1. Introduction

Three-dimensional (3D) fluorescence microscopy is an important method for biological and medical research. Multiphoton excitation greatly facilitates 3D fluorescence microscopy due to the nonlinear absorption process that confines the fluorescence emission to the focal region. This enables 3D scanning microscopy without a confocal pinhole or descanning. The confinement of excitation to the focal region further minimizes photobleaching and photodamage from volumes outside the focal region [1].

In general, the speed of a microscope is limited by the number of detected photons per pixel integration time to produce a measurable signal relative to noise. The number of detected photons is limited by the specimen fluorophore concentration, fluorophore saturation and specimen photodamage. For most biological samples, the excitation peak intensity is restricted to about 200 GW/cm², which is reached by a tightly focused 7–15 mW femtosecond laser beam. At higher excitation power, photodamage often occurs and the confinement of imaging volume degrades. As this limits the maximum fluorescent signal, the speed of multiphoton single-spot imaging is ultimately limited by the concentration of fluorophores in the specimen. This limitation can be circumvented by parallelization of the imaging process using multiple excitation foci, whose peak intensities all lay below the saturation or damage threshold. The technique is known as Multifocal Multiphoton Microscopy (MMM) [2–4]. The intrinsic 3D resolution provided by the nonlinear excitation process negates the need for a pinhole array as in Nipkow-disk based confocal microscope [5, 6].

The MMM has been shown to be particularly suitable for live cell investigations since it combines the advantages of standard multiphoton microscopy such as optical sectioning and suppression of out-of-focus phototoxity with high recording speeds. Its utility has been proven in applications such as studying receptor sorting in COPI vesicles based on fluorescence resonance energy transfer FRET MMM microscopy [7] and the 3D distribution of acidic organelles inside live neuroendocrine cells transfected with green fluorescent protein GFP [8]. Further, it is relevant to imaging tissue structures without physiological motion artifacts in living animals or patients [9, 10], and to extracting statistically significant measurements from a large sample volume [11–13].

Thus, the MMM provides real time 3D resolved images of a biological specimen which can be observed using a digital camera or by eye. In this instrument we take this approach even further by also parallelizing the detection using a CCD camera, which divides the imaging sensor into 16 independent segments. The multi-segment camera significantly reduces camera read noise, increasing the sensitivity and the maximum possible frame rate of the multiphoton images.

2. Instrument design and characterization

The technical design of the MMM consists of a densely packed squared microlens array (Adaptive Optics, Cambridge, MA); each microlens transmitting a collimated beam at a slightly different angle, illuminating the back aperture of the objective lens. The foci are scanned rapidly across the image plane by an orthogonally arranged resonant mirror-system to ensure full coverage of the focal plane [Fig. 1(a)]. A Ti-sapphire laser (Coherent Mira 900) generates a maximum output power of 1.6 W at 780 nm, with ~150 fs pulses at a repetition rate of 76 MHz, and illuminates a 6×6 microlens array. The number of foci was chosen so each focus on average carries a fractional power of approximately 10 mW which lays below the damaging threshold of the sample. This value includes the power loss in the optical beam path that is over 75%. A resonant scanner with a fast (10240 Hz) and slow (640 Hz) axis (Electro-Optical Products Corp, Glendale, NY) scans the foci [Fig. 1(b)] across a total region of 64 µm×64 µm.

The fluorescence is collected by a CCD camera and is synchronized with the scanner. The CCD has a maximum frame rate of 1428 frames per second but it is operated at 640 frames per second in this instrument. This system can be slowed down to acquire weaker fluorescent signals by reducing the frame rate. The camera is a cooled back-thinned CCD with a quantum efficiency of over 80% in the visible range and a read out noise of 11±1 electrons at 1428 fps, 8±1 electrons at 714 fps, and 6+/-1 electrons at 357 fps. For the actual speed of 640 frames per second, the read noise is 8 electrons. The pixels are square and 21 µm on a side. It was custom developed for fluorescence microscopy, consisting of 16 segments, each containing 16×64 pixels, composing a final 128×128 image size. Each of the camera segments has its own readout amplifier for which the threshold can be adjusted individually.

While the system design is primarily optimized for power throughput, it remains important to quantify the image resolution. In Fig. 1(d), the intensity profile of a 0.3 µm green fluorescent latex particle (Bangs Laboratories) is shown. The bead was immobilized in agar. The plot shows the axial resolution of the system is approximately 1.0 µm as characterized by the full-width-at-half-maximum of the point spread function. The effective point source is imaged with a 60×1.2 NA objective lens (Nikon, Plan Apo 60×WI: N.A. 1.2) at 780 nm excitation wavelength. The data were acquired by imaging 800 xy image planes laying 100 nm apart and profiling the intensity along a single pixel in z. The signal level values are taken directly from the brightest pixel in the images without fitting or averaging.

 

Fig. 1. (a). Principle of the Multifoci Multiphoton Microscope (MMM). (b) Sketch of the MMM setup. For illustration purposes, in (a) and (b) an array of only 3×3 beams is shown. (c) Still image of the foci in the focal plane. (d) Intensity profile along the axial axis. Abbreviations: Ti:Sa: Ti-sapphire Laser, MLA: Microlens Array, L: Lens, RSM: Resonant Scan Mirror, DC: Dichroic Mirror, TPB: Two-Photon Block Filter, CCD: Segmented Charged Coupled Device, Inv. M.: Inverted Microscope.

Download Full Size | PDF

The axial resolution is primarily limited by the illumination geometry. Theoretical resolution values of an ideal two-photon microscope under these conditions are 0.3 µm and 0.7 µm in the radial and axial directions respectively. To maximize excitation light throughput, we illuminated the back aperture with collimated light in a rectangular shape under-filling the back aperture of the objective lens, resulting in a diminished axial resolution in comparison to its theoretical limit. The resolution loss due to residual interference effects of neighboring beamlets causes only slight reductions in resolution, as the foci are sufficiently far apart [14]. Since the system signal-to-noise level is limited by the read noise of the CCD camera, we intentionally degrade the radial resolution to maximize the photon collection in each CCD pixel by choosing the system magnification such that the CCD pixel size is slightly larger than the radial point spread function in the image plane. Therefore the complete image of a focus resides in one single pixel.

3. Instrument evaluation by imaging calcium transients in myocytes

Among the many biological events that happen in millisecond time scales, the occurrence of spontaneous and induced calcium release has functional implications with regard to tissue electrophysiology and pathophysiology. As an interesting model to evaluate the performance of this high speed microscope, we studied myocyte contractions driven by calcium transients. Calcium waves are known to propagate at typical speeds of approximately 105 µm/sec. driving contraction waves.

 

Fig. 2. The spontaneous contraction of cardiac myocytes labeled with fluo3. The data are extracted from a 0.625 sec. movie sequence, consisting of a series of 400 images, each acquired 1.56 ms apart form each other. The number in the images indicates their place in the sequence each being 78 ms (50 images) apart. The images were smoothed in time with a Gaussian filter with four frames standard deviation. The dark current has been subtracted and the image was corrected using the flat field image.

Download Full Size | PDF

In order to resolve the calcium concentration with micron level resolution, images must be acquired with temporal resolution better than 100 Hz. We measured spontaneous calcium wave generation and propagation in fluo 3 loaded myocytes with 640 Hz, generating a 0.625 second movie sequence, consisting of 400 single images 1.56 ms apart from each other. The cardiac myocytes are freshly dissociated adult rat cells loaded with ~2 micromolar of the acetoxymethyl ester (AM) of fluo 3 in solution for ~30 minutes at room temperature.

A selection of acquired images with temporal separation of 78 ms is shown in Fig. 2. Fluo 3 fluorescence increases in the presence of calcium. The calcium wave propagation can be directly measured by the fluorescent intensity emitted. In Fig. 3(a), the calcium propagation is shown at different cross sections of the myocyte along with the corresponding white light image in which the indicated area has reached its half peak. The intensity profiles in Fig. 3 are taken from the same dataset shown in Fig. 2. The wave propagates 40 µm in 382 ms which indicates an average velocity of 105 µm per second which is in accordance with the literature [15].

 

Fig. 3. (a). White light image of cardio myocyte shown in Fig. 2(b) Fluorescent intensity over time measured at areas 1, 2 and 3, cross sectioning the cell at 20 µm distances (a). The data indicate a wave propagation of 105 µm per second along the cell. The intensity plots are taken from the fluorescent image dataset shown in Fig. 2.

Download Full Size | PDF

4. Discussion

One of the major speed limitations of the former MMM designs is the relatively low frame rate of the CCD camera used. The major noise sources of any CCD camera are its dark noise and read noise. The dark noise of the CCD camera is a function of a number of device parameters such as temperature, pixel size, and integration time. For high speed imaging, dark noise contribution is typically small since integration time is short. In contrast, read noise represents a more serious limitation. The read noise originates from the electronic noise of the readout amplifier of the CCD chip. The amplifier noise is a function of readout frequency. Therefore, increasing CCD frame rate invariably results in an increase in read noise and a decrease in sensitivity. This reduction in detector sensitivity occurs with a concomitant decrease signal level as high speed implies shorter pixel residence time of a scanning microscope. Therefore, the speed of a MMM system is ultimately limited by the signal to noise level of the image. A path that we have taken in this MMM design is to improve the sensitivity of the CCD detector at high speed. This camera’s unique performance is mainly based on the segmentation of its CCD chip.

Most CCD cameras have a single segment and all the pixels on the chip are read out through a single output amplifier stage. In contrast, in this custom designed CCD chip, the device is segmented in to 16 regions; each region has its own readout amplifier. By parallelizing the readout of the CCD chip, one can improve the overall readout frame rate of the device without incurring higher read noise. It is informative to compare the noise characteristics of this segmented camera with a non-segmented counterpart. Since our CCD has 16 segments, the read noise of a single segment system would be at least four times higher assuming that the readout amplifier follows at best a square-root dependency on frequency [16]. While there is significant advantage in using a segmented CCD for high speed, high sensitivity imaging, there are also disadvantages. Specifically, since each segment has a different readout amplifier, each segment can have different signal gain and offset levels resulting in a non-uniform intensity image even with uniform illumination [Fig. 4(a)]. While these gain and offset levels can be tuned, they are a function of chip temperature and do vary overtime. This problem can be overcome in the future design of this CCD chip that may include an automated gain and offset adjustment circuitries.

In addition to read noise, our CCD camera also has a smaller but not negligible dark noise. Operating at a frame rate of 640 frames/sec, a typical dark image of the CCD camera is shown in Fig. 4(a) along with a histogram of the signal distribution in one typical segment [Fig. 4(b)] and histogram of the mean value of the dark signal [Fig. 4(c)] of all segments. At this speed, we measure a dark noise of about 3 electrons and a read noise of 8 electrons.

 

Fig. 4. (a). Dark image of the 128×128 pix CCD camera at 640 frames/sec. The picture includes the dark and read noise of the camera. (b) Histogram of the signal of one typical segment (segment #12, framed in (a)). (c) Histogram of the mean signal values of all segments. The average dark signal of all segments is 8.7 electrons.

Download Full Size | PDF

It is instructive to consider the signal to noise of the fluorescent bead (Fig. 1) and myocyte (Fig. 2 and Fig. 3) images acquired using this segmented CCD. In addition to the CCD dark and read noises, the total noise of the image is further composed of the noise generated by the sample background and the shot nose of the signal itself. The noise generated by the sample is affected by the fluorescence of the sample and its embedding. In the case of the imaging of the fluorescent bead, shown in Fig. 1, the background fluorescence is close to zero. Due to sample preparation of the cardiac myocyte, the background fluorescence is significant and contributes to almost 1/3 of the total noise. An overview of the signal and noise consideration in the imaging of these two samples using this instrument is given in Table 1. We have further calculated the signal to noise ratio for a hypothetical single segment CCD (assuming read noise has quadratic dependence on readout frequency) for comparison. Taking into account the quantum efficiency of this segmented CCD, we see that about 12–15 photons from the sample must be incident on this device to achieve unity in signal to noise level. For the estimated conventional CCD with only a single readout channel, this number will increase by a factor of 4.

Table 1. Comparison of the presented 16 segment camera and a comparable single segment camera for two different image applications. The numbers represent the detected electrons of the system characteristics at an imaging speed of 640 frames per second.

View Table

5. Conclusion

We introduced a multiphoton microscope which is capable of rapidly generating three dimensional data in the sub micron scale over an area of 64 µm×64 µm using multi-foci excitation and a high-speed, high-sensitivity CCD camera.

With the presented system, the noise equivalent signal per pixel at a frame rate of 640 Hz is approximately 10–15 photons for typical biological sample. The maximum frame rate for a typical conventional single-segment CCD with the same set of amplifiers to achieve the same sensitivity would be 160 Hz. Therefore our system improves the imaging speed by four times. We demonstrate the use of this instrument to quantify 3D resolved calcium wave propagation in cardiac myocytes and establish a propagation velocity of 105 µm per second. This instrument should have broad applicability in a variety of biomedical problems where high speed imaging with 3D resolution is critical. To our knowledge, the data presented in this report are the fastest images of biological specimen as yet acquired with multiphoton microscopy.

A number of improvements may further improve the performance of this system. First, due to the Gaussian shape of the laser beam, the intensity distribution of the foci is inhomogeneous and falls off at the image corners. A homogeneous intensity distribution can be achieved by replacing the microlens array with a diffraction optical element (DOE) which allows power variations between different optical paths within 1% [17]. Since the variation in illumination intensity of the foci is relatively static, it can be easily corrected by calibration, if necessary. Although software correction works well to improve the uniformity of the images, optical correction has the advantage of generating images with uniform signal to noise, which allows high frame rates. Second, the imaging speed of the microscope could be even further increased if the total available laser power would be greater. As discussed, this system was limited to 6×6 foci as the maximum laser power of 1.6 W enables the illumination of 36 foci with a fractional average power of 10 mW. The only way to increase the imaging speed is to further parallelize the illumination. A new line model of Ti-sapphire lasers with a maximum output power to 5W will enable the illumination of 100 foci at the same power level.