1.

Introduction

Intracellular ion signalling microscopy (e.g., Ca ²⁺ flux microscopy) imposes substantial demands on detector technology, which can fundamentally be filtered down to two key requirements—sensitivity and speed. The detector must be sensitive to: 1. detect the weak signal of low dye concentrations; 2. cope with the lower photon fluxes afforded by shorter exposure times (complementing fast frame rates); 3. detect the weaker photon fluxes afforded by reduced excitation powers (reducing photobleaching of dyes and photodamage to tissues, thereby lengthening experimental lifetimes); and 4. overcome the significant readout noise detection limit of high-speed readout rate. High frame rates are required to facilitate the recording of fast calcium flux processes, in accordance with the major temporal resolution requirements of intracellular ion signalling studies. Ca ²⁺ sparks have been recorded in a variety of cell types, including smooth muscle cells.^{1 2 3 4 5} These events represent the spontaneous release of Ca ²⁺ from ryanodine sensitive intracellular stores, and are characterized by their rapid rise time (20 to 95 ms), decay (27 to 60 ms), and limited spatial spread (<2 μm²).² Currently available imaging systems can record spark activity in whole cells, but few are capable of delivering frame rates in excess of 30 fps, particularly while maintaining good resolution in terms of signal to noise (S/N) and spatial definition, which clearly limits the ability to resolve temporal events accurately. The ideal imaging system would allow researchers to record calcium release events from whole cells at frame rates that would allow a more accurate determination of spark kinetics across the entire cell. Furthermore, a particularly sensitive detector is necessary not only to deliver resolvable signals at short exposure times, but also to yield low noise kinetic plots from smaller, more localized intracellular regions of interest.

We have previously reported⁶ use of the world’s first back-illuminated electron multiplying charge-coupled device (EMCCD) (BV-EMCCD, BV denoting optimization for visible and near IR) from Andor Technology in providing optimal signal-to-noise (S/N) performance at significantly reduced input signal levels. The EMCCD is an advanced CCD camera design, offering unsurpassed sensitivity performance, and has been shown to yield markedly improved S/N under ultra-low-light conditions at high speed operation, facilitating the demanding fast speed low-light microscopies such as single molecule detection, intracellular ion signalling microscopy, and 4-D microscopy. The detector makes use of a new CCD architecture^{7 8 9 10 11 12} that unites the sensitivity of an intensified CCD (ICCD), with the inherent advantages of a CCD. The EMCCD camera used in the previous preliminary study⁶ contained a 128×128 pixel sensor format, currently capable of frame rates in excess of 500 fps at 10-MHz pixel readout, and was used to illustrate the combined effects of both back-illuminated sensor technology for maximum photon conversion efficiency, and EMCCD technology for eliminating the camera detection limit.

The specific effects of these technologies is to optimize the two fundamental parameters of sensitivity, detection limit, and quantum efficiency, as recapped next.⁶

Therefore, the BV-EMCCD, a device that combines both electron multiplying CCD technology and back-thinned CCD technology, can be expected to offer both single photon detection sensitivity (by virtue of having eliminated the readout noise) and the ultimate in photon conversion efficiency (therefore improved shot noise). A back-illuminated EMCCD device can theoretically offer improvements of a factor of ×2 (or greater at some wavelengths) in signal/shot noise ratio over a front-illuminated EMCCD device. Indeed such improvements had been previously experimentally verified.⁶

This work describes some ultra-low-light experiments carried out with the iXon DV887 (BV), a back-illuminated EMCCD containing a 512×512 frame-transfer chip reading out at 10 MHz. At this readout speed, the camera can be operated at >30 fps in full frame readout, and much faster still with subarray selection or binning, as is shown. The camera has been incorporated into an ultra-low-light Nipkow spinning disk confocal microscopy arrangement for rapid calcium flux measurements on cells isolated from rabbit urethra, guinea pig bladder, and vas deferens. The results presented are instrumental in verifying three principal benefits of this technical innovation: 1. the significant influence on fundamental S/N considerations in extremely low-light cellular fluorescent microscopy scenarios; 2. the temporal/spatial resolution ability required to identify and record transient Ca ²⁺ spark sites and waves; and 3. the ability of the system to record these events over long durations. In determining the improvements in sensitivity and resolution, the BV-EMCCD was directly compared (using the same cells) to the technology format that before the advent of the EMCCD camera had been considered the standard for low-light detection at fast readout speed—a GenIII+ICCD. The temporal ability was demonstrated through recording of rapid Ca ²⁺ spark and wave events within the smooth muscle cells, over extended kinetic series. Sensitivity was such that kinetic profiles can be plotted of these localized events using very small region of interest (ROI) averaging, while maintaining sufficient S/N for accurate measurement.

2.

Experimental Description

3.

Results and Discussion

To perform a direct performance comparison between the BV-EMCCD camera and the Gen III+ICCD, a series of Fluo-4 loaded rabbit urethra smooth muscle cells were imaged, taking a short kinetic series (10 frames) with one camera, then performing the same analysis with the other camera directly after with the same cell still in place. The order in which the images were recorded was alternated for each subsequent cell, to rule out photobleaching effects on the differing S/N levels. In general, largely due to the rapid confocal nature of the technique and the short 33-ms exposure times employed (2×2 binning of the ICCD was necessary to achieve this exposure), the fluorescent images represent extremely low photon fluxes. All cells imaged were within the same field of cells and subjected to identical dye loading times. However, it was obvious that some cells were inherently weaker than others. This is due to slightly different loading efficiencies, focal fine position, and also, the laser power was purposefully varied slightly to produce different degrees of signal intensities. The important point was that each cell was imaged like-for-like, varying only the camera.

Figure 1(a) shows one of the cells recorded with each camera type, both before and after applying a smoothing algorithm. It is clear that both signal contrast and resolution are markedly better in the case of the images recorded with the BV-EMCCD. The effect of smoothing is to enhance the contrast further between fluorescent signal and background, at the expense of some resolution. However, even after smoothing of each image, the same improvements are noticeable in the BV-EMCCD images compared to the corresponding ICCD image. Figure 1(b) shows a representative series of comparative cell images from this experiment, each having been smoothed to provide clearer visualization, particularly of the ultra-weak images. It is clear that signal intensity and image contrast were better throughout the series for the images recorded with the EMCCD, a direct function of a higher S/N ratio afforded by the higher QE of the EMCCD detector (both types of detector have overcome the read noise detection limit). This pattern has been observed previously for the iXon DV860, 128×128 sensor, versus the same Gen III+ICCD, albeit across a series of different cells within the same field.⁶ The degree of improvement is particularly apparent for cell 3, for which the photon flux was so low that it could barely be detected at all with the ICCD, but is certainly identifiable with the EMCCD. The superior signal of the EMCCD is to be expected, given the markedly restricted QE of ICCDs compared to the unrestrained QE of the BV-EMCCD. Indeed, at the 516-nm λ _max of Fluo-4 dye, the BV-CCD should have >×2 QE than a GenIII+photocathode. This should result in >×1.4 increase in signal-to-shot noise ratio.

Figure 1

(a) Comparative raw and smoothed low-light fluorescent images recorded from a live smooth muscle cell using the BV-EMCCD and the Gen III+ICCD. (b) Further smoothed images recorded of Fluo-4 loaded smooth muscle cells using both the BV-EMCCD and the Gen III+ICCD; 33-ms exposure time per frame; high EM or ICCD gain setting throughout.

Furthermore, the resolution from the BV-EMCCD is markedly better, and one can more easily identify areas of intracellular dye compartmentalization, for example. This resolution improvement arises to a large extent from the different electron amplification technology used in the EMCCD, which virtually eliminates the pixel cross talk that is inherent in ICCD measurements (yielding a form of instrumental blurring),¹² but also from the slightly larger area of the 2×2 binned pixel required to run at a faster readout speed (and shorter exposure time). In comparison, the EMCCD produces images of stark resolution—postacquisition smoothing can be applied artificially through software if required, but this is the researcher’s choice. It is interesting also to note that, in terms of sensitivity, a greater number of photons should hit the slightly larger pixel area of the 2×2 binned ICCD than that of the nonbinned BV-EMCCD, yet the latter camera clearly maintained a clear S/N advantage (without the need to sacrifice resolution for speed).

It is important to note that all of the images shown here are of extremely weak signals, and would be lost entirely within the read noise floor if it were not for the respective signal amplification technologies.

To provide a better depiction of the temporal capabilities of this latest 10-MHz BV-EMCCD camera, a series of cells were chosen from which fast Ca ²⁺ sparking events could be imaged in extended kinetic series, either with full frame or subarea (the latter can be tailored to achieve both faster speeds and to match the dimensions of the muscle cells). The events either occurred naturally within the cell or were induced through controlled injection of caffeine to stimulate Ca ²⁺ release. Generally, it is highly desirable to record as many images as possible within the duration of a sparking event or wave, such that a clear spatial and temporal profile of the intracellular calcium activity can be generated. In fact, one approach that is currently used to get maximum temporal resolution is to employ a line scanning technique.⁵ To obtain such high temporal resolution (500 frames/sec), a significant sacrifice in spatial resolution is required (i.e., the scanned region is only 1 pixel diam).

Figure 2(a) shows a series of selected images from a 1100-frame kinetic series recorded of Fluo-4 loaded guinea pig bladder smooth muscle cell with the BV-EMCCD, operated with 256×256 pixel subarray at 59 frames/sec. The montage of consecutive images represent a Ca ²⁺ spark event, which can be seen to rapidly appear close to the cell membrane and spread into the cell. A smaller 100×100 pixel subarray was then placed on top of the sparking region of this cell, such that a frame rate of 113 frames/sec was obtainable from a more targeted field of view. This function is readily achievable from within the Andor imaging software, requiring the customizable subarray box simply to be dragged on top of the interesting region. Figure 2(b) shows a montage of consecutive images derived from one of the sparking events at this faster frame rate, from which the dynamics of the event occurring within this field of view can be more readily discerned.

Figure 2

(a) Consecutive BV-EMCCD fluorescence images from a 1500-frame kinetic series at 59 frames/sec, 256×256 subarray, showing the initiation and spread of a single Ca ²⁺ spark. (b) Consecutive images from the same cell at 113 frames/sec, 100×100 subarray. The cell section is shown against the black background, from which no fluorescence is observed.

Kinetic plots, shown in Fig. 3, were generated from a ROI function within the Andor software, from a 2.7-μm ROI over the spark site shown in Figs. 2(a) and 2(b), therefore showing comparative kinetics at 59 and 113 frames/sec. In all cases, the Ca ²⁺ spark events are clearly recognizable above the basal fluorescence level. Figures 3(b) and 3(d) show one of the spark profiles from each series expanded out to give a clearer depiction of the number of data points present within a given event, each data point derived from a single image capture. It is clear from this figure that differences in activation and decay kinetics of the spark are clearly identifiable.² The corresponding 113 frames/sec kinetic plot shown in Fig. 3(d) clearly shows more data points per event. To highlight this comparison further, Fig. 3(e) shows expanded kinetic profiles, plotted as a function of frame number, at 59 frames/sec (from Fig. 2) and 113 frames/sec. The spark events are clearly more temporally defined at the faster speeds, yet maintain sufficient S/N despite the shorter exposure times employed. This ability to go to higher temporal resolution while maintaining the sensitivity to deal with the lower signal levels associated with shorter experiments is highly desirable for recording such Ca ²⁺ spark events.

Figure 3

Kinetic plots derived from ROI averages of the 1100-frame kinetic series of Fig. 2: (a) 59 frames/sec; (b) 59 frames/sec expanded spark of the asterix event in (a); (c) 113 frames/sec; (d) 113 frames/sec expanded spark (the baseline slope is due to an offset artifact that has since been rectified); and (e) 59 and 133 frames/sec expanded and overlaid.

The ability of this system to record calcium release events at high temporal resolution is further demonstrated in Fig. 4. In this experiment, the effects of applying caffeine (10 mM) were examined on a quiescent smooth muscle cell. Figure 4(a) shows a montage of images selected from a 1100-frame kinetic series, at 60 frames/sec, of a Fluo-4 loaded bladder smooth muscle cell, during which caffeine was added to induce a Ca ²⁺ release from ryanodine sensitive stores. As the fourth panel of Fig. 4(a) demonstrates, it was possible to observe the initial site of Ca ²⁺ release, which then spread across the entire cell (subsequent images). Figure 4(b) shows kinetic plots recorded from two regions of interest, one at the point of Ca ²⁺ release and one close to the extremity of the cell [bottom of the cell as viewed in Fig. 4(a)]. By using such an analysis, one can build up a clear profile of the response, decay, and magnitude of the wave event, and can also measure the time taken to spread across the cell. Without a camera of sufficient sensitivity as well as temporal and spatial resolution, rapidly spreading events such as this would be very difficult to characterize in terms of both confidently identifying the localized region of weak initial activity, and recording this initiation with sufficient data points.

Figure 4

(a) Selected fluorescence images from a 1100-frame kinetic series at 60 frames/sec, showing the progression of a caffeine-induced Ca ²⁺ transient event in guinea pig bladder myocyte. (b) Kinetic plots derived from offset ROIs within the cell.

Recently Sergeant et al.;¹ demonstrated the presence of spontaneous transient inward currents in specialized pacemaking cells in the rabbit urethra, and later suggested that these may be due to formation of a spontaneous Ca ²⁺ wave in these cells.⁴ To test this idea and to demonstrate that continuous recording of these events was possible over a long duration, we examined activity in an isolated interstitial cell from the rabbit urethra loaded with Fluo-4-AM. Figure 5(a) shows a selection of images taken from a 1500 kinetic series of images acquired at 30 frames/sec. It is clear from these images that Ca ²⁺ oscillations arise from one region of the cell and are propagated throughout the cell. To analyze the spread of this wave throughout the cell, two 50×50-pixel ROIs were placed at opposite ends of the cell. The resultant plots, shown in Fig. 5(b), clearly show an overlaid series of offset wave events, the peaks arising from the bottom ROI appearing consistently ahead of those from the top ROI, from which the transverse time of the wave is measurable. It is interesting to note that the delay between the peaks varies slightly from wave to wave and that, in contrast to the guinea pig smooth muscle, no distinct spark-like events could be detected. To illustrate the ability to measure these events at faster speeds, kinetic series were recorded at 60 and 120 frames/sec also. Figure 5(c) shows a section of the ROI kinetic plot obtained at 30 frames/sec, compared to a wave recorded at 120 frames/sec. While it is clear that the progressively shorter exposure times employed to afford greater temporal resolution resulted in a lower S/N ratio, the camera’s ultra-sensitive performance was such that even at 120 frames/sec, the Ca ²⁺ wave profile could still be clearly discerned. It is important to note that the subarray used to give 120 frames/sec was sufficient to image the entire cell at this speed without need for pixel binning.

Figure 5

(a) Selected fluorescence images from a 1500-frame kinetic series at 30 frames/sec, showing the progression of a Ca ²⁺ wave as it progresses though a rabbit urethral cell. (b) Kinetic plots derived from offset ROIs within the cell. (c) Comparison of kinetic plots generated from 30 and 120 frames/sec kinetic series of the same cell (the 120 frame/sec images used the same field of view, varying only the exposure time).

During analysis, ROIs are employed, selectable from within the Andor imaging software, from which average values may be calculated from the pixels contained within. If small ROIs are used, this feature may allow researchers to examine events that are spatially smaller than sparks, i.e., Ca ²⁺ quarks.¹⁴ Figure 6 shows a series of kinetic profiles generated from the studied spark site of the 59 frames/sec kinetic series of images examined in Fig. 2. The profiles were generated from three different sizes of ROI placed at the center of the sparking region, which when the ×100 objective and ×1.5 c-mount lens are accounted for, correspond to ROIs of ∼2.7×2.7 μm², 1.3×1.3 μm², and 0.7×0.7 μm². It is clear that the S/N ratio decreases as the ROI area decreases, as expected. However, camera sensitivity is such that the sparks can be confidently identified, even from the smallest ROI area.

Figure 6

Kinetic plots generated from ROIs of different sizes, 2.7, 1.3, and 0.7 μm²; the ROIs placed within the cell spark site analyzed in Figs. 2 and 3.

4.

Conclusions

The significant boost in ultra-sensitive imaging performance offered by the new back-illuminated EMCCD is demonstrated here in the context of measured improvements to a particularly demanding ultra-low-light microscopy technique. Improvements are in the form of sensitivity, resolution, and speed parameters that are to a large extent interdependent. This enhanced sensitivity offered by this camera complements its ability to deal with the lower photon fluxes associated with the shorter exposure times inherent to fast frame rate operation. Furthermore, through intelligent choice of subarray selection, very fast frame rates (>100 frames sec) are achievable without sacrifices to pixel resolution. The higher sensitivity will also prove entirely useful in enabling use of both lower fluorophore concentrations, lower excitation powers, higher optical magnifications and smaller ROI analysis, conditions employed for the most part in the current study. The BV-EMCCD will undoubtedly be used in the optimization of a number of challenging low-light microscopy approaches (and indeed low-light macroscopic approaches also), including single molecule microscopy^{15 16} and 4- or 5-D live cell microscopy.