3.1.

Imaging Retinal SHG from Kidney Cells and Cultured Neurons

Our first trials were carried out on a strain of human embryonic kidney cells (HEK 293), which expresses the mammalian rhodopsin protein in their membranes. In the course of these experiments, it was discovered that the reconstitution of external all-trans retinal was necessary for the production of SHG, and moreover, there was no apparent difference between the rhodopsin expressing cells and the control cells, in terms of the SHG signal. Both strains have shown very strong SHG signals after the addition of all-trans retinal (ATR) into the extracellular medium (see next). We therefore concluded that all-trans retinal by itself is sufficient to produce SHG in these cells and presumably in other types of cells. The retinal molecules bind to the outer leaflet of the cell membrane by lipophilic attraction and orient themselves even in the absence of rhodopsin.

In Fig. 1 we show simultaneous images of SHG and TPEF of HEK 293 cells and a composite image of the two. Cells were grown on cover glass in culture prepared from a cell line that does not express rhodopsin. Similar images were acquired of the rhodopsin expressing cells, in which the signal strength was comparable (not shown). The main feature is that SHG originates predominantly from the membrane, although some retinal was internalized, as can be seen from the fluorescence signal from within the cells, presumably accumulated in internal membranes of the endoplasmic reticulum (ER) and Golgi apparatus. The multiphoton excited fluorescence—although very weak compared with the SHG (for presentation purposes the TPEF channel is enhanced in the image)—was detectable only after incubation with 50-μM retinal for a period of roughly 20 min. Therefore, the endogenous autofluorescence cannot account for this signal. What is striking is that the SHG signal from the membrane is accompanied only by a very weak fluorescence compared to the cytoplasmic TPEF. The reasons for that are still not clear at this point. The image in Fig. 1 was taken shortly after modifying our two-photon scanning microscope for the purpose of SHG microscopy,⁴¹ and it does not include the lock-in amplifier, later added to the setup. This explains the relatively high average power (∼20 mW at the sample) that was needed to produce an image with such clarity.

Figure 1

Imaging retinal SHG of kidney cells. Composite image of SHG (blue) and TPEF (green) HEK 293 cells (not expressing rhodopsin) stained with all-trans retinal 50 μM in extracellular solution, laser power ∼20 mW at 850 nm, circularly polarized light, and 30-μs pixel dwell time. 20× magnification objective lens NA 0.5 was used with the normal condenser lens NA 0.8. Below the individual SHG (right) and TPEF (left) images are shown. The TPEF image is enhanced using a look-up table adjustment.

Figure 2 shows an example of SHG imaging of rat hippocampus pyramidal neurons stained with ATR. It shows a high spatial resolution both in the plane of the image and in the axial (z) direction. The latter is a result of the nonlinear optical nature of the SHG process. Even though the neurons are grown on a cover glass and their morphology is essentially planar, unlike the situation in the real brain, the cell’s somata are seen at the first image and the dendrites are clearly visible at the second image, which is focused 2-μm closer to the cover glass. Here we used lock-in detection to amplify the SHG signal. High-frequency modulation of 800 kHz using an EO modulator and a short time constant of 10 μs allow noise rejection and high gain while maintaining relatively high data rates. The scanning pixel dwell time was 30 μs, which allows three time constants per pixel. In the raw data of the two images of Fig. 2, the full (12-bit) dynamic range is used and the signal at the soma is even saturated at some points, whereas the background is essentially zero. Note that the laser was polarized in the direction of the y axis of the image and therefore the horizontal membranes of the dendrites and soma appear to have a stronger signal. This is in agreement with the assumption that the retinal molecules are oriented roughly perpendicular to the membrane surface.

Figure 2

Imaging retinal SHG of neurons. SHG images of rat hippocampus pyramidal neurons in dissociated culture, stained with 25-μM all-trans retinal. The two images are separated by 2 μm in the z axis (where z is the vertical distance from the cover glass surface). An Olympus objective lens of 60× magnification NA 0.9, and condenser lens NA 0.8 were used. Laser average power at the sample ∼4 mW at 900 nm, linearly polarized light (in the y axis of the image), 30-μs pixel dwell time, lock-in amplifier used with 800-kHz modulation frequency, and time constant of 10 μs. No Kalman frame averaging and no image processing was done on the images except for some enhancement of the lower pixel values by limiting the maximum value in the look-up table to 1332 (from the original 4095).

3.2.

SHRIMP Signal Reflects Membrane Voltage

To measure the changes in SHG signal as a function of membrane potential, we used the patch clamp technique. In this procedure, a thin glass electrode filled with the right salt solution is introduced into the cell’s cytoplasm, and a giga-ohm seal is formed by the membrane around it. A pyramidal neuron from rat hippocampus, which was grown in dissociated culture on a glass substrate, was patched and stimulated to fire action potentials. After verifying that the cell was healthy and excitable by electrically recording action potentials in current clamp, we switched to voltage clamp, where the cell membrane was held at −70 mV with respect to the ground electrode in the extracellular medium. An image of the patched neuron is shown in Fig. 3(a). The laser was then scanned over the cell at the same line (shown in green) repeatedly, and an x-t image was recorded of the SHG signal while the voltage was switched 89 times between 0 and −70 mV every 40 scan lines (200-ms interval). In Fig. 3(b), we show the temporal mean of the line scan where the two graphs correspond to the two voltages. The two peaks correspond to the points where the laser beam crosses over the left and right membrane (vertical) walls of the neuron, where the curvature of the membrane yields a strong signal. The difference between the signals at the two voltages at the peaks are 18.4 and 18.9 for the right and left membranes, respectively (calculated for five points around the peaks). The significance of these values was tested both in the time and spatial domains. First, two time series corresponding to the two voltages were generated from the value at the membrane (average of three pixels centered at the peak of the x profile) at every time point. We then performed a standard t test for two independent populations, which resulted in the different values for the means and a significance parameter p=1.66⋅10⁻⁴, N=3560. The second significance test was done by taking the two x profiles (for the two voltages) shown in Fig. 3(b) and performing a t test for two paired populations (the values depend on the x position of each pair) on five points centered at the peak. The result of this test was again two significantly different populations with p=4⋅10⁻³, N=5. We note that taking more than five points around the peak actually results in a smaller p value (more significant difference), but the difference between the two mean values for the two populations is slightly reduced and can lead to underestimation. This can be easily understood when one considers that away from the membrane, the signal approaches zero and so does the difference. A similar test was done on the left membrane data with similar results.

Figure 3

Retinal SHG is sensitive to membrane voltage. (a) SHG image of rat hippocampus pyramidal neuron under patch clamp. The green line is the chosen xt line scan going through the soma membrane. Fastest scan speed of 2 μs per pixel was used without lock-in detection, and laser power 5 mW at the sample at 880 nm. (b) x profile averaged over time for the two voltages. The difference profile is also shown in blue. The difference in percentage is also shown at the two peaks (the left and right membrane walls). (c) Percentage change in SHG signal as a function of voltage step. The slope obtained from linear regression through the origin with correlation coefficient R=0.99498 and p<0.0001 significance parameter.

We repeated this procedure for a voltage step of 40 mV (−70 and −30 mV) and 20 mV (−70 and −50 mV), the results from these voltage clamp experiments are summarized in Fig. 3(c), where we plotted the percentage change in SHG for the left- and right-hand membranes. The result of the linear regression gives a slope of 25.4±0.7 increase in SH signal for 100-mV depolarization of the cell’s membrane with correlation coefficient R=0.9949 and p<0.0001 significance parameter. Due to photobleaching and to deterioration in the health of the cell during scanning, the experiments with the voltage step of 40 and 20 mV (which were done after the 70-mV step scan) yielded a less significant result using the t tests than the first experiment. We nonetheless include these results, since they are consistent with the expected linear dependency of SHG on voltage.

3.3.

SHRIMP Toxicity and Phototoxicity

All-trans retinal is a hydrophobic molecule and therefore requires the addition of a nonpolar solvent, which is undesirable but necessary in many other labeling procedures of different fluorescent markers. Although other groups seem to be able to image neurons stained with retinal using ethanol,³⁸ our neurons seem to be healthier with the DMSO as the solvent rather than the ethanol. In our hands, the DMSO became toxic for hippocampal cultured neurons at concentrations greater than 0.05. The retinal/DMSO when added to the cell culture seemed to become toxic when incubated for more than 20 min without wash. The toxicity was evaluated by observing the integrity of the cells’ membrane under a conventional light microscope. However, the cells were well stained and healthy if this time limit was kept during labeling. Rather than toxicity due to the solvent, the greatest challenge in these types of experiments was to perform patch clamp recording on the previously labeled neurons, since we found it difficult to obtain a good electric seal (∼1 GΩ is needed) between the patch pipette and the cell membrane after staining, presumably due to the deposition of retinal and DMSO molecules on the cellular membranes, which create a kind of screening of membrane by adsorbing to the glass of the pipette.

Photodamage was observed in some cases when the laser average power was increased beyond 10 mW at the sample (and using a 0.9-NA objective lens) and the laser was pointed to a particular point at the membrane for duration of more than 100 ms for ∼50 times intermittently. In these cases a visible change—strong bleaching and a distortion of the membrane nearby—could be seen in the neuron’s membrane at the point, judging by the SHG image. We did not observe dramatic changes in cell health when the laser light was kept under 10 mW at the sample, and regular scanning was used. To illustrate this point, the cells in Fig. 2 look perfectly healthy, as can be seen by the integrity of their membrane and their processes. Unhealthy neurons develop beaded dendrites.

3.4.

Reverse Polarity of the SHRIMP Signal

We found it surprising that for retinal the SHG signal actually increases with depolarization of the membrane, whereas in all other studies that use other molecules for SHG measurements of membrane potential, the signal decreases with depolarization. To explain the differences in behavior, one has to consider the permanent molecular dipole of the dye and its direction in the membrane as well as the induced dipole moment change. It is usually assumed in the two-level model that the excited state possesses a larger dipole moment than the ground state.¹⁵ It is known that upon light excitation of ATR, there is a net negative charge moving away from the β-ionone ring toward the aldehyde group^{33 42} (see Fig. 4 for molecular structure of ATR). Langmuir-Blodgett trough experiments²⁷ suggest that the molecular alignment of ATR in the membrane is such that the aldehyde group faces the water interface and the ionone ring is inside the lipid bilayer. It is also known that both the permanent dipole (μ_g) and the induced change in the dipole (Δμ) point toward the ionone ring.^{33 34} We can conclude therefore that the small permanent dipole of ATR is directed toward the cytoplasmic side and is in the same direction as the electric field in the membrane at resting conditions in which the cell is more negative than the extracellular environment. This is in contrast to the amphiphilic styryl dyes such as di-4-ANEPPS,⁹ which have the opposite polarity in the membrane. We therefore propose that the reverse polarity of the dipole moment and subsequently Δμ may explain at least qualitatively why the response to voltage change has the opposite effect on SHG intensity.

Figure 4

Chemical structure of all-trans retinal aldehyde.

It seems reasonable to assume that the mechanism for the SHG response to voltage changes in the membrane-bound retinal is mainly of electro-optic origin through the induced dipole moment change,^{15 18} but we cannot rule out additional realignment contributions.²² However, the increase in SHG with decreasing electric field strength suggests that reorientation is not the main contributor to the response and may even be a weakening factor in the response.

3.5.

SHRIMP Bleaching and Wavelength Dependency

As discussed in the introduction, SHG can be obtained off-resonance and therefore SHG imaging can be made with significantly reduced photobleaching, which can in some cases create a serious problem. For example, to optically detect fast biological events like action potentials that take place on the time scale of milliseconds, one cannot use the laser beam to scan a wide area. There is a tradeoff between field of view and temporal resolution. In the extreme case when the laser beam is parked at one position on the membrane and the laser shutter is opened, we notice very fast photobleaching of the SH signal at a rate of tens of milliseconds with moderate laser power of less than 10 mW at 850 nm (data not shown). We also noticed partial recovery if one allows for some time to pass between openings of the shutter, possibly from 2-D diffusion along the membrane. The process of SHG bleaching in the membrane is not well understood, although some models exist. One such model supports a light-induced flip-flop mechanism²³ of the amphiphilic dye molecules going from one leaflet of the membrane to the other, and thus spoiling the asymmetrical distribution and reducing SHG. The bleaching could also be due to light-induced conformational change from all-trans to cis configuration, which has a smaller Δμ, or simply a result of bond breaking and loss of HRS from these molecules. Since surface SHG depends quadratically on the number density of molecules that generate HRS,^{21 25} SHG bleaching would occur even at a faster rate than bleaching TPEF, which depends linearly on the number of fluorescent molecules.

We wanted to explore the possibility of reducing SHG photobleaching by operating at a wavelength away from resonance and preferably at a longer wavelength. Since ATR has a single photon resonance around 380 nm, we decided to test the bleaching rate at 900 nm with the Ti-sapphire laser and at 1060 nm with the Nd:glass laser. Indeed, the results shown in Fig. 5 convince us that at this range, longer wavelength reduces photobleaching.

Figure 5

Photobleaching of retinal SHG. (a) Time averaged (20 frames) SHG image of a neuron stained with ATR. Laser used was an ultrafast Nd:glass laser from High-Q Lasers, operating at 1060 nm. Pixel clock of 10 μs (2 s per frame). (b) Threshold image showing the selected membrane pixels, which are used for averaging. (c) Photobleaching curves for two excitation wavelengths 900 nm (Ti-sapphire laser) and 1060 nm (Nd:glass laser). 20 frames were taken sequentially as an xyt movie. Each point in the graph represents the average pixel value of that frame calculated for the membrane pixels [shown in Fig. 4(b)].

Figure 5(a) shows the resultant time-average (20 frames) SHG image of a neuron stained with ATR. In Fig. 5(c) we see the photobleaching curves for two excitation wavelengths: 900 and 1060 nm. 20 frames were taken sequentially as an xyt movie (2 s per frame). The filters in front of the PMT were changed accordingly to 450- and 530-nm center wavelengths with a 20-nm band. Each point in the graph represents the average pixel value for one frame calculated at the membrane pixels [shown in Fig. 5(b) after applying a threshold to exclude background pixels]. It is clear from the graph that the bleaching rate is faster for the shorter wavelength. The double exponential fit reveals that there are at least two characteristic decay time scales: a fast and a slow decay. Interestingly, the 1060-nm data shows an initial fast decay t₁=3.5 frames (even faster than the 900-nm data, though with a small amplitude) and then a very slow decay t₂=169 frames. The transmission of our system (including the 60×IR2 Olympus objective lens) was measured in this wavelength range, and the laser power marked in the figure is after this correction has been made. Although the average power at 1060 nm in our experiments is two times greater, and pulse width is actually shorter (100 fs) than at 900 nm (150 fs for the MIRA laser), which means higher peak intensities, and accepting that the other parameters are quite similar (repetition rate, objective back aperture filling, and beam quality), photobleaching is less severe at this longer wavelength, image quality is not degraded, and longer experiments duration involving intensive imaging can be performed.

3.6.

SHG versus TPEF

Optical techniques have had a great impact on neuroscience in the past and improvements in microscopy methods are changing our view of the CNS. Studying electrical signals in neurons using light only can take this field into a whole new level of sophistication and understanding. The quest for optical recording from neurons has therefore been ongoing for the last three decades and is nicely illustrated in some of the publications of Webb, who also pioneered the use of multiphoton microscopy in biological research⁴² and who recently also contributed to SHG microscopy,^{43 44 45} and to whom this issue of the Journal of Biomedical Optics is a tribute. We present a technique that relies on SHG, which like TPEF is a nonlinear optical phenomenon that requires very large photon fluxes. SHG has obvious advantages for membrane imaging. It is an exquisitely sensitive technique to measure interfaces^{46 47} and thus appears ideal for biological samples. We do not expect it, however, to replace TPEF, since it does not have the sensitivities that fluorescence dyes possess for chemical environments. In addition, fluorescence-based techniques have the obvious advantage of imaging green fluorescent protein (GFP) and its variants. Rather, the two techniques seem to complement each other in their functionalities and are easily realized in a laser scanning microscope, as is shown in this work and in numerous previous studies. Many of the recent applications of SHG microscopy make use of an intrinsic signal from arrangements of certain proteins such as collagen that possess the required symmetry for SHG, thus revealing specific tissue structure.^{11 44 45} Moreover, given that rhodopsin is a HRS protein that can produce SHG, we propose the possibility of imaging the retina using SHG. Although the geometry has to be changed to image the live eye, contrast images could reveal with high resolution the density distribution of rhodopsin proteins and of oriented aggregates of retinal in the retina, and this could possibly lead to medical applications in diagnosing defects in the retina, as well as to the study of photochemical and photobiological mechanisms of vision.