1999) 4142–4145. [12] Ji-Xin Cheng, Lewis D. Book, X. Sunney Xie, Opt. Lett. 26 (17) (2001) 1341– 1343. [13] K.P. Knutsen, J.C. Johnson, a.E. Miller, P.B. Petersen, R.J. Saykally, Chem. Phys. Lett. 387 (4–6) (2004) 436–441. [14] Dan Oron, Nirit Dudovich, Yaron Silberberg, Phys. Rev. Lett. 90 (21) (2003) 213902. [15] Ji-xin Cheng, Andreas Volkmer, Lewis D. Book, X. Sunney Xie, J. Phys. Chem. B 105 (2001) 1277–1280. [16] Ji-Xin Cheng, Andreas Volkmer, X. Sunney Xie, J. Opt. Soc. Am. B 19 (6) (2002) 1363–1375. [17] Ji-Xin Cheng, Andreas Volkmer, X. Sunney Xie, J. Opt. Soc. Am. B 19 (6) (2002) 1363–1375. [18] Eric O. Potma, Wim P. de Boeij, Peter J.M. van Haastert, Douwe A. Wiersma, Proc. Natl. Acad. Sci. 98 (4) (2001) 1577–1582. [19] Ji-xin Cheng, Eric O. Potma, Sunney X. Xie, J. Phys. Chem. A 106 (37) (2002) 8561–8568. [20] Richard D. Schaller, Joseph Ziegelbauer, Lynn F. Lee, Louis H. Haber, Richard J. Saykally, J. Phys. Chem. B 106 (34) (August 2002) 8489–8492. [21] Xiaolin Nan, Ji-Xin Cheng, X. Sunney Xie, J. Lipid Res. 44 (11) (2003) 2202– 2208. [22] Eric O. Potma, X. Sunney Xie, Chemphyschem: Eur. J. Chem. Phys. Phys. Chem. 6 (1) (January 2005) 77–79. [23] Ji-xin Cheng, Andreas Volkmer, Lewis D. Book, X. Sunney Xie, J. Phys. Chem. B 106 (34) (August 2002) 8493–8498. [24] Michiel Müller, Juleon M. Schins, J. Phys. Chem. B 106 (14) (April 2002) 3715– 3723. [25] Tak W. Kee, Marcus T. Cicerone, Opt. Lett. 29 (23) (2004) 2701–2703. [26] Hideaki Kano, Hiro-O Hamaguchi, Opt. Exp. 13 (4) (February 2005) 1322– 1327. [27] Vasilis Ntziachristos, Nat. Meth. 7 (8) (August 2010) 603–614. [28] Iain Robinson, Michael Andreas Ochsenkühn, Colin J. Campbell, Gerard Giraud, William J. Hossack, Jochen Arlt, Jason Crain, J. Biophot. 3 (3) (August 2010) 138–146.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
Contents lists available at ScienceDirect
Methods journal homepage: www.elsevier.com/locate/ymeth
Nonlinear optical methods for cellular imaging and localization A. McVey a, J. Crain a,b,⇑ a b
School of Physics, The University of Edinburgh, EH9 3JZ Scotland, United Kingdom National Physical Laboratory Teddington, TW11 0LW, United Kingdom
a r t i c l e
i n f o
Article history: Received 20 December 2013 Revised 28 February 2014 Accepted 2 March 2014 Available online xxxx Keywords: Imaging Nonlinear optics Biophysics Fluorescence Raman spectroscopy
a b s t r a c t Of all the ways in which complex materials (including many biological systems) can be explored, imaging is perhaps the most powerful because delivering high information content directly. This is particular relevant in aspects of cellular localization where the physical proximity of molecules is crucial in biochemical processes. A great deal of effort in imaging has been spent on enabling chemically selective imaging so that only specific features are revealed. This is almost always achieved by adding fluorescent chemical labels to specific molecules. Under appropriate illumination conditions only the molecules (via their labels) will be visible. The technique is simple and elegant but does suffer from fundamental limitations: (1) the fluorescent labels may fade when illuminated (a phenomenon called photobleaching) thereby constantly decreasing signal contrast over the course of image acquisition. To combat photobleaching one must reduce observation times or apply unfavourably low excitation levels all of which reduce the information content of images; (2) the fluorescent species may be deactivated by various environmental factors (the general term is fluorescence quenching); (3) the presence of fluorescent labels may introduce unexpected complications or may interfere with processes of interest (4) Some molecules of interest cannot be labelled. In these circumstances we require a fundamentally different strategy. One of the most promising alternative is based on a technique called Coherent Anti-Stokes Raman scattering (CARS). CARS is a fundamentally more complex process than is fluorescence and the experimental procedures and optical systems required to deliver high quality CARS images are intricate. However, the rewards are correspondingly very high: CARS probes the chemically distinct vibrations of the constituent molecules in a complex system and is therefore also chemically selective as are fluorescence-based methods. Moreover,the potentially severe problems of fluorescence bleaching and quenching are circumvented and high-resolution three dimensional images can be obtained on completely unlabelled specimens. We review here aspects of CARS and Multiphoton fluorescence techniques to cellular localization and measurement. Ó 2014 Published by Elsevier Inc.
1. Motivation Despite their chemical specificity, high resolution and optical sectioning capability, fluorescence-based imaging techniques all share one fundamental limitation. Namely, that the molecular constituents of interest must be stained by incorporating ’’foreign’’ labels to create image contrast. In the most benign cases, these labels act as robust but passive dyes which have no effect other than to render visible the molecules to which they are attached. In some circumstances, however, the dyes are neither robust nor passive. Specifically, prolonged exposure to the excitation beam often leads to permanent damage to the dye molecules. This ⇑ Corresponding author. School of Physics, The University of Edinburgh, EH9 3JZ Scotland, United Kingdom E-mail address: [email protected] (J. Crain).
photobleaching degrades image contrast and forces compromises between illumination intensity, image quality and acquisition speed. Image quality may also suffer from fluorescence quenching whereby non-radiative electronic relaxation processes become significant and suppress fluorescence yield. Beyond these practical issues connected to obvious loss of signal, other problems are more subtle. Namely, free radical reaction products generated during photobleaching may be toxic for biological specimens. Moreover, fluorescent labelling itself can cause non-negligible perturbations to chemical or biophysical pathways and to cellular functions the severity of which cannot be assessed a priori. Despite intensive worldwide efforts to design more robust fluorescent labels with tailored properties and the use of further chemical additives and stabilizers such as anti-fading agents, new imaging modalities that do not rely on fluorescence at all offer a very elegant alternative for applications across the physical and life sciences. Multiphoton
http://dx.doi.org/10.1016/j.ymeth.2014.03.002 1046-2023/Ó 2014 Published by Elsevier Inc.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
2
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
methods partially mitigate these risks and we consider aspects of the method below. An alternative strategy involves the inherent and specific molecular vibrations of molecules in complex systems as the basis for imaging contrast in unstained samples. In principle, either infrared (IR) or Raman spectroscopy is potentially suitable as the basis for vibrational microscopies but both have serious disadvantages in practical applications. Specifically, the long excitation wavelengths required for IR imaging severely degrade spatial resolution. Also, strong water absorption complicates applications in aqueous systems including biological cells. The combination of infrared spectroscopy with scanning nearfield optical microscopy (IR-SNOM) [1,2] or tip-enhanced Raman spectroscopy (TERS) [3] can provide signal enhancement and spatial resolution below the diffraction limit. Raman scattering avoids these limitations through the use of visible excitation sources but the intrinsically small scattering cross section makes detection very difficult (especially in the presence of background autofluorescence) and usually precludes application in 3D microscopy of biological systems. In what follows we will show that a multiphoton variant of spontaneous Raman scattering is emerging as a powerful tool for 3D microscopy of complex systems and that it may offer significant advantages over incoherent laser-induced fluorescence and spontaneous Raman techniques for certain applications. 1.1. Physical principles . 1.2. Two-photon fluorescence Insertion of fluorophores into samples allows samples otherwise indistinct to light microscopy to be examined with far greater resolution and accuracy. By exciting an electron within the flourophore from the ground state to a higher electronic state the emission of a photon at a Stokes shifted (red-shifted) frequency can be observed. Single-photon fluorescence is induced by a single photon causing the excitation and as such the excitation is proportional to the intensity (I) of the incident light as every photon has an equal probability of causing emission. It is therefore necessary for the photon to be of high-energy in order to match the required excitation gap (typically UV or visible light). Two-photon fluorescence (TPF) occurs when two photons are absorbed simultaneously (within 10(18) seconds of each other) [4] by an electron inducing a excitation from the ground state to a higher order electronic state. Any two photons which sum to match the energy gap between the ground and excited state are sufficient but in practice they are usually of the same wavelength. As energy is inversely proportional to wavelength, the required wavelengths are twice as long as those necessary for single photon absorption. This increase in wavelength from UV–vis to nearinfrared significantly reduces the photodamage caused by incident photons within the sample and allows for deeper penetration depths for live tissue imaging [5]. Fig. 1 shows the energy level scheme for two photon excitation. Since TPF involves the combination of two photons, the fluorescence emission is proportional to the square of the incident intensity. This phenomenon means that in order to generate TPF a high density of photons must be present in the sample effectively confining TPF to a small focal volume within the sample. This is extremely advantageous for imaging live samples as only the focal volume is subject to photodamage at any one time. Unfortunately the high density of photons (1020–1030 photons per square cm) [4] required to generate TPF means that peak laser powers must be higher than those needed for single photon
Fig. 1. Energy level diagrams for CARS (left) and two-photon fluorescence (right). CARS is an excitation to electronically non-resonant energy levels (light grey) by three incident photons, frequency xp and xS respectively to create a fourth at a blue-shifted frequency, xas . This limits the energy absorption by the sample. Twophoton fluorescence is an electronically resonant excitation to a higher electronic energy level (black) by two incident photons (usually of the same frequency, xS ). By combining two photons photodamage is significantly reduced within the sample and the high localization of the excitation within the sample allows optical sectioning.
fluorescence increasing the potential for photobleaching within the focal volume. This limits TPF to pulsed laser sources (between 100 femtoseconds and 1 picosecond) which allow peak laser powers (and thus peak density of photons) to remain high while keeping the overall power incident on the sample relatively low. TPF shares many of the same incident light conditions required to generate a signal with CARS (and indeed other nonlinear microscopy techniques) and as such lends itself well to multimodality within a CARS microscope. 1.3. Coherent anti-Stokes Raman scattering Coherent anti-Stokes Raman scattering (CARS) is the nonlinear optical analogue of spontaneous Raman scattering. Unlike two photon processes, it is based on an electronically non-resonant interaction so that energy absorption by the sample is minimized thereby rendering the process largely non-invasive and free from bleaching. To understand the CARS process and its place in the diverse family of nonlinear optical phenomena we begin with the conventional expansion of the induced polarization ~ P in powers of electric field strength:
~ P ¼ vð1Þ E1 þ vð2Þ E1 E2 þ vð3Þ E1 E2 E3
ð1Þ
The v expansion coefficients are the tensor susceptibilities of order n. In linear optics the polarization ~ P is linearly dependent on the electric field strength E1 and the susceptibility vð1Þ is related to the dielectric constant of the material. This leading term accounts for all aspects of linear optical phenomena including refraction, Rayleigh scattering, absorption (when vð1Þ is complex) and birefringence (when the explicit tensor properties are included). Processes arising from the second-order response include second harmonic generation, hyper-Rayleigh scattering and sum/difference frequency generation. This lowest order nonlinear susceptibility is zero for all materials with inversion symmetry. The third order susceptibility is responsible for four-wave mixing in which three optical fields interact to produce a fourth one and at least some of its tensor components are non-zero for any symmetry. Physically, the four-wave mixing process can be understood by recognizing first that a single input field causes an ðnÞ
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
oscillating polarization in a dielectric medium. This is re-radiated (and phase-shifted) as Rayleigh scattering (linear optics). Addition of a second optical field also drives oscillations of the dielectric and the interference of the two leads to harmonics in the polarization at sum and difference frequencies. If a third field is introduced it beats with the sum and difference frequencies generating the fourth field. The coupling between the waves occurs only when energy and momentum are both conserved. Coherent Anti-Stokes Raman scattering (first demonstrated in 1965) [6] is a special case of this four-wave mixing process involving two pump and one Stokes input fields (with frequencies xp and xS , respectively). The energy level scheme and phase matching conditions describing these processes are shown in Fig. 1. The pump and Stokes fields coherently stimulate oscillations at the difference frequency xp xS . A third incident photon (typically also from the pump beam) is then scattered leading to emission at xCARS ¼ 2xp xS as prescribed by energy conservation. The emission is therefore blue-shifted relative to the incident beams which helps detection above one-photon-induced autofluorescence. The emission direction is determined by the momentum conservation or phase matching condition which requires jDkj n ¼ ð~ kCARS ð2~ kp ~ ks ÞÞ n p where n is a measure of the interaction length. For the pump and Stokes input fields the expression for vð3Þ has the general structure (See for example Lotem et al.) [7]
vð3Þ ¼
SR
XR ðxp xS Þ iCR
þ
Stp
xtp xp iCtp
ð3Þ þ vnonres:
ð2Þ
Term 1 accounts for Raman scattering events where XR corresponds to a Raman mode frequency (see Figure), SR is the Raman cross-section and CR is the corresponding linewidth. Term 2 arises from twophoton electronic resonances in which the corresponding cross-section and linewidth are given by Ctp and Stp , respectively. The third term is a background non-resonant susceptibility. It is approximately constant due to its nearly flat frequency response. From this expression it is clear that vibrationally-resonant enhancement of vð3Þ and the corresponding CARS signal occurs when the frequency difference xp xS is equal to a that of a vibrational Raman transition XR in the specimen. The intensity of the CARS emission depends on the square of the number of scatterers in the focal volume and also non-linearly on the excitation field intensities (Ip and IS , respectively) according to ICARS / jvð3Þ j2 I2p IS and can be several orders of magnitude larger than the spontaneous Raman signal. The sensitivity of the CARS technique is limited by the background arising from the vibrationally non-resonant electronic contributions to vð3Þ . The polarization properties of CARS emission have been explored in detail by Oudar et al. [8] The important result for imaging applications is that if the orientation of linearly polarized pump and Stokes beams is set so that they are inclined at an angle relative to each other then the polarization properties of the resonant and non-resonant parts of the third-order polarization will be different. This difference can be exploited to reduce the non-resonant contribution using polarization-resolved detection as we will see later. Finally we note that the nonlinear CARS signal is only generated at the focal point where the intensities of the optical fields are maximal. Therefore, as with multi-photon fluorescence imaging, CARS is also inherently confocal and can, in principle, be used for 3D optical sectioning. 1.4. Technical developments The basic physical principle of the CARS process outlined above was first exploited for imaging applications in Duncan’s CARS microscope [9]. This first system successfully demonstrated the principle of CARS microscopy and was capable of imaging in wavenumber
3
ranges >2000 cm1 using visible radiation from picosecond dye lasers. However the electronic contributions to vð3Þ degraded sensitivity. The use of visible light sources exacerbates this problem by exciting two-photon resonances (term 2 in vð3Þ ) thereby increasing the background. It was therefore evident from the earliest days of CARS microscopy that efficient methods for reduction of the nonresonant background were necessary if it was to become a powerful and versatile imaging method [10]. However, these did not come quickly. It was not until almost twenty years later when advances in ultrafast light sources enabled Zumbusch et al. to use tightly focussed near infrared (femtosecond) pulses to improve sensitivity by avoiding two-photon electronic resonances [11]. Further improvements have been realized by narrowing spectral bandwidths and polarization-resolved detection [12] and possibly the use of chirped laser pulses [13] and coherent pulse shaping techniques [14]. The latter method aims for maximal background suppression by exploiting the different spectral responses of the resonant and non-resonant terms of the susceptibility tensor. A further technical step in CARS imaging of heterogeneous samples comprising objects in solution (including intracellular constituents) came from the realization that the CARS radiation profile depends on the physical size of the scatterer. This is a consequence of the fact that for objects that are small compared to the pump excitation wavelength (i.e., those with small interaction lengths n) the phase matching condition is relaxed and can be met for all ~ kCARS . The CARS radiation profile is therefore effectively symmetric in the forward and backward directions for copropagating excitation beams. This is not the case for larger objects where the phase matching condition is more restrictive and the CARS emission is biased towards the forward scattering direction. This property can also be exploited to improve image quality in certain cases. Specifically, Forward-detected CARS (F-CARS) microscopy emerges as most suitable for imaging objects of a size comparable to or larger than the excitation wavelength. For smaller objects, the F-CARS signal becomes dominated by the non-resonant solvent background. In this case, however, the backpropagating resonant component can be detected in so-called Epi-geometry thereby rejecting the forward-biased background. Epi-CARS (E-CARS) [15] therefore provides a sensitive means of imaging objects having an axial length much smaller than the excitation wavelength. A full discussion of the radiation pattern and its dependence on sample geometry (shape as well as size) is given by Cheng [16]. For CARS to be capable of monitoring dynamical processes in living cells or other systems increasing acquisition speed is critically important. Initially, CARS microscopy images were constructed by mechanical scanning of the sample. The resulting data collection speed was prohibitively slow. This too has largely been overcome by scanning the two collinearly combined laser beams over a static specimen [17]. Acquisition times are reduced by two orders of magnitude. Scanning of the two beams of different wavelength does however cause potential aberration that limits the region of overlap of the two focal spots and thereby the effective field of view. The use of CARS imaging in real biological environments [18] is still at an early stage with relatively few but very active research groups engaged in both technique development and applications. So far effort has focussed on strong CARS signals such as those observed from lipids [19–21]. These are rich in aliphatic CH2 modes which have a prominent stretch feature around 2,800 cm1. Despite their postulated importance in membrane biology, direct observation of lipid rafts in cells remains a challenge: Potma and Xie have used CARS to see separate lipid domains within a single lipid bilayer [22]. For their initial demonstration, they use a model system of giant unilamellar vesicles made of a 50–50 mixture of two lipids, 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC). At room temperature, these lipids form two distinct phases. The two lipids show only
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
4
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
minor differences in the C–H stretching vibrations, which provide only a slight contrast for the image. To improve the contrast, Xie and Potma use deuterated DSPC and regular DOPC. The greater difference between the C–D and the C–H stretching vibrations makes it easy to distinguish the two lipids when they segregate into separate domains. A further increase in speed has been enabled by so-called multiplex CARS imaging [23–26]. Here the same physical principle applies as in conventional single-frequency CARS. The difference is that Stokes light source is not scanned in frequency. Instead a broadband Stokes source is used to allow simultaneous resonant excitation over a wide range of Raman frequencies XR . The broadband source can be generated from ultrafast light pulses passed through a tapered optical fiber. CARS microscopy thus has the potential of becoming a major intrinsically chemo-specific optical imaging technology with the following specific advantages: Vibrational contrast is intrinsic to complex samples. Fluorescent probes are not necessary problems with photobleaching, cytotoxicity or other undesirable effects are circumvented. Strong and directional signals are obtainable at only modest average incident laser power tolerable by biological samples. CARS being a non-linear process the signal provides inherent confocal 3D optical sectioning. CARS emission is typically higher in frequency than autofluorescence and hence is detectable against an strong background. In the case where near-infrared excitation beams are used, there is reduced scattering thereby allowing deeper penetration. 1.5. CARS microscopy - methods and beam delivery We describe here a CARS system consisting of two Coherent Mira titanium-sapphire lasers (Mira 900-P and 900-D) pumped
by a Coherent Verdi V-10 diode-pumped solid-state laser. Our apparatus uses the 785 nm output of a Ti:Sapphire oscillator (200 mW, 76 MHz, 150 fs pulses) fed into a tapered nonlinear fiber, 9 mm long with a 2 mm diameter waist to create a continuum that spans the wavelength range (500–1100 nm). A spectrally narrow portion of the same oscillator output is used for the pump and probe light. This allows the flexibility of tuning necessary for creating the correct wavelength spacing for producing Raman shifts of the magnitude of 2000–3000 cm1. The two lasers are overlapped spatially using the optical setup shown in Fig. 2, a combination of mirror M3 and a beamsplitter (BS). A pulse locked loop system (Coherent Synchrolock) is used to maintain overlap between two laser pulses temporally at pulse lengths of 2 ps and repetition rate of around 80 MHz. Unfortunately, as the two laser beams derive from different laser cavities any slight alteration to the cavity length of one of the lasers will cause the two pulses to walk away from each other. For a CARS signal to be generated the two laser pulses must be precisely overlapped both spatially and temporally at the sample. The Synchrolock contains a phase shifter that allows this to be achieved by adding an arbitrary phase ð0 360 Þ to one of the pulse trains. To ensure the beams are overlapped a portion of the two incident beams is passed through two photodiodes a slow GaAsP photodiode (G1116 Hamamatsu) and a fast silicon P-I-N chip (818-BB-02, Newport). The fast photodiode allows the user to resolve individual pulses meaning the beams can be brought approximately into phase. The response time of this photodiode is not enough to completely overlap the pulse trains. It is therefore necessary to use the slow photodiode, which with a rise time of 4000 ns integrates the signal over several hundred pulses, to fine-tune the phase correlation until the trains are completely aligned within the sample. Neutral Density (ND) filters are used to control the pulse power from each laser beam. ND1 and ND2 primarily control the ratio of the two beams matches the 2 : 1 ratio of pump to Stokes beams as outlined in xCARS ¼ 2xp xS . ND3 is used to tune the power
Fig. 2. Schematic setup of our CARS/Multiphoton microscope. Two Coherent Mira lasers are overlapped spatially using mirror M 3 and beamsplitter (BS). The pulses of the lasers are overlapped temporally using the Phase locked loops (PLLs) and fine-tuned using a GaAsP photodiode and Silicon P-I-N chip. The beams are passed through an aperture (diameter W 0 ) and Faraday Isolator before being expanded to 5 mm diameter by lenses L1 and L2 then to fill the back focal plane of the microscope objective by Lscan and Ltube . Galvonometer scanning mirrors Mx and M y allow fast raster scanning across the sample plane with acquisition times of a few seconds for a 5122 pixel image. Signal wavelengths are separated from the incident by longpass dichroic mirrors (DM) and further filtered using bandpass filters before being detected using photomultiplier tubes (PMT). Brightfield images are obtained using a CCD camera illuminated either with a halogen light or an infrared LED.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
5
Fig. 3. Left. Image of polystyrene beads of diameter 1 lm imaged using the CARS setup. Beads are clearly distinguished both from the background and from beads directly next to each other with a pixel per micron resolution in the lateral direction of as great as 90:65 4:00. The generated CARS signal is vastly stronger than the non-resonant background effects and the signal-to-noise ratio (SNR) is as great as 3 : 1. Right. Image of a polystyrene bead of diameter 2 lm imaged using the CARS setup. The resolution of the system is greatly enhanced compared to previous images obtained using the system.
Fig. 5. The growth of two ecoli MG1655 microcolonies on M9 agarose as imaged using two-photon microscopy. The ecoli contain a GFP expressing plasmid (pcH60) which is excited using light of wavelength 896 nm causing two-photon fluorescence of the GFP protein. Signal is acquired in PMT2 allowing the possibility of simultaneously imaging a colony using CARS and multiphoton microscopy. Image intervals are at 30 min and are chosen to be well within the reproduction rate of the bacteria. The sample was imaged from between 17.5 and 25.5 h after the sample was placed and sealed onto the M9 agarose. The scale bars are 5 lm.
Fig. 4. Z stack of MG1655 microcolony 25.5 h after being placed and sealed onto M9 agarose. Images are acquired using epi-detection of two-photon excited GPF plasmid pcH60 within the E. coli. The stacks are produced by imaging from the coverslip into the agarose using an inverted method. Each slice is separated by 0.2 lm in the z direction. The scale bars are 5 lm.
incident on the sample to limit effects of photobleaching from secondary processes (e.g. Two-photon fluorescence).
The overlapped beams are passed first through an aperture of diameter w0 and then a Faraday Isolator in order to limit effects of back-reflections on the stability of the laser cavities. To ensure the maximum CARS signal is obtained the beams are expanded to fill the back aperture of the microscope objective. This is done using two sets of lenses. Lenses L1 and L2 perform an initial beam expansion from 0.8 to 5 mm. The beams are then passed through the x and y scanning mirrors, Mx and My mounted on scanners (6215HSM60, Cambridge Technology). These mirrors and the controlling servoamplifier circuits (677215HHJ, Cambridge Technology) allow for fast raster scanning through the sample plane. The beams are then expanded to fill the back aperture of the microscope objective by lenses Lscan and Ltube . The CARS microscope has been designed to allow maximum flexibility while maintaining rigidity and simplicity of use. The user has the option of utilising either a commercial microscope (NIKON Eclipse TE300) or a custom-built system to achieve the desired images. The custom-built microscope allows simultaneous acquisition from two photomultiplier tubes (R2896, Hamamatsu), one setup for forward detection, the other for epi-detection. This allows multimodal acquisitions in real time. The desired CARS signal is separated from the incident wavelengths by long wave pass dichroic mirrors (683dcxr Chroma) and further filtered using two band pass filters (HQ590/20, Chroma). Bright field images are recorded using
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
6
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
a CCD camera illuminated by either a halogen light or infrared LED depending on the sensitivity of the sample to external light. Either microscope can be fitted with a piezo objective scanner (Physik Instrumente) below the microscope objective to allow optical sectioning of the sample up to 100 lm. The detected signal is displayed using a custom-built program adapted from Nonlinear Optical Microscopy Scanning and Imaging System (NOMSIS) provided by Anthony Lee of Texas A& M University. 1.6. Methods of image optimization CARS images of bulk dodecane (a liquid hydrocarbon with a large proportion of C-H aliphatic bonds) are initially imaged. Bulk measurements are taken to allow characterisation of the area within the sample plane where CARS signals are generated. Laser wavelengths of 896 and 714 nm are used for the master and slave lasers respectively to give a Raman shift of 2845 cm1. CARS signals are optimised using polystyrene beads of fixed diameter (2 and 1 lm respectively) dried onto a glass coverslip. This system geometry directly mirrors the geometry used for live imaging of samples but provides a stable and reproducable medium from which to test the signal characteristics. Fig. 3 shows polystyrene beads of 1 lm (left) and 2 lm (right) diameter as imaged using the above settings. Lateral resolution is shown to be as great as 90.65 ± 4.00 pixels per micron which is of a quality greater than that required to image at a sub-cellular level (cellular level is defined here as being of the order of 1 lm) and signalto-noise ratio (SNR) is greater than 3:1. This is in excess of
diffraction-limited conventional confocal Raman microscopy techniques [27]. 1.7. Live cell imaging Using the microscope system described above we describe imaging of an Escherichia coli colony: E. coli samples of strain K-12 MG1655 containing a GFP expressing plasmid (pcH60) are grown from overnight stationary phase to exponential phase in LB and tetracycline at a concentration of 1 ll per ml. Samples are then diluted in PBS to concentrations of the magnitude of 1 109 cells per ml. 1.7.1. Microscope slide Prepared samples are imaged using a custom-made microscope slide. The microscope slide is fitted with a GeneFrame (Thermo Scientific) which is filled with a low fluorescing media, M9 agarose. 5 ll of E. coli solution is pipetted onto the surface of the M9 agarose and allowed to dry before a coverslip is secured in place. The sealed microscope slide is inverted and placed onto the custom-build microscope and a suitable bacterium is selected for imaging using the brightfield microscope. 1.7.2. Axial resolution Stacks of images of E. coli (MG1655) were recorded using excitation of GFP plasmid pcH60 inserted within the bacteria directly. The sample is scanned initially at low pass rate and short dwell time per pixel (2562 pixel image at 20 ls dwell per pixel) to locate
Fig. 6. (A) Forward CARS image of fibroblast cells using the C-H aliphatic bond excitation at 2845 cm1 during their infection with mouse cytomegalovirus genetically modified to express GFP with the immediate early promoter 3. Two-photon fluorescence images of the corresponding area can be made simultaneously using the epimounted PMT (B). This allows direct comparison of the two modalities and significantly enhances understanding. Superimposing the CARS and two-photon images (D) allows the user to distinguish directly between lipid droplets and inclusion droplets (both showing black on the two-photon image but lipids showing bright red in the CARS image). The brightfield image of the sample area (C) is made using the infrared diode.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002
A. McVey, J. Crain / Methods xxx (2014) xxx–xxx
the bacterium on the PMT output. The pixel dwell is then increased for acquistion of high resolution images (5122 pixels) and scanned at intervals of 0.2 lm in the axial direction. Fig. 4 shows a stack of a microcolony of E. coli imaged 25.5 h after being placed on the sample. Laser power incident on the sample is approximately 10:0 0:1 mW. The pixel dwell is 30 ls and each image plane is imaged 3 times (to reduce pixel-to-pixel variance) resulting in a sample plane acquisition time of 24.01 ± 0.01 s. Individual E. coli are clearly distinguishable from each other throughout the stack.
1.7.3. Temporal measurements Stacks of images of the E. coli microcolonies are taken every 30 min to allow for suitable tracking of division of bacterium within the colony. Fig. 5 is a compilation of the image plane centred on the bacterial colony from each stack. Images are separated temporally by 30 min and are from 17.5 h after sealing the microscope slide to 25.5 h. Clear division of the imaged bacteria is seen for both microcolonies at suitable lateral resolution to allow accurate tracking of individual bacteria within the microcolony. Images are obtained with a dwell per pixel of 30 ls and are scanned three times to reduce the pixel to pixel noise. Power incident on the sample is approximately 10.0 ± 0.1 mW and the sample shows no photobleaching during the process of imaging the microcolony over 48 h.
1.7.4. Multimodal imaging of live samples High resolution CARS and two-photon fluorescence microscopies can be combined to image the same sample at the same time. Fig. 6 [28] shows the combination of high resolution CARS (A) and two-photon fluorescence (B) microscopy of the same area of fibroblast cells during infection with a mouse cytomegalovirus modified to express GFP with the immediate early promoter. Superimposing the two techniques (D) it is possible to distinguish between the high concentration lipid droplets (bright red in A and D but black in B) and inclusion droplets (dark in A, B and D). The bright field image of the corresponding region (C) made using the infrared diode as fibroblast cells are susceptable to even small intensities of light is included for comparison. CARS images are taken using dwell per pixels of 40 lm and are averaged over 4 scans of the sample plane resulting in an image acquisition time of about 11 s. The power incident on the sample is estimated to be 28 mW.
7
References [1] D.T. Schaafsma, R. Mossadegh, J.S. Sanghera, I.D. Aggarwal, J.M. Gilligan, N.H. Tolk, M. Luce, R. Generosi, P. Perfetti, A. Cricenti, G. Margaritondo, Ultramicroscopy 77 (1-2) (1999) 77–81. [2] D. Vobornik, G. Margaritondo, J.S. Sanghera, P. Thielen, I.D. Aggarwal, B. Ivanov, N.H. Tolk, V. Manni, S. Grimaldi, A. Lisi, S. Rieti, D.W. Piston, R. Generosi, M. Luce, P. Perfetti, A. Cricenti, J. Alloys Compd. 401 (1–2) (2005) 80–85. [3] Bruno Pettinger, Bin Ren, Gennaro Picardi, Rolf Schuster, Gerhard Ertl, Phys. Rev. Lett. 92 (9) (2004) 096101. [4] Martin Oheim, Darren J. Michael, Matthias Geisbauer, Dorte Madsen, Robert H. Chow, Adv. Drug Deliv. Rev. 58 (7) (2006) 788–808. [5] C. Buehler, C.Y. Dong, B.R. Masters, P.T.C. So, IEEE Eng. Med. Biol. Magazine 18 (5) (1999) 23–30. [6] P.D. Maker, R.W. Terhune, Phys. Rev. 137 (3A) (1965) A801–A818. [7] Haim Lotem, R.T. Lynch, N. Bloembergen, Phys. Rev. A 14 (5) (1976) 1748– 1755. [8] JeanLouis Oudar, Robert W. Smith, Y.R. Shen, Appl. Phys. Lett. 34 (11) (1979) 758–760. [9] M.D. Duncan, J. Reintjes, T.J. Manuccia, Opt. Lett. 7 (8) (1982) 350–352. [10] Eric O. Potma, David J. Jones, J.-X. Cheng, X.S. Xie, Jun Ye, Opt. Lett. 27 (13) (2002) 1168–1170. [11] Andreas Zumbusch, Gary R. Holtom, X. Sunney Xie, Phys. Rev. Lett. 82 (20) (1999) 4142–4145. [12] Ji-Xin Cheng, Lewis D. Book, X. Sunney Xie, Opt. Lett. 26 (17) (2001) 1341– 1343. [13] K.P. Knutsen, J.C. Johnson, a.E. Miller, P.B. Petersen, R.J. Saykally, Chem. Phys. Lett. 387 (4–6) (2004) 436–441. [14] Dan Oron, Nirit Dudovich, Yaron Silberberg, Phys. Rev. Lett. 90 (21) (2003) 213902. [15] Ji-xin Cheng, Andreas Volkmer, Lewis D. Book, X. Sunney Xie, J. Phys. Chem. B 105 (2001) 1277–1280. [16] Ji-Xin Cheng, Andreas Volkmer, X. Sunney Xie, J. Opt. Soc. Am. B 19 (6) (2002) 1363–1375. [17] Ji-Xin Cheng, Andreas Volkmer, X. Sunney Xie, J. Opt. Soc. Am. B 19 (6) (2002) 1363–1375. [18] Eric O. Potma, Wim P. de Boeij, Peter J.M. van Haastert, Douwe A. Wiersma, Proc. Natl. Acad. Sci. 98 (4) (2001) 1577–1582. [19] Ji-xin Cheng, Eric O. Potma, Sunney X. Xie, J. Phys. Chem. A 106 (37) (2002) 8561–8568. [20] Richard D. Schaller, Joseph Ziegelbauer, Lynn F. Lee, Louis H. Haber, Richard J. Saykally, J. Phys. Chem. B 106 (34) (August 2002) 8489–8492. [21] Xiaolin Nan, Ji-Xin Cheng, X. Sunney Xie, J. Lipid Res. 44 (11) (2003) 2202– 2208. [22] Eric O. Potma, X. Sunney Xie, Chemphyschem: Eur. J. Chem. Phys. Phys. Chem. 6 (1) (January 2005) 77–79. [23] Ji-xin Cheng, Andreas Volkmer, Lewis D. Book, X. Sunney Xie, J. Phys. Chem. B 106 (34) (August 2002) 8493–8498. [24] Michiel Müller, Juleon M. Schins, J. Phys. Chem. B 106 (14) (April 2002) 3715– 3723. [25] Tak W. Kee, Marcus T. Cicerone, Opt. Lett. 29 (23) (2004) 2701–2703. [26] Hideaki Kano, Hiro-O Hamaguchi, Opt. Exp. 13 (4) (February 2005) 1322– 1327. [27] Vasilis Ntziachristos, Nat. Meth. 7 (8) (August 2010) 603–614. [28] Iain Robinson, Michael Andreas Ochsenkühn, Colin J. Campbell, Gerard Giraud, William J. Hossack, Jochen Arlt, Jason Crain, J. Biophot. 3 (3) (August 2010) 138–146.
Please cite this article in press as: A. McVey, J. Crain, Methods (2014), http://dx.doi.org/10.1016/j.ymeth.2014.03.002