We describe a custom one-photon (confocal) and two-photon all-digital (photon counting) laser scanning microscope. The confocal component uses two avalanche photodiodes (APDs) as the fluorescence detector to achieve high sensitivity and to overcome the limited photon counting rate of a single APD (∼5 MHz). The confocal component is approximately nine times more efficient than our commercial confocal microscope (fluorophore fluo 4). Switching from one-photon to two-photon excitation mode (Ti:sapphire laser) is accomplished by moving a single mirror beneath the objective lens. The pulse from the Ti:sapphire laser is 109 fs in duration at the specimen plane, and average power is ∼5 mW. Two-photon excited fluorescence is detected by a fast photomultiplier tube. With a ×63 1.4 NA oil-immersion objective, the resolution of the confocal system is 0.25 μm laterally and 0.52 μm axially. For the two-photon system, the corresponding values are 0.28 and 0.82 μm. The system is advantageous when excitation intensity must be limited, when fluorescence is low, or when thick, scattering specimens are being studied (with two-photon excitation).
- cardiac muscle
- excitation-contraction coupling
optical sectioning techniques, such as confocal microscopy (14) and two-photon microscopy (6), provide the means to localize molecules in living cells with high spatial and temporal resolution. With these techniques, the understanding of cell function is being extended to cellular volumes (“microdomains”) on the order of 10− 15 liters (16). Furthermore, events within such microdomains of individual cells may be visualized as part of a much larger scene, such as an entire group of interconnected cells. Through the use of two-photon microscopy, such scenes may be located deep (e.g., ∼100 μm) within a living, reasonably intact tissue (3). The physiologist studying live tissue in this situation is usually faced, however, with measuring low levels of light. It may be necessary to use low illumination intensity to reduce photodamage to the specimen (12). There may be only a few fluorescent molecules, or the concentration of fluorescent dye must be kept low to reduce perturbation of the cell (2). In a confocal system, spatial resolution can be improved (up to a point) by rejecting fluorescence, further reducing the signal. In thick specimens, light scattering may reduce the efficiency of fluorescence excitation and detection (3). Because we are interested in conducting studies of the type described above, we have constructed a combined confocal and two-photon microscope in which the unique features are related to efficient detection of low levels of fluorescence. This is achieved through simple, efficient optical design, the use of novel detectors (for such microscopes), and photon counting, which offers improved accuracy and signal-to-noise ratio (S/N) at low light levels (compared with analog methods). The use of photon counting also allows the system to be “all digital,” with the consequent advantages of accuracy and ease of signal processing. Thus the microscope is a “photometric” device, in that light is quantified accurately, for the purposes of quantification of fluorescent molecules. We also present a simple technique for measuring duration of the ultrashort pulse of infrared (IR) light used for two-photon excitation at the specimen plane. This technique allowed us to measure the group velocity dispersion (8, 15) introduced by the glass in the microscope.
METHODS AND MATERIALS
Figure 1 presents a schematic diagram of the optical components and the fluorescence detectors. The major optical component is a Nikon Diaphot 300 microscope (Nikon, Garden City, NY) with two Nikon objective lenses (L4), an oil-immersion ×63 lens (NA 1.4) and a ×60 “water objective” lens (NA 1.2). The microscope and associated optics and lasers are all mounted on a large vibration isolation table (4 × 6 ft) (Technical Manufacturing, Peabody, MA) with the “clean top” surface provided with mounting holes. Most optical components (holders and mirrors) were purchased from New Focus (Sunnyvale, CA). The design of the confocal laser “scan side” of the instrument is conventional and has been described previously (13). Differences to the system, also described previously, include the method of beam scanning, beam intensity modulation, fluorescence detection, and digital image construction. On the “confocal side,” the beam from an argon ion laser is incident on a single mirror (SM), located in the exit pupil of the scan lens (L3). This mirror is rotated by galvanometers around two orthogonal axes (x, horizontal or line scanning; y, vertical or frame scanning). The line-scanning galvanometer (model 6800, Cambridge Technology, Watertown, MA) is mounted in a custom, inertially balanced holder mounted on the shaft of the large (frame scanning) galvanometer (model 6900, Cambridge Technology). This system offers several advantages over dual-mirror systems in that the design is simplified, light losses are minimized, and alignment is simplified. The intensity of the beam is modulated using an electrooptic device (i.e., a Pockels cell, model 350-50, Conoptics Danbury, CT). This device provides “blanking” of the beam during horizontal and vertical “flyback” of the scan mirror. In addition, it provides precise, programmed control of beam intensity, which is distinctly advantageous compared with the use of conventional neutral-density optical filters mounted in wheels. The L3 scan lens (Fig. 1) is a Nikon ocular (×10, wide-field CFW) adapted to fit into the side video port of the microscope. Motorized focusing (z-axis) is accomplished with a Nikon Remote focus accessory (stepper motor) controlled through a serial computer interface (RS 232). Fluorescence emission from the specimen is “descanned” by the scan mirror, and, after passing back through appropriate dichroic beam splitters (EFSP1 and DCLP1) and filters (BPF2), is directed by a mirror (M1) into an aperture. This aperture is relatively large (∼1.0-mm diameter) because the Airy disc (i.e., the image of the point source of emitted light) is imaged “at infinity.” Light passing through the aperture is essentially parallel and is focused by a lens (L1) into the detection system. This system consists of a 50%–50% beam splitter (BS1), which directs light equally into two photon-counting modules, each consisting of an integrated avalanche photodiode (APD) and amplifier/discriminator (SPCM-AQ-121; EG&G Optoelectronics). The L1 lens focuses the light to a region smaller than the diameter of the active area of the photodiode (∼180 μm).
To this system, we have added a multiphoton excitation system, consisting of a titanium (Ti):sapphire laser (Tsunami; Spectra-Physics Lasers, Mountain View, CA) pumped with a 5-W solid-state (NdYVO4) green laser (Millenia; Spectra-Physics Lasers). After passing through a pair of high-index prisms for precompensation of group velocity dispersion (GVD), the IR beam (typically 800 nm) from the Ti:sapphire laser is combined with that from the argon ion laser (typically 488 nm) at a short-pass edge filter (EFSP1; centered at 720 nm; Omega Optical, Brattleboro, VT). The two beams are aligned before reaching the scan mirror, through the adjustment of the MP2 mirror and the EFSP1. A portion (5%) of the IR beam is picked off by a beam splitter (BS2) and directed into a monochromator (Oriel, Stamford, CT) and video camera for observation of the spectral output of the Ti:sapphire laser and assurance of mode-locked operation. For two-photon imaging, the stage of the microscope is covered to exclude ambient light. A dichroic mirror (DCLP2; 605 nm) is positioned to reflect emitted light into the rear epifluorescence port of the microscope, in which is positioned a filter (EFSP2; center wavelength 600 nm) for excluding reflected red light and a lens (L2) to assure that the emitted light (which is “nondescanned”) impinges on the active area of a photon-counting photomultiplier tube (PMT; model 9124, Electron Tubes; amplifier discriminator model 1182, EG&G).
For safe operation, the entire table is enclosed around its perimeter to prevent stray laser light from leaving the table. In addition, interference filters have been placed in the eyepiece tube of the microscope to prevent the excitation laser light (at 488 and 800 nm) from reaching the eyepieces. An IR viewer is used to align the IR beam and to check for any IR light emerging from the eyepieces.
The essential electronic components of the microscope are 1) a microcomputer equipped with a variable-scan digital video capture board, a digital input/output (I/O) board, and a digital-to-analog (D/A) output board; 2) a custom pulse counter that emulates a variable-scan digital video camera; and 3) driver circuitry for the mirror galvanometers. Control of most functions is through a “virtual instrument” consisting of a LabView program (LabView and IMAQ, National Instruments) operating the digital video capture board, D/A board, and I/O board in the microcomputer (operating system Windows NT 4.0). The digital I/O board generates horizontal and vertical synchronization pulses and a pixel clock signal, all of which are used to gate the counting circuitry and to synchronize capture of the digital data by the variable-scan video capture board. Voltage waveforms, representing the desired scan mirror excursions, are generated by the D/A board. Actual driving voltage is derived from these signals by driver circuitry provided with the galvanometers (Cambridge Technology). Pulses (transistor-to-transistor logic) from the detectors (APDs and PMT) are counted by a custom counter (gated by the pixel clock) and sent as 8-bit digital data to the variable-scan digital video capture board. The custom gated counter has a maximum count rate of 50 MHz per channel. It has two separate channels that each produce a number (0–255) representing the counts detected during a pixel. The numbers from the two channels can be sent separately to the digital video capture board, or they can be added to each other before being sent. This provides the capability for dual-channel imaging (confocal) or the use of two APDs to improve S/N (see below). Images are simultaneously displayed and stored in real time (either in the computer RAM or on a hard disk). In addition, a D/A circuit within the custom counter produces analog voltage signals that are viewed on an oscilloscope and that could be used with an analog variable-scan video capture board.
System calibration and characteristics: low light detection.
APDs are advantageous for measuring low levels of light, primarily because they have higher quantum efficiency (QE) than PMTs at the wavelengths of interest. They are also ideal for confocal microscopy, where the emitted light is descanned and can be focused to a small point, corresponding to the small active area of an APD. Furthermore, APDs with particularly small active areas with consequent low noise can be utilized. The QE of the APDs we used increases from 0.45 to 0.60 over the wavelengths from 500 to 600 nm (range of the emission spectrum of Ca-fluo 4), whereas the QE of an appropriate PMT falls from ∼0.15 to ∼0.05 over that same range. To estimate the improved detection of fluo 4 fluorescence that might be provided by an APD compared with a PMT, we integrated the numerical product of the fluorescence emission spectrum of Ca-fluo 4 and the QE of our APDs and PMT. The APD produced 4.2 times as much signal from fluo 4 as did the PMT.
Photon counting, which can be done with either APDs or the PMT, is particularly advantageous when levels of light are low (1). Photon counting eliminates the variability arising from the distribution of current pulse amplitudes and provides an absolutely accurate quantification of “black level,” which arises only from “dark counts.” The latter is particularly advantageous when using a low background fluorescence (F0) to construct ratioed fluorescence signals, such as those used to study cardiac Ca2+ sparks (4, 10). The only significant disadvantage is that maximum count rates in photon-counting systems are limited by system “dead times,” or that time during which the system is insensitive to another incoming photon. Thus it is necessary to correct the recorded count rates for system dead time when count rates are high. The correction factor required for a particular APD is provided by the manufacturer. However, the use of a correction factor increases uncertainty (reduces S/N) of the measurement, and therefore correction factors should be kept as low as possible. The photon-counting modules we used had a dead time of ∼40 × 10− 9 s, and the corrected count rate as a function of measured count rate is shown in Fig.2 A (filled symbols). It can be seen that the corrected count rates become appreciably different (i.e., >10%) for count rates >2.0 MHz. This problem can be partly overcome through the use of two APDs such that the count rate at each is 50% of what it would be otherwise (Fig. 2 A, open symbols). This reduces the correction factor that must be applied. Thus the correction factor for a two-APD detector at a recorded count rate of 10.0 MHz is 1.37, whereas that for a single APD detector is 2.0. The use of correction factors is perfectly valid, provided that a good estimate of the measured count rate is obtained. This requires that pixel durations be long enough that adequate numbers of counts are accumulated. For example, in a single-APD system, count rates of 5.0 MHz can be measured accurately with pixel durations of 10.0 μs, but not 1.0 μs. With 10.0-μs pixels, each pixel will contain, on average, 50 counts. The variance of a sample of such pixels (representing, for example, the peaks of Ca2+ sparks) will be 7.07, corresponding to 0.707 MHz. With pixel durations of 1.0 μs, however, the variance will be 2.24, corresponding to 2.24 MHz. Thus much larger and more variable correction factors will be used for pixels 1.0 μs in duration compared with pixels of 10.0 μs in duration, despite the fact that the “true” count rate (i.e., 5.0 MHz) is exactly the same. As shown in Fig. 2 A, this problem is reduced significantly by the use of two APDs.
To obtain the intensity-response function for the confocal microscope, a sample of fluorescein (10.0 μM) was illuminated with laser light over an ∼20-fold range of intensities through the use of the Pockels cell. As described above, images could be obtained separately from each APD. Distributions of pixel values (count rates) obtained from each APD are shown in Fig. 2 B (solid line) as well as their sum, as produced by digital addition of the two images in the custom counter. In the case shown, the mean count rates from each APD were 1.53 and 1.35 MHz, reflecting the fact that the beam splitter (BS1, Fig. 1) did not split the light perfectly into two equal beams. The mean of the summed image was 2.85 MHz, as expected. At low light levels (<5 counts/pixel, or typically 0.5 × 106counts/s) the distributions of count rates was well fit by the Poisson distribution (not shown). At higher light levels, the distributions approached normal (as shown). S/N (defined as mean/variance) were 1.0, as expected. The intensity-response function for the two-APD system is shown in Fig. 2 C, in which the summed average count rate is plotted as a function of input light intensity. Recorded count rates deviated significantly from linearity at count rates >5.0 MHz, indicating undercounting due to system dead time. Because the two APDs had the same dead time (according to the manufacturer) and an insignificant difference in recorded count rates (12%), a correction factor for one-half the summed recorded count rate was used on the images, resulting in a linear intensity-response function (Fig. 2 C, open triangles).
As discussed above, the use of a correction factor reduces S/N. In the case of uncorrected images, S/N increased from 1.0 to 1.1 over the range shown (Fig. 2 C), whereas that of corrected images decreased from 1.0 to 0.6 over the same range. The increase in S/N of uncorrected images is actually an artifact due to the filtering effect of system dead time. On the other hand, S/N decreased in the corrected images as mean count rate increased, because of large correction factors that are applied to pixels that happen to have count rates greater than the mean. S/N can be improved by using longer pixel durations, which results in more counts per pixel and a more reliable (i.e., less variable) estimate of the true count rate and, hence, of the true correction factor.
In the two-photon fluorescence detection system, the dead time of the amplifier discriminator (model 1181, EG&G PARC) is 20 ns, much greater than the duration [7.5 ns, full width at half-maximum (FWHM)] of the current pulses through the PMT (model 9124, Electron Tubes). Thus the dead time of the two-photon detection system is similar to that of the two-APD system. Although large QE would be desirable in a two-photon imaging system, the emitted fluorescence is not descanned, and therefore a wide-area detector, such as a PMT, must be used. Because count rates (even at the peak of Ca2+sparks) recorded with the two-photon system were generally <2.0 × 106 counts/s, it was not necessary to apply correction factors to these data.
Efficient two-photon excitation requires ultrashort pulses of light at the specimen plane (5). It is known, however, that the duration of the pulse of light increases as it travels through glass, such as microscope objectives (8). The amount of such broadening can be calculated, provided that the amount of glass and its refractive index are known. This is generally not known for commercial microscopes and objective lenses, however, so a method of determining “pulse width” at the specimen plane was devised. The ability to determine pulse width accurately at the specimen plane also aids in compensating GVD through the use of external “precompensation optics.” In our system, precompensation for GVD is provided by a pair of high-index prisms (SF- 10) mounted in micromanipulators, according to methods described earlier (7). The polarization of the output beam is rotated 90° through the use of a periscope (MP1) before precompensation. Pulse width was determined from the autocorrelation function (Fig.2 D) measured at the specimen plane by using an aspheric lens (New Focus ×60, NA 0.6) to collimate the light emerging from the objective lens (Nikon ×60 oil, NA 1.4) before directing it into an autocorrelator (SP 409-08, Spectra Physics). The aspheric lens added a negligible amount of glass to the system. We found that this arrangement was easier to use than the methods described previously (8,15), in which the GVD compensation is inferred from the maximization of two-photon fluorescence emission. With 86.5 cm of air between the prisms, and with 1.8 cm of prism glass in the path, the pulse width at the specimen plane was 109 × 10− 15s. The total compensation was thus 4,075 fs2, similar to that reported by others (8, 15). The spectral output was approximately Gaussian, centered at 800 nm, with an FWHM of 11 nm, as determined with the monochromator and video camera system (Fig. 1).
Point-spread functions (PSF; Fig. 3) were measured with the use of subresolution fluorescent beads (0.1 μm in diameter) immobilized on a glass coverslip that forms the bottom of our standard physiological recording chamber. A three-dimensional data set was acquired by first focusing on a plane immediately above the bead and then scanning with the automatic stepper set in 0.1-μm steps. With a ×60 1.4-NA oil-immersion objective, the lateral FWHM of the confocal PSF was 0.25 μm and the axial FWHM was 0.52 μm. For the two-photon system, with the same objective lens, the corresponding values were 0.28 and 0.82 μm. For these measurements, pixel dimensions were 0.05 μm laterally and 0.1 μm vertically. Three-dimensional images of the PSF were constructed, and vertical sections (X–Z planes) are shown in Fig. 3.
We compared the optical efficiency of our custom confocal with that of a Bio-Rad MRC 600 confocal microscope (Bio-Rad Laboratories, Melville, NY). Efficiency will be defined here as the ratio of the number of photons exciting a sample (per pixel) to the number detected (per pixel). The sample was a uniform solution of fluorescein (1.0 μM) in a physiological recording chamber, and the chamber, containing exactly the same sample, was used with both microscopes. The MRC 600 was put into photon-counting mode and adjusted as described in the supplied manual, with the exception that the gain was set to the maximum. Excitation power, at the specimen plane, was measured with a power meter (LaserCheck; Coherent, Auburn, CA) having a resolution of 0.01 μW. The MRC 600 was set to normal scan mode [768 × 512 pixels, 0.275 μm/pixel, 1.5 μs/pixel (estimated)]. Power at the specimen plane was 5.5 μW (or μJ/s), corresponding to 2.16 × 107 photons/pixel (there are 2.46 × 1018 photons per joule at a wavelength of 488 nm). The mean number of detected photons per pixel was 2.8, giving an actual efficiency of 1.30 × 10− 7. For the custom confocal microscope, a power of 0.97 μW was used to produce 26.7 photons/pixel (mean) in a smaller image (256 × 256 pixels, 0.10 μm/pixel, 10.0 μs/pixel). The efficiency of this system was thus 1.12 × 10− 6. Thus the efficiency of the custom confocal is ∼ 8.6 times that of the MRC 600. Because the spatial resolution of our system is somewhat better than that of the MRC-600 (13), it seems reasonable that the relative efficiency would be even greater if the systems were compared at exactly the same spatial resolution.
Confocal images of cardiac sparks.
We are interested in recording Ca2+ sparks in cardiac cells. For this application, the confocal microscope is the clear choice, because it provides superior spatial resolution to the two-photon microscope. However, no measurements of Ca2+sparks with the two-APD system have been published previously. The preparation of rat cardiac ventricular cells for this purpose was according to the methods described previously (10). Ca2+sparks were recorded in the “line-scan” mode, in which we scanned a single line 25.6 μm in length in 3.0 ms. Of this time, 0.44 ms are used for horizontal flyback. Thus the pixel “dimensions” are 10.0 μs and 0.1 μm. As discussed previously, the relatively long pixel duration limits the distance that can be scanned but increases S/N. The small pixel size of 0.1 μm is necessary to satisfy the Nyquist criterion, given the dimensions of the confocal PSF. Through the use of the Pockels cell, the intensity of emitted fluorescence in the cardiac cell at rest (F0) can be adjusted within the range of 1.0–2.0 MHz. Because the peak fluorescence pseudoratios (ΔF/F0) of cardiac Ca2+ sparks seldom exceed 4 or 5, this permits the measurement of Ca2+ sparks with a minimal correction factor and adequate S/N. An example of a Ca2+ spark with a peak ΔF/F0 of ∼2 is shown in Fig. 4 A. The distribution of amplitudes of all Ca2+ sparks recorded in this experiment is shown in Fig. 4 C (open bars). According to theory, Ca2+ spark amplitude histograms are limited at their lower end by noise (9). This limitation was characterized by the distribution of peak ΔF/F0 (Fig. 4 C, hatched bars) obtained from images in which no Ca2+ spark occurred. Clearly, Ca2+ sparks having ΔF/F0 <0.5 cannot be measured with this system because of noise. Ca2+ sparks recorded with two-photon excitation (Fig. 4 B) can appear similar to those recorded with one-photon excitation. The amplitude histogram (Fig. 4 D), however, is skewed to larger Ca2+ sparks because greater noise (at the level of illumination used) limited the ability to measure small Ca2+ sparks.
Two-photon imaging is expected to be advantageous, particularly in thick, scattering specimens (3), but provides no advantage in spatial resolution in thin specimens. This is illustrated in Fig.5, in which the same pollen grain is imaged with the two systems (Fig. 5, A and B). These pollen grains are characterized by spikes or spines on their surface that provide a convenient illustration of spatial resolution. The optical section provided by two-photon excitation (Fig. 5 B) is clearly not as thin as that provided by the one-photon confocal system. The lower axial resolution of the two-photon system (Fig. 2) is apparent as increased haze around the sharp spines. We have recently published the first confocal images of Ca2+ within individual vascular smooth muscle cells of the walls of living pressurized resistance arteries (11). This is a thick specimen for which two-photon imaging should be advantageous. Accordingly, we stained a rat mesenteric artery with an amine-reactive fluorescein dye [5-(and-6)-carboxy-2′,7′-dichlorofluorescein diacetate, succinimidyl ester, Molecular Probes, Eugene, OR], and the same region of living artery wall was imaged with the two systems (Fig. 5,C and D). In this case, the two-photon image is superior, revealing more clearly the individual vascular smooth muscle cells and providing a better optical section of endothelial cells that are folded into the arterial wall (visible as the “trough” running down the center of the image).
We present a simple, efficient, and cost-effective combined confocal and two-photon laser scanning microscope. It provides spatial resolution that is comparable to or better than that of typical commercial microscopes, while using approximately nine times less light (in confocal mode). This provides substantial benefits for physiological studies. For example, voltage-clamp studies of Ca2+ sparks in cardiac cells typically involve repetitive scanning of the same region of the cell thousands of times during a series of voltage-clamp pulses (10). The high efficiency of our custom confocal microscope means that such experimental protocols can be much more prolonged, enabling the collection of much more data, because there is less risk of photodamage to the cell and less photobleaching of the Ca2+ indicator dye. The detection of fluorescent Ca2+ indicator signals in intact arteries under physiological conditions (11) is also greatly facilitated by the efficient signal collection, because the rate of loss of fluorescent indicator dyes from smooth muscle cells is relatively high at mammalian temperatures compared with room temperature.
The use of photon counting in this microscope provides high precision of light measurement and obviates some sources of noise (1). Most importantly, black levels, which are very important in fluorescence ratios (e.g., ΔF/F0), are very low in this instrument and are not subject to analog offsets in amplifiers. Whereas the instrument excels at measuring low levels of fluorescence, its dynamic range is limited by the dead times of available photon-counting systems.
Although not demonstrated here, a major advantage of this instrument over typical commercial instruments is its adaptability for physiological experiments in which various stimuli (e.g., voltage-clamp pulses) may have to be given in exact temporal relationship to image acquisition. Because all functions of the instrument are readily controlled and accessible, the timing of physiological stimuli and data acquisition is facilitated.
Finally, by incorporating both one- and two-photon fluorescence-excitation systems, versatility is obtained for studying both thin and thick specimens. The addition of a Ti:sapphire laser that is tunable to the system also makes possible the two-photon excitation of fluorophores over a wide range of wavelengths (5).
We acknowledge the invaluable assistance of Dr. Victor A. Miriel with the preparation of isolated arteries used in this study.
Address for reprint requests and other correspondence: W. G. Wier, Dept. of Physiology, Univ. of Maryland School of Medicine, 655 West Baltimore St., Baltimore, MD 21201 (E-mail:).
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.
- Copyright © 2000 the American Physiological Society