|Home | About | Journals | Submit | Contact Us | Français|
Probing biological structures and functions deep inside live organisms with light is highly desirable. Among the current optical imaging modalities, multiphoton fluorescence microscopy exhibits the best contrast for imaging scattering samples by employing a spatially confined nonlinear excitation. However, as the incident laser power drops exponentially with imaging depth into the sample due to the scattering loss, the out-of-focus background eventually overwhelms the in-focus signal, which defines a fundamental imaging-depth limit. Herein we significantly improve the image contrast for deep scattering samples by harnessing reversibly switchable fluorescent proteins (RSFPs) which can be cycled between bright and dark states upon light illumination. Two distinct techniques, multiphoton deactivation and imaging (MPDI) and multiphoton activation and imaging (MPAI), are demonstrated on tissue phantoms labeled with Dronpa protein. Such a focal switch approach can generate pseudo background-free images. Conceptually different from wave-based approaches that try to reduce light scattering in turbid samples, our work represents a molecule-based strategy that focused on imaging probes.
Multiphoton fluorescence microscopy is now the most popular technique for imaging scattering biological samples with sub-cellular resolution [1,2]. Due to the nonlinear intensity dependence of the absorption process, multiphoton excited fluorescence is mostly generated from the laser focus where the probability of having two (or more) simultaneous incident photons is the highest. This confined excitation enables efficient capture of scattered fluorescence photons by a wide-field detector without a confocal pinhole. Such intrinsic optical sectioning dramatically improves the detection sensitivity and hence imaging depth into scattering samples [3,4]. Consequently, multiphoton microscopy has become an indispensable tool for biomedicine [5,6].
However, multiphoton microscopy will eventually lose contrast when imaging deep into scattering samples [7–11]. As demonstrated in Fig. 1(A) , the two-photon fluorescence image of fluorescent beads embedded in a turbid 3D sample quickly fades away with the increasing depth. But images can be acquired much deeper if the laser power is increased to compensate for the scattering loss as shown in Fig. 1(B). Nevertheless, image contrast deteriorates with depth as the fluorescence from the out-of-focus beads grows and eventually dominates the signal from the focal region. At this fundamental depth limit, the inherent optical sectioning breaks down. Obviously, this loss of image contrast cannot be overcome by further increasing the excitation efficiency which would enhance signal and background equally.
Extensive efforts have been invested in improving the image contrast of multiphoton microscopy by employing a number of strategies, such as adaptive optics to pre-compensate for the scattering loss [12,13], imaging with longer wavelengths , optical phase conjugation , ex vivo chemical cleaning reagent , differential aberration imaging  and focal modulation . It is worth noting that many of these methods focus on tailoring the incident laser beams to reduce the scattering effect (including the tissue induced aberration and refractive index mismatch) of the turbid sample on the light waves. Relative little work has been carried out to search for special imaging probes themselves.
Herein we explore the utility of an emerging class of reversibly switchable fluorescent proteins (RSFPs) for high-contrast multiphoton imaging. GFP and its variants are widely used as genetically encoded reporters in bio-imaging [19–21]. Unlike conventional FPs, RSFPs can switch between a fluorescent and a non-fluorescent state upon light illumination in a reversible manner. Such unique photochromism has been employed for a variety of applications including studying intracellular protein trafficking , sub-diffraction resolution fluorescence microscopy [23–28], removing auto-fluorescence via optical lock-in , photochromic FRET [30,31], and most recently as a genetically encoded micro-viscosity reporter . By harnessing the switching ability of RSFPs, our strategy preferentially turns on or off the RSFPs at the focal plane while keeping the protein in the out-of-focus background in the opposite state. We demonstrate two distinct techniques, multiphoton deactivation and imaging (MPDI) and multiphoton activation and imaging (MPAI), on tissue phantoms made of live bacterial and mammalian cells labeled with Dronpa-3.
Dronpa and its closely-related mutants Dronpa-2 and Dronpa-3 are among the first discovered and best studied RSFPs. In this report we chose to use Dronpa-3 for demonstration, as it exhibits a higher quantum yield of bright-to-dark switching when compared to the original Dronpa . Like most RSFPs, Dronpa-3 is initially in its bright state. As shown in Fig. 2(A) , irradiating Dronpa-3 at 488nm will generate green fluorescence emission. However, prolonged 488nm illumination converts Dronpa-3 to a dark state. Then irradiation at 405nm efficiently recovers the original bright-state. The structural basis for its photochromism is believed to be the cis-trans isomerization of the internal chromophore, accompanied by a change in the protonation state .
We then evaluate two-photon induced photoswitching kinetics of Dronpa-3. Figure 2(B) shows the two-photon fluorescence images of Dronpa-3 expressing E. coli cells excited by a pulsed laser at 920nm which is around the two-photon absorption peak of GFP . Repeated scanning and imaging over the same cells displays a graduate signal attenuation due to the bright-to-dark switching. As expected, a higher 920nm laser power leads to a notably faster kinetics of switching-off. In both tests, brief irradiation at 405nm subsequently recovers the original bright-state intensity. Hence, the bright-to-dark switching-off kinetics is indeed dependent on the two-photon excitation intensity at 920nm. Similarly, in Fig. 2(C), we confirmed that the dark-to-bright switching-on kinetics is dependent on the two-photon excitation intensity at 800nm. Therefore, RSFPs offer on-off molecular states that can be modulated by external two-photon light illumination.
Before elaborating on MPDI and MPAI, we need to first discuss two-photon imaging of regular fluorophores. Using the criterion that the in-focus signal (S) and the out-of-focus background (B) are equal [7–11], the fundamental depth limit can be defined as
where Vin, is the focal volume, Vout, is the total sample volume along the beam path but excluding the focal volume, r, is the distance from the optical axis, z, is the axial distance from the surface, C is the local fluorophore concentration, and I is the laser intensity, and τ is the pixel dwell time. We assume there is no fluorophore saturation or bleaching, and identical fluorescence collection efficiency between the signal and the background at the wide-field detector.
Equation (1) could help us to quantify how the laser intensity distributes within the scattering sample when the above fundamental depth limit is reached. The fluorophores are normally distributed throughout the 3D volume of the sample. Thus, the number of out-of-focus fluorophores is almost always orders-of-magnitude larger than that of the in-focus ones. Therefore, if the generated fluorescence signal is comparable between the focal volume and the out-of-focus volume, the laser intensity at the focus will have to be much more intense than the out-of-focus counterpart, despite the scattering loss. When coupled with photoswitching transitions, the high focal laser intensity could offer additional discrimination mechanisms against background, which is exactly being employed in MPDI and MPAI.
We now propose to carry out MPDI as shown in Fig. 3(A) : make a differential image before and after switching off the bright Dronpa-3 preferentially at the focus volume. The pre-deactivation image of Dronpa-3 labeled sample, quickly acquired at 920 nm, would be the same as the image obtained with regular GFP. In contrast, the post-deactivation image will depend on the switching kinetics whose rate should be dependent on the local intensity of 920nm (Fig. 2(B)). As analyzed earlier for Eq. (1), the 920nm laser intensity at the focus is much more intense than the out-of-focus counterpart at the imaging-depth limit. In a non-depletion region, this would lead to a more efficient deactivation at the focus after the switching scanning process at 920nm. Consequently, as illustrated by Fig. 3(A), Dronpa-3 will be primarily turned off at the focus in the post-deactivation image. The resulting difference image between pre- and post- deactivation would effectively cancel the out-of-focus background. Therefore, we expect a much improved S/B ratio for MPDI with Dronpa-3.
We then demonstrate MPDI on tissue phantoms made of live bacterial or mammalian cells. Dronpa-3 expressing E. coli cells are embedded in 3D low melting-point agarose gel (2%). The resulting sample is highly scattering due to the packed E. coli cells in 3D. As demonstrated in Fig. 4(A) , the out-of-focus background is substantial for a quick pre-switching scan at a depth of 250 µm. The corresponding near-unity S/B indicates that this depth is close to the fundamental imaging-depth limit for regular fluorophores. After performing one relatively slow (or multiple fast) deactivation scanning at 920nm, the post-deactivation image indeed became dimmer. The difference image offers a satisfactory contrast with much reduced out-of-focus background. We note that prolonged deactivation led to poor MPDI contrast due to depletion within the whole sample. To demonstrate MPDI in mammalian cells, we further constructed a “two-layer” composite sample in which a dense layer (120 µm) of Dronpa-3 expressing E. coli cells serve as the “scattering” background and H2B-Dronpa-3 expressed HEK 293T cells on a coverslip are the targeted objects. As shown in Fig. 4(B), the MPDI image is effectively free from out-of-focus background generated from the E. coli cells in the “scattering” layer.
Inspired by the spirit of MPDI, we now propose a different but related technique: multiphoton activation and imaging (MPAI) with RSFPs. This is possible because of the reversible manner of the cis-trans isomerization. As shown in Fig. 3(B), we will implement three sequential steps: first, switching Dronpa-3 in the entire sample volume of interest to the dark state with a prolonged 488 nm illumination; second, preferentially activating Dronpa-3 from the focus to the bright states by scanning the focal plane with a pulsed laser at 800 nm with a proper pixel dwell time of ξ; third, taking a regular two-photon image at 920 nm. Similar to the deactivation in MPDI, the two-photon activation process will occur predominately at the focal plane in a non-saturation region, and therefore, the subsequent two-photon imaging of the photoactivated population should exhibit a superior image contrast.
MPAI is demonstrated on similar tissue phantom samples as in MPDI. As shown in Fig. 5(A) , the out-of-focus background is overwhelming for the regular two-photon images of the Dronpa-3 (bright state) expressing E. coli cells at a depth of 140µm. A brief 488nm laser illumination efficiently switched off Dronpa-3 completely in the sample volume. After that, two-photon activation by scanning with a 800nm pulsed laser and subsequent two-photon imaging with a 920nm pulsed laser resulted in a remarkable image contrast of cells at the focal plane. We note that over-activation (with too strong 800nm laser or too slow scanning) led to poor MPAI contrast. The result on H2B-Dronpa transfected HEK 293T cells in the “two-layer” sample, shown in Fig. 5(B), also proves a much lower background and improved contrast for chromosome features inside the cell nucleus. Thus, we have demonstrated MAPI in providing superior contrast and hence deeper penetration of two-photon microscopy for scattering samples.
In Fig. 1, fluorescent beads were purchased from Invitrogen (F13080: blue/green 1.0 µm beads), mixed with 5% Intralipid (Sigma I141) which serves as the scattering species, and then embedded in 2% agarose gel in three dimension. In live bacteria imaging experiments presented in Figs. 2, ,4,4, and and5,5, Dronpa-3 expressing E. coli cells were prepared as previous reported , and then harvested by centrifugation, resuspended and embedded inside an imaging spacer using 2% agarose gel with or without non-fluorescent polystyrene beads (0.9 µm diameter in average from Sigma-Aldrich CLB9). In the HEK 293T mammalian cell imaging experiments presented in Figs. 4 and and5,5, “two-layer” composite samples consisting of a background layer and a target layer were prepared as follows. An imaging spacer with a 120 µm thickness (Sigma GBL654004) was first pasted onto a piece of micro coverglass (VWR 48393-150). Inside the spacer, 10 µL of Dronpa-3 expressing E. coli cells in 2% agarose gel together with non-fluorescent beads serving as the scattering background was added. This background material was then sealed inside the spacer by a 35-mm glass bottom petri dish on the top. The imaging target objects, HEK 293T cells, were cultured on the petri dish and transiently transected with H2B-Dronpa-3 plasmids which were constructed as reported previously .
All the two-photon fluorescence images were taken on a Leica TCS SP5 MP inverted microscope equipped with a Mai Tai HP laser (690-1040 nm tunable emission) and a HCX PL APO CS 20X DRY microscope objective (N.A. = 0.70). A non-descanned epi photomultiplier tube (PMT) detector placed directly behind the objective was used for fluorescence collection in combination with a 680 nm Short Pass emission filter. Images were processed with LAS AF software unless otherwise stated.
Images of fluorescent beads (Fig. 1) were taken under 840 nm with a pixel dwell time of 20 µs. Images of the Dronpa-3 expressing E. coli samples (Figs. 4(A) and 5(A)) were taken under 920 nm (99 mW) with a pixel dwell time of 40 µs. For Fig. 4(A), a slow deactivation scanning in x-y at 920 nm (99 mW) with a pixel dwell time 160 μs was performed before taking the post-deactivation image. For Fig. 5(A), an activation scanning in x-y at 800 nm (2.9 mW) with a pixel dwell time of 80 μs were performed in prior to the subsequent imaging at 920 nm. The pre-deactivation images of H2B-Dronpa-3 transfected HEK 293T samples (Fig. 4(B)) were taken under 920 nm (99 mW) with a pixel dwell time of 40 µs. The post-deactivation image, after 40 frames of two-photon deactivation scanning at 920nm (99 mW), was taken under 920 nm (99 mW) with a pixel dwell time of 40 µs. Images of H2B-Dronpa transfected HEK293T samples (Fig. 5(B)) were taken under 920 nm with a pixel dwell time of 40 µs. The two-photon activation was performed with 4 scanning frames of 800 nm (40 mW) with a pixel dwell time of 40 µs. Note that due to the strong sample scattering of the prepared tissue phantoms, the actual laser power reached at the actual focal plane is lower than the measured total power transmitted through the microscope objective. All images were acquired with 512 by 512 pixels.
Both MPDI and MPAI of RSFPs are based on the insight that the laser intensity at the focus has to be much more intense than the out-of-focus counterpart despite the scattering loss, when the depth limit is reached for regular fluorophores as defined in Eq. (1). When coupled to the photoswitching kinetics of RSFPs, such laser intensity disparity translates to an additional mechanism for discriminating signal over background in the non-depletion or saturation region. Hence, although MPDI and MPAI seem to adopt opposite approaches, the S/B are comparable between MPDI and MPAI in theory. However, there is a profound difference in terms of noise. The background rejection is performed numerically in MPDI, after detection, rather than physically, before detection as in MPAI. Although the average position-dependent backgrounds can be effectively canceled out for every scanning pixel in MPDI, the shot noises from the two backgrounds would add in quadrature, which sets a higher limit for the distinguishable in-focus signal size for MPDI.
Radically different from the existing approaches that mostly focus on reducing scattering loss by tailoring the incident light waves, the current work represents a molecule-based strategy that harnesses the special photophysics of imaging probes themselves. In terms of practical performance, the recently engineered new generation of RSFPs [31,36] with high photon output, faster photoswitching rate and different colors has the prospect to further improve MPDI and MPAI.
Finally, it is constructive to make connections to a few other related techniques. Cyclic sequential transition of Dronpa-3 has been harnessed to increase the spatial resolution in one-photon imaging . In addition, the switching of RSFPs was used to improve the image contrast in the optical lock-in detection . However, the background there is mainly from the auto-fluorescence which cannot be modulated by light. Moreover, the recently proposed stimulated emission reduced fluorescence (SERF) microscopy can also improve the image contrast for thick scattering samples . From a spectroscopy perspective, MPDI of RSFPs is very similar to the attenuation effect of stimulated emission in SERF microscopy.
We thank Lu Wei, Zhixing Chen, Louis Brus, Rafael Yuste, Darcy Peterka and Virginia Cornish for helpful discussions. W.M. acknowledges the startup funds from Columbia University, and grant support from Kavli Institute for Brain Science.