CellSpecks: A Software for Automated Detection and Analysis of Calcium Channels in Live Cells

Abstract

To couple the fidelity of patch-clamp recording with a more high-throughput screening capability, we pioneered a, to our knowledge, novel approach to single-channel recording that we named "optical patch clamp." By using highly sensitive fluorescent Ca²⁺ indicator dyes in conjunction with total internal fluorescence microscopy techniques, we monitor Ca²⁺ flux through individual Ca²⁺-permeable channels. This approach provides information about channel gating analogous to patch-clamp recording at a time resolution of ∼2 ms with the additional advantage of being massively parallel, providing simultaneous and independent recording from thousands of channels in the native environment. However, manual analysis of the data generated by this technique presents severe challenges because a video recording can include many thousands of frames. To overcome this bottleneck, we developed an image processing and analysis framework called CellSpecks capable of detecting and fully analyzing the kinetics of ion channels within a video sequence. By using randomly generated synthetic data, we tested the ability of CellSpecks to rapidly and efficiently detect and analyze the activity of thousands of ion channels, including openings for a few milliseconds. Here, we report the use of CellSpecks for the analysis of experimental data acquired by imaging muscle nicotinic acetylcholine receptors and the Alzheimer's disease-associated amyloid β pores with multiconductance levels in the plasma membrane of Xenopus laevis oocytes. We show that CellSpecks can accurately and efficiently generate location maps and create raw and processed fluorescence time traces; histograms of mean open times, mean close times, open probabilities, durations, and maximal amplitudes; and a "channel chip" showing the activity of all channels as a function of time. Although we specifically illustrate the application of CellSpecks for analyzing data from Ca²⁺ channels, it can be easily customized to analyze other spatially and temporally localized signals.

Figure 1

Illustration of CellSpecks algorithm processing steps: separation of noise from signal of nAChRs channel activity. (A) shows fluorescence trace from a single pixel (gray) and the final signal trace (black) calculated by CellSpecks. (B) illustrates the processing steps required to go from the original (gray) trace to the signal (black) trace. First, temporal smoothing creates a blur (blue) trace. A mode value is calculated for the entire blur trace (purple dotted line). SD, σ, is calculated for all fluorescence points from the original trace that are below the mode value (dark blue dotted line). Noise threshold is set as 2σ + mode (red dotted line) and a noise trace generated (green). Subtracting noise (green trace) from the original trace (gray), the signal trace (black) is obtained. In this figure, all time-varying quantities (original trace, blur, noise, signal) are plotted with solid lines, whereas time-invariant variables (mode, σ, noise threshold) are plotted with dotted lines.

Figure 2

CellSpecks graphical user interface. The CellSpecks interface includes the capacity to audit the detection and analysis process for quality and accuracy. The map of detected channels is shown in the Algorithm Output Images windows (left). Map is generated from a 5000 frames stack captured by imaging a 40 × 40 μm² region of an oocyte plasma membrane expressing muscle (αβγδ) nAChRs. Channels are represented by bright pixels at their determined location. The brightness of a pixel corresponds to the maximal amplitude event generated by the channel. Clicking on any pixel brings up the trace window (right), displaying the time traces such as raw (red), baseline or noise (green), and processed or signal traces (black) of the channel at that location. Results of any modification are displayed in the Algorithm Output Images. Information such as mean open and close times, P_O, peak amplitudes, lifetimes and amplitudes of all events for all channels, channel chip showing the activity of all channels detected as a function of time, and channel location maps for all channels can be exported as ASCII files for publication-quality plots and further analysis.

Figure 3

CellSpecks closely reproduces the channel locations and gating kinetics of simulated functional channels with random location and gating kinetics. Distributions of mean open times (A), mean closed times (B), and mean P_O (C) of true values (first row) and values estimated by CellSpecks (second row) for all channels are shown. (D) Channel-chip (image exported directly from CellSpecks) representation in which the horizontal axis is time (2 s total) and the vertical axis is the channel number (1–50)—i.e., each horizontal line represents one channel. The red and black represent the open and closed states of the channel, respectively. (E) The actual noisy (SNR = 5) time trace for a single channel and (F) the trace identified by CellSpecks are shown. (G) The number of channels (left panel) and number of events (right panel) identified by CellSpecks (green bars) versus the actual values (red bars) as functions of SNR. (H) Channel location maps from CellSpecks (green circles for SNR = 5 and blue × for SNR = 10) and true locations (yellow circles for SNR = 5 and red circles for SNR = 10).

Figure 4

Detection of hundreds of individual nAChR channels by CellSpecks. (A) A channel map showing the locations of 850 nAChRs channels within a 40 × 40 μm² membrane patch is shown. The map was generated automatically by CellSpecks after identifying coordinates of all channel sites through a 25 s imaging period, during which the oocyte was polarized to −150 mV in the presence of 1 μM acetylcholine. (B) An example of single-channel recordings (SCCaFTs) resulting from Ca²⁺ influx during channel opening when the membrane was hyperpolarized to −150 mV. Traces are obtained by monitoring fluorescence from regions of interest (one pixel, corresponding to a 0.33 × 0.33 μm² of plasma membrane) centered on three of the channel locations shown in (A).

Figure 5

Automated analysis of channel parameters. After locating the active sites (850) in the image field (Fig. 4A), CellSpecks can automatically generate analyses of the fluorescence events (5187) measured during the record, calculating the open times, close times, amplitudes for all events, and mean P_O for each corresponding region of interest. These data sets are then used to statistically analyze channel populations. (A) Distribution of the events duration for all of the detected nAChR channels shown in Fig. 4A. Data are fitted by single exponential decay (solid line) with the time constant 8.6 ms. (B) Distribution of the corresponding closed times (intervals between events) fitted by a double exponential decay function (solid line) with time constants of 123.8 and 1191.3 ms. (C) Distribution of the mean open probabilities calculated for the corresponding channels. Data are fitted by a double exponent decay function with P_O1 of 0.00041 and P_O2 of 0.0047. (D) Plot displays the maximal events amplitude distribution obtained measuring the peak fluorescence for each detected event. Distribution are fitted by a Gaussian function with a peak amplitude of 73 ΔF/Fo × 100 (solid line). To see this figure in color, go online.

Figure 6

Characterizing the activity of Ca²⁺-permeable ion pores formed by Aβ42 oligomers in a 40 × 40 μm² plasma membrane patch of an X. laevis oocyte imaged through TIRFM. The stack has 5000 frames recorded at 400 frames/s. In this example, 1 μg/mL of Aβ42 oligomers was applied to the bathing solution, and Ca²⁺ influx to the cytoplasm was enhanced by applying a hyperpolarizing potential of −80 mV. (A) Distribution of the channels’ mean events duration (channels mean open time) for 930 active sites. Double exponential fit (solid curve) yields decay constants of 7.2 and 20.1 ms. (B) Distribution of the mean closed times (times at which events were not present) fitted by a double exponential decay with constants values of 0.473 and 6034 s, respectively. The plot in (C) shows the corresponding distribution of channels P_O fitted by a double exponential decay function (solid line), with the curve yielding values of P_O1 = 0.0084 and P_O2 = 0.012. (D) The corresponding distribution of maximal amplitude per channel in which variations in amplitude among different channels are evident. (E) A sample fluorescence trace representing the Ca²⁺ influx through a single Aβ42 pore. Multiple conductance levels during individual events are indicated in the fluorescence trace and clearly shown in the idealized trace (F). (G) shows a single event in which the pore opens up to four conductance levels. Channel-chip representation of the gating of all channels detected in the stack (H) with progressively zoomed-in versions are shown in (I) and (J).