Optically activated, customizable, excitable cells

Genetically encoded fluorescent biosensors are powerful tools for studying complex signaling in the nervous system, and now both Ca2+ and voltage sensors are available to study the signaling behavior of entire neural circuits. There is a pressing need for improved sensors, but improving them is challenging because testing them involves a low throughput, labor-intensive processes. Our goal was to create synthetic, excitable cells that can be activated with brief pulses of blue light and serve as a medium throughput platform for screening the next generation of sensors. In this live cell system, blue light activates an adenylyl cyclase enzyme (bPAC) that increases intracellular cAMP (Stierl M et al. 2011). In turn, the cAMP opens a cAMP-gated ion channel. This produces slow, whole-cell Ca2+ transients and voltage changes. To increase the speed of these transients, we add the inwardly rectifying potassium channel Kir2.1, the bacterial voltage-gated sodium channel NAVROSD, and Connexin-43. The result is a highly reproducible, medium-throughput, live cell system that can be used to screen voltage and Ca2+ sensors.


Why are biosensors important?
Genetically encoded, fluorescent biosensors are powerful tools for studying cell signaling in real-time [1,2]. They are minimally invasive, and since they are genetically encoded their expression can be targeted to specific cell types and tissues [3][4][5]. However, the short wavelengths of light needed to image green biosensors can heat and damage the brain [6]. There is a need for better biosensors that emit red [7] or near-infrared [8] light because these longer wavelengths enable investigators to image deeper into thick tissues. Furthermore, increases in the signal to noise ratio of these sensors empowers scientists to collect more information from ever increasing numbers of cells [9,10]. An additional benefit of brighter sensors with high SNR is that they can be detected at lower expression levels. Higher levels of Ca 2+ sensor expression levels can lead to epileptiform activity [11].
A proven approach to expanding the fluorescent biosensor toolbox involves screening thousands of prototypes, evolving for optimal function [12][13][14][15]. One way this has been done is expressing mutagenized libraries of Ca 2+ sensors in E. coli. One study used a microfluidic measurable change in cAMP levels in bPAC-expressing HEK293 cells as seen with R-cADDis, a red fluorescent cAMP sensor [28]. We transduced HEK293 cells with BacMam viruses to co-express bPAC and R-cADDis (S1 Table). The following day, we stimulated the cells with 20 milliseconds of blue light and then collected images continuously with 561 nm excitation light that does not activate bPAC ( Fig  1A). Upon the blue light stimulation, the R-cADDis red fluorescence increases (Fig 1B). cAMP levels within the cell rise over the time frame of~100 seconds ( Fig 1C). Control cells with no bPAC did not show an increase in red fluorescence. The system response is remarkably reproducible and the same cells can be repeatedly stimulated. Indeed the only limitation is that too much blue light given over a short time will create too much cAMP and saturate the sensor. The slow decrease in fluorescence of R-cADDis is most likely due to the phosphodiesterases present in the cell which work to eliminate cAMP [29].

Coupling the actuator with an ion channel
To couple cAMP production with channel activation, we expressed a cyclic nucleotide-gated ion channel. We tried both the rat olfactory CNG and the human heart HCN2 channel (S1 Table). The rat olfactory CNG channel is normally gated by cyclic guanosine monophosphate (cGMP) [30]. However, several mutations (δ61-90/C460W/E583M) can be introduced [31] that renders it sensitive to cAMP. In theory, this mutant channel should couple bPAC activation to a current that depolarizes the cell, creating Kuhl-C cells. To create Kuhl-H cells, we expressed the human HCN2 pacemaker channel which is also cAMP-gated and is hyperpolarization-activated (S1 Table) [32]. Once the cell is hyperpolarized, the HCN2 channel can open when bound to cAMP.
We activated the Kuhl-C or Kuhl-H cells with blue light to increase cAMP. Fig 2 illustrates that 20ms of blue light activation of bPAC, produces an increase in the red fluorescence of Ca2 + sensor R-GECO1 [7], indicating a gradual increase in whole-cell Ca 2+ over several minutes. Interestingly, blue light stimulation also produced bright subcellular flashes of fluorescence in both the Kuhl-C and Kuhl-H cells. The source of this Ca 2+ is not clear, but one possibility is that increases in cAMP could be causing brief IP 3 channel opening, or perhaps mitochondrial release [33] (Fig 2C and 2G). These small flashes of Ca 2+ last~20 seconds (Fig 2D and 2H and S1 Movie). The control cells without bPAC showed no such responses (S1 Fig).

Addition of Kir2.1 coupled with bPAC's cAMP activity leads to fluctuating whole-cell Ca 2+ transients
The co-expression of bPAC and CNG, or HCN2, produced reliable Ca 2+ transients, but these were long-lived events with slow kinetics. Introducing a voltage-regulated inward-rectifying potassium channel should increase the speed and amplitude of these transients by creating a greater driving force on Ca 2+ to enter the cell. Wild type HEK293 cells have a resting membrane potential of only -25 mV, but Kir2.1 expression shifts their resting potential to -70mV [20,[34][35][36].
We co-expressed Kir2.1, bPAC, R-GECO1, and a cyclic nucleotide-gated channel in HEK293 cells to produce Kuhl-CK and Kuhl-HK cells (S1 Table). In the Kuhl-C/HK cells, there is some degree of spontaneous activity, meaning Ca 2+ fluorescence fluctuations, before blue light stimulation. There are Ca 2+ transients that occur in 85% of the cells, but the activity is largely uncoordinated. Each cell has its own "signature" change in Ca 2+ levels, as seen in the R-GECO1 fluorescence traces (Fig 3B). Ca 2+ transients can be seen for up to 30 minutes following the blue light stimulus (S2 Fig). The histogram shows the average cumulative number of fluorescence peaks per cell over 340s (Fig 3E). The time scale is the same for each bar within  Table). B) The cells were illuminated with 20ms of 480nm light. This activation of bPAC increases cAMP levels which increases R-cADDis fluorescence. Brightness and contrast are the same for both images. C) The traces are a discontinuous recording of R-cADDis in the same cell (100s on, 20s off, 100s on), illustrating that the bpac response can be activated multiple times. Twenty milliseconds of blue light at 20s and 140s activates bPAC which produces bursts of cAMP for 100s. This is a stereotyped response that is repeatable with multiple doses of blue light.

PLOS ONE
We wanted to break down each Ca 2+ transient into quantifiable components that could be compared across different conditions (Fig 3F). We defined a transient as a change in

PLOS ONE
fluorescence that exceeded the standard deviation of the variability in the baseline by a factor of 26 [37]. For each fluorescent peak, we quantified the total cumulative peaks and the total number of peaks per ROI (S4-S10 Figs). The normalized 4F of the peak (amplitude), the full width of the peak at half maximum (FWHM) and the inter-peak interval (S4-S10 Figs). Also, we examined the total number of Ca 2+ transients recorded per experiment and recorded the time it took for a Ca 2+ transient to occur post blue light stimulation.

Voltage-gated sodium channel NavD creates faster Ca 2+ transients
The bacterial sodium channel, NavRosDg217a (NavD) is characterized by a rapid activation state that depends on the cell depolarizing to -30mV, and a slow inactivation state (hundreds of milliseconds) that starts during the depolarization [34]. Co-expressing the NavD channel with Kir2.1 can potentially create a faster excitable system because the Kir2.1 hyperpolarizes the cell after each spontaneous depolarization caused by the sodium channel. We chose NavD because the mutant is slower than the mammalian sodium channels and in theory, we could create a system with depolarizations such that membrane depolarization events could be captured with slower imaging speeds (10Hz). A recent paper by Chen et al. found that varying the levels of Kir2.1, and Cav1.3 or Nav1.5 affected the membrane oscillations as recorded by electrophysiology [32]. We hypothesized that different levels of the Kir2.1 and NavD would lead to different corresponding Ca 2+ responses.
HEK293 cells have endogenous connexin proteins, primarily connexin-45, that create gap junctions that electrically couple the cells [38,39]. However, Kirkton and Bursac [20] found that when action potentials occur in a confluent layer of HEK293 cells expressing Kir2.1 and Nav1.5 they do not uniformly spread through the monolayer. Additional Connexin-43 produced rapid and robust action potentials. This is consistent with several other studies that have shown that including the Cx-43 protein increases the electrical gap junction coupling of HEK293 cells [20,34,40].
To test this we systematically transduced cells with varying ratios of Kir2.1 and NavD with and without Cx-43 while holding the viral concentrations of bPAC, CNG and R-GECO1 constant. The Kuhl-CKNa cells were transduced (S1 Table) and~48 hours later the cells were stimulated with 20ms of blue light and images were collected with 561nm excitation. Upon activation of bPAC there were varying responses in each condition that are quantified (S4 and S5 Figs). Interestingly there seems to be a significant effect on the responses in the Kuhl-CKNaCx cells (Fig 4D) depending on the amount of (Kir2.1):(NavD) VG/μL applied.
To determine the optimum (Kir2.1):(NavD) ratio we examined the average cumulative chance that a cell will fire by analyzing the total number of peaks for a well/total number of points collected from Inter-Peak interval (Fig 4B and 4D). Using the data from the Kuhl-CKNa cells we used the total number of peaks per ROI (S4D Fig) and ran a Dunnett's test (S2 Table) to examine if there was a statistical significance between the conditions. (Kir2.1 1e8): (NavD 5e8) VG/μL had a significant increase ( ���� p � .0001) in the average cumulative number of peaks per cell ( Fig 4B). In the Kuhl-CKNa system, there is a 35% chance that a cell will produce a significant Ca 2+ peak and out of the cells that were suprathreshold there is a 43% chance of a second peak within 66s of the initial fluorescence peak (S4F Fig).
While there appears to be a range for optimal concentration using comparable amounts of the two channels in Kuhl-CKNaCx cells (Kir2.1 1e8) and (NavD 1e8) VG/uL evokes a significant response because there is a 40% chance that out of 150 cells a significant peak will occur in any given cell in 210s. Of the cells that were suprathreshold (S5D Fig Table) to examine if there was a statistical significance between the conditions. The experimental trials that were statistically greater than the control were marked ( ���� p � .0001, ��� p � .0002) on the graph (Fig 4D).
Introducing the NavD and Kir2.   Table). B) Shown is the average cumulative fluorescence peaks per cell across 210s for each indicated viral concentration of Kir2.1 and NavD (n = 150 cells per condition). The duration of imaging post blue light stimulus is the same for each condition (colored bar) in the graph.

PLOS ONE
has a Ca 2+ bursting rosette pattern following a stimulus, while HCN2 seems to have a spreading wave-like pattern (S2 and S3 Movies).

Repeatability
Several iterative rounds of optimization were used to identify the right expression levels of each component to increase activity in terms of the total Ca 2+ transients in every cell, and the total number of cells within the field of view having a significant Ca 2+ peak (S4-S11 Figs). The optimized Kuhl-KCNaCx cells show remarkable, consistent activity when stimulated with blue light. In theory these Kuhl cells could be used to screen prototype activity sensors, new GECI and GEVI biosensors. Such a screen, however, would depend critically on consistent well to well activity. To measure how consistent the activity is from well to well, we integrated the activity seen in 14 different wells of cells. The representative cartoon (Fig 5A) depicts all the components transduced into the Kuhl-KCNaCx cell (S1 Table). One hundred percent of the 600 cells in the field of view will produce a significant peak. Fifty percent of the cells are likely to produce a second significant peak within~94s ±32. There is a 100% chance that 1,500 cells will produce a significant peak within 15 minutes (S10F Fig). Thirty percent are likely to produce a second significant peak within~148s±89. Tukey's multiple comparison test did not find a significant difference between the average cumulative number of peaks per cell for the 14 trial wells with Kuhl-CKNaCx.

Comparing green and red sensors in the same cell
While slow and laborious, an advantage of patch clamp fluorometry is that whole-cell voltage changes can be directly compared with fluorescence changes [41]. The all-optical Kuhl cell approach does not have the advantage of such an independent measurement. However, we reasoned that different colored sensors could be co-expressed and simultaneously imaged to arrive at a comparative measure. For example, a green sensor of known properties could be compared to a new one with red fluorescence. To test the feasibility of this approach, we used an optical splitter to compare the responses of R-GECO1 and G-GECO1 within the same cell.
The Kuhl-KCNaCx cells were transduced and 48 hours following the cells were imaged. In Fig 6 an image created by the OptoSplit II is shown where the green and red images of both sensors are simultaneously collected on two different portions of the EMCCD camera (S5 Movie). The fluorescence response from both sensors is plotted in (Fig 6C and 6D). This simple analysis shows that differences can be easily measured (Fig 6C and 6D) where the Tau on/off of both signals was measured. R-GECO1 Tau on was consistently slower than G-GECO1 and R-GECO1 Tau off was consistently faster than G-GECO1 Tau off . Previous photophysical characterization of G-GECO1 and R-GECO1 show that there are relatively large differences between the two sensors. G-GECO1 has a faster K on and R-GECO1 has a faster K off [7]. G-GECO1 has a greater K d, kinetic with a difference of 347 nM 1 .

Green voltage sensors can be screened in Kuhl-CKNaCx cells
ArcLight is a downward going voltage sensor, meaning that it gets brighter when the membrane is hyperpolarized, and dimmer when it is depolarized [42]. It is pH sensitive and has been demonstrated to work in single-cell fly neurons, but requires high-intensity illumination to achieve an adequate signal-to-noise ratio [43]. Slow depolarizations can be seen in the Kuhl-CKNaCx cell system (Fig 7) with a reasonable SNR. Comparing baseline fluorescence variation (standard deviation) with the amplitude of the fluorescence change reveals from 15 different cells an SNR of 5.5. The Kuhl-CKNaCx system shows promise for screening voltage sensor prototypes (S6-S8 Movies).

Discussion
This Kuhl cell system was designed to be customizable and there are several important considerations to take into account when customizing for a particular activity sensor. Multiple R-GECO1 fluorescence and their respective responses can be compared and plotted over time. The kinetics of both the red and green sensor from the same ROI is measured and plotted below the fluorescence trace. D) Representative traces from the upper right ROI in panels 6B. Both the R-GECO1 and G-GECO1 fluorescence and their respective responses in the same cell can be compared and plotted over time. The Tau on/off of both the red and green sensor from the same ROI is measured and plotted below the fluorescence trace. The difference between G-GECO1 and R-GECO1 can be seen because it has a~2 fold increase in the speed of its kinetics (K d ).
https://doi.org/10.1371/journal.pone.0229051.g006 PLOS ONE channels could be exchanged for the ones used in this study. For example, a faster voltagegated sodium channel would most likely create a faster action potential in the cells.
The Kuhl cells can potentially be used in medium throughput screens for better activity biosensors. The throughput of any screen, however, depends on the time budget necessary to scan a 96 or 384 well plate. In this study we arbitrarily defined activity as fluorescence peaks that exceeded the SNR of 26. That is, the peak was 26 times larger than the standard deviation of the baseline noise. Using this threshold, we found that out of 150 cells, imaging for 80s±30 was sufficient to detect 100 cells that would produce a Ca 2+ peak above threshold. Recording the activity in a well for 80s would eliminate false negatives. The imaging time, plus the plate translation time, should result in a time budget of roughly 100 minutes per 96 well plate. While 400 to 500 wells a day is medium throughput, it is considerably better than the manual patch clamp fluorometry methods in use today.
We have shown that screening both red and green GECIs simultaneously in the Kuhl-CKNaCx system is possible, and that one sensor can be used to benchmark the other. Individual cells' dual GECI response can be measured for its brightness, kinetics (Tau), SNR, and sensitivity.
The improvement of the genetically encoded voltage indicators-GEVIs-has been slow due to limitations in the throughput involved in manual, whole-cell patch clamp fluorometry [13,18]. In principal, the system described here will enable investigators to screen without field stimulation or any manual intervention. Voltage changes along the membrane of the HEK293 cells are visible and can be tracked in the Kuhl-CKNaCx cells using the GEVI Arc-Light, since the action potential in these cells is quite slow, lasting seconds~4s. We suggest that because we were able to benchmark Ca 2+ indicators simultaneously that this should also work for voltage sensors. The prototype sensors could be expressed in Kuhl cells that also express a benchmark sensor such as GCAMP7f or Voltron [1,44].

Plasmids
The HCN2 plasmid was kindly donated by Joan Lemire at Tufts University from Dr. Michael Levin's lab. The NavRosDg217a was reverse translated using a human codon preference and synthesized as a gBblock by IDT (Coralville, Iowa). The cyclic nucleotide-gated olfactory channel (from Rattus norvegicus) has the following mutations, δ61-90/C460W/E583M and was synthesized as a gBblock by IDT [45]. pGEMTEZ-Kir2.1 was a gift from Richard Axel, Joseph Gogos & Ron Yu (Addgene plasmid #32641).

Transduction
HEK-293 cells were prepared and transduced following the manufacturer's recommended protocol. Briefly, HEK293 cells were plated in 96 well plastic plates (48 x10 3 cells/well) and transduced with the appropriate mix of viruses and the HDAC inhibitor sodium butyrate (2mM final concentration). Two days later, the media was exchanged with PBS, tyrodes buffer, or Fluorobrite. The best results were consistently obtained with PBS.

An ASI modular microscope
The experiments (Figs 1-5 and 7) were conducted using wide field imaging on an ASI-XYZ stage fitted with a modular infinity microscope. The objective imaged onto the detector chip of a Hamamatsu ORCA-Flash 4.0 scientific-CMOS camera. The images were collected using either MATLAB scripts that controlled the camera, ASI stage, SH1-Thorlabs shutter, and ThorLabs DC4100 Four Channel-LED Driver, or by using μManager [46]. A custom illumination system was arranged. Briefly, a dichroic mirror was positioned at the entrance of the microscope to combine blue light LED illumination with 561 laser illumination. An SH1-Thorlabs shutter was used to create brief illumination from a blue ThorLABS LED (488nm) for rapid stimulation of bPAC. At 100% the blue light illumination was 70mW/cm 2 . The yellow illumination was provided with a Sapphire laser (561nm, 50mW, Coherent). The laser beam was steered with two mirrors (arranged in a periscope) into an entrance aperture of the ASI microscope. Before entering the microscope, the laser beam passed through a 50 0 diffuser (ED1 C50 MD, Thorlabs) placed in the focus of an f = 20 mm aspheric lens (ACL2520U, Thorlabs) that collimated the beam for further traveling into the microscope.
Optosplitter. To simultaneously collect images of both G-GECO and R-GECO, we used an optical splitter (CAIRN optosplit II, Faversham Kent ME13 8UP). This makes it possible to simultaneously collect both the green and red images on a single camera. An Olympus IX81 was fitted with LED illumination (470 nM, 565 nM, ThorLabs) where the power to the LEDs can be independently adjusted. A Pinkel filter set was used to simultaneously image both G-GECO and R-GECO (GFP/DsRed-2X-A-000, Semrock, Rochester, New York). The CAIRN splitter was positioned in between the microscope and EMCCD camera such that a dichroic filter (561 nM, Semrock, Rochester, New York) split the green and red emission into two optical paths. The red emission passed through an additional band pass filter (617/70, Semrock, Rochester, New York) to eliminate any potential green emission from the G-GECO.
Analysis. Image data was stored in a Z-stack tiff file and loaded into the FIJI distribution of the ImageJ software [47]. The initial frame of the stack was eliminated from all analysis to correct for photoactivation of R-GECO1 upon bluelight stimulation of bPAC [48]. The cells were selected using a freehand ROI surrounding the cell of interest. The average total intensity pixel value within the ROI for each frame was collected using the time series analyzer plugin and saved as a .csv file. The raw fluorescence trace data was loaded into MATLAB.
The analysis was done in MATLAB using the Find Peaks function in the Signal Processing Toolbox. The raw fluorescence traces were rescaled from zero to 10,000 and normalized to zero. The data was smoothed using 'sgolay' function. The main peak prominence threshold was set to 50.5 with an SNR of 26±10. The Signal-to-noise ratio was defined as the average of the peaks amplitudes that were defined using the peak finder thresholder by the standard deviation (S.D.) of the signal before blue light stimulus. We examined the SNR for Kuhl-CKNaCx cells over 5 individual trials which were averaged over 15 trials. The FindPeaks function returns a vector with the local maxima (peaks) from the ROI trace data. The Peak Finder in MATLAB marks each fluorescence peak by its location in time and records the number of peaks per ROI. In addition, it returns the widths of the peak and the prominence of the peak.
The time that a fluorescence peak occurred during imaging was recorded. The inter-event interval is the absolute difference between each consecutive peak. The FWHM and inter-event interval time was adjusted for exposure times. The total cumulative number of peaks for an entire trial was measured. Each fluorescence peak was measured from zero to the maximum peak intensity (normalized F) using the MATLAB Peak Finder. The time and total quantity of a fluorescence peak in a ROI were recorded. To determine the chance that a cell will fire twice was determined by (total number of peaks for a well/total number of points collected from Inter-Peak interval).