Design for Fast Optogenetic Screen In Mammalian Cells

Abstract Genetically encoded fluorescent biosensors are proving to be powerful tools in neuroscience. The GCaMP6 Ca2+ sensor is widely used, and there are now many proof-of-principle versions for many second messengers that show promise. Improving these has been challenging because testing them involves a low throughput, labor-intensive processes. Our goal was to create a live cell system that uses a simple, reproducible, optogenetic process for testing prototypes of genetically encoded biosensors. Blue light was used to activate an adenylyl cyclase enzyme from the soil bacterium Beggiatoia (bPAC) that increases intracellular cAMP (Stierl et al. 2011) as detected by the red sensor R-CaDDis. In turn the cAMP opened a cAMP gated channel (olfactory cyclic nucleotide gated channel, CNG, or the hyperpolarization-activated cyclic nucleotide gated channel, HCN2). This produced slow Ca2+ transients as detected by R-GECO1.2. To speed these transients up, we added the inwardly rectifying potassium channel 2.1, Kir2.1, and the bacterial voltage gated sodium channel (NAVROSD). This is a modular system in which the kinds of channels, and their relative amounts, can be tuned to produce the cellular behavior crucial for screening a particular biosensor in an automated format.


32
Why are Ca 2+ sensors important?

33
The calcium ion is an important second messenger in cellular processes ranging from 34 neuronal function and muscle contractility, to fertilization and embryogenesis (Bear,Connors,35 sensors are being coupled with functional magnetic resonance imaging (fMRI) (Barandov et al. 48 2019;Ghosh et al. 2018). 49 There are now several Ca 2+ sensors that can sense Ca 2+ transients in cells with 50 millisecond response times and that are bright enough to image in living tissue (Ouzounov et 51 al. 2017). Most of these sensors have blue excitation and green emission peaks. Currently, the 52 sensors with the fastest kinetic response time to Ca 2+ flux are the Ultrafast GCaMP6f, jRCaMP 53 and jRGECO1a (Helassa et al. 2016;Kerruth et al. 2019). The recent jGCaMP7 sensor makes 54 it possible to capture individual action potentials and can track large neuronal populations 55 when using two photon or wide field imaging (Dana et al. 2019). 56 There is still a need for better Ca 2+ sensors that are red (Zhao et al. 2011), or near IR 57 (Qian et al. 2019), because these longer wavelengths enable investigators to image deeper 58 6 In theory, we could use similar channels to create cells that do not require field 104 stimulation, and have faster action potentials and whole cell Ca 2+ transients. We wanted the 105 system to be automatable, inexpensive, modular, and useful in many different cell types. Our 106 goal was to create a screening platform for biosensors that is consistent, reliable, and capable 107 of quickly screening thousands of prototypes. 108

109
Establishing an Optical Actuator 110 How can we actuate a biosynthetic cell line using light? To develop a optogenetically 111 controlled screen we chose to use bPAC (blue photo activated cyclase), a blue light activated 112 adenylyl cyclase from the soil bacteria Beggiatoia (Stierl et al. 2011). bPAC enzyme is 113 activated with 480 nm light which then converts ATP into cyclic adenosine monophosphate 114 (cAMP). R-caDDis is a fluorescent cAMP sensor that is excited with 561 nm light (Bernal 115 Sierra et al. 2018). We began by examining whether blue light stimulus can produce a 116 measurable change in cAMP levels in  We transduced HEK-293 cells with bPAC (2x10^11 Vg/mL, viral titer) and R-caDDis (5 x 118 10^10 Vg/mL) ( Fig.1 A). The following day the cells were stimulated with 10 milliseconds of 119 blue light and then images were collected continuously with 561 nm excitation that does not 120 activate bPAC. Upon blue light photoactivation, the R-CaDDis red fluorescence increased. 121 The quantification of the red fluorescence reveals that cAMP levels within the cell rise over the 122 time frame of ~100 seconds (Fig.1C). An obvious increase in red fluorescence can be seen 123 upon blue light activation and is demonstrated in the images of HEK-293 cells taken pre and 7 10 seconds post blue light stimulation (Fig.1B). Control cells with no bPAC did not show an 125 increase in red fluorescence. The system response is remarkably reproducible and the same 126 cells can be repeatedly stimulated. Indeed the only limitation is that too much blue light given 127 over a short period of time will create too much cAMP and saturate the sensor. The slow 128 decrease in fluorescence of R-CaDDis is most likely due to the phosphodiesterases present in 129 the cell which work to eliminate cAMP (Tewson et al. 2016). transduced with bPAC (the actuator) and red-caDDis, the red fluorescence cAMP sensor. B) 132 The cells were briefly illuminated with 20ms of 480nm light. This activation of bPAC increases 133 cAMP levels which increases R-cADDis fluorescence. Brightness and contrast were the same 134 for both images. C) 20 ms of blue light activates bPAC which produces bursts of cAMP over a 135 period of 100 seconds. This is a stereotyped response that is repeatable with multiple doses of 136 blue light. 137

138
To examine the dose-response relationship for blue light and cAMP production, the 139 intensity of the blue light was systematically varied during stimulation while quantifying the r-140 CaDDis fluorescence. HEK-293 cells were transduced with bPAC and R-caDDis, and each 141 well in a 96-well plate was given a 15 milliseconds dose of blue light. For each well, the power 142 of the LED 480nm light was increased stepwise from 0% to 100% in increments of five using a 143 THORLABS LED controller. At 100% power the light source measured 70mW/cm 2 . Figure 2  144 reveals a near-linear dose response in which increasing levels of blue light produces increases 145 in cAMP levels. Above 50% power to the LED, there was an increased variability in the 146 8 response that is most likely due to saturating the sensor with cAMP. It is possible that lowering 147 the level of bPAC expression would make it possible to create a system with a greater range of 148 tunability before saturating the sensor with cAMP. The fluorescence change, delta F, was used 149 to determine the cAMP level and all points in the dose response are an average of multiple 150 trials (Fig. 2). 151 Coupling the Actuator with an ion Channel

156
To couple cAMP production with depolarization, we expressed a cyclic nucleotide-gated 157 ion (CNG or HCN2) channel. The CNG channel is normally gated by cyclic guanosine 158 monophosphate (cGMP) (Fesenko, Kolesnikov, and Lyubarsky 1985). However, several 159 mutations can be introduced (Rich et al. 2000) that render it sensitive to cAMP. In theory, this 160 mutant channel should couple bPAC activation to a current that depolarizes the cell. In addition 161 the HCN2 is cAMP gated and must be hyperpolarized to activate(K. Chen et al. 2018). 162 We expressed a CNG channel (1.26 x 10^11 Vg/mL), bPAC, and R-GECO1 in HEK 163 cells (Fig. 3A). We triggered the activation of bPAC with blue light. Figure three illustrates that 164 upon 20ms of blue light activation of the enzyme, the red fluorescence of R-GECO increased. 165 The opening of the CNG channel increased Ca 2+ levels in the cell which was quantified by an 166 increase in red fluorescence levels of R-GECO1 [16] (Fig. 3). The Ca 2+ levels (Fig 3.    We choose the NavRosDg217A for two reasons. First, the human sodium channel has 196 an enormous coding region that exceeds the carrying capacity of most viruses. Bacterial 197 channel subunits are much smaller. Secondly, we wanted to create a system with long slow 198 depolarizations such that membrane depolarization events could be captured with slower 199 imaging speeds (10Hz). We transduced the bacterial channel NavRosDg217A (Nguyen,200 Kirkton, and Bursac 2016), Kir2.1, CNG, and R-GECO1 into HEK-293 cells. There is some 201 degree of spontaneous activity, but when we activate bPAC with 20ms of blue light there were 202 robust responses in all of the wells. There were obvious Ca 2+ transients that occurred in 90% 203 of the cells, but the activity was largely uncoordinated. Each cell has its own "signature" 204 change in Ca 2+ levels, as seen in the R-GECO1 fluorescence traces (Fig.4B). The 205 fluorescence traces encode the change in the intracellular Ca 2+ concentration levels of the cell. 206 Optimizing the Activity

211
Our goal was to create a highly reproducible system that can be optimized. We propose 212 that different levels of the Kir2.1 and NavD would lead to varying responses. We systematically 213 transduced cells with varying ratios of Kir2.1 and NavD while holding the levels of bPAC and 214 CNG channels constant. The different ratios of the channels lead to different responses. To 215 quantify the Ca 2+ imaging data we analyzed data using a custom made Matlab Program (Fig.  216 5A). We wanted to break down each Ca 2+ transient into components that could be analyzed 217 and looked at in multiple ways. For each Ca 2+ transient we examined the prominence of the 218 peak, the full width of the peak at half maximum (FWHM), and the inter-peak interval. Kir2.1 to NavD ratio of 14:1. 228 The Peak Finder in Matlab marks the peaks by their location in time and their maximum 229 fluorescence intensity. The peak intensity was then measured from the base fluorescence to 230 the peak that was pre-determined using the Matlab Peak Finder (Fig. 5A). The peak intensity 231 of the events does not appear to be random. Histograms of peak intensity (Fig. 5D) show what 232 appear to be units of activity as though the Ca 2+ response is integrating one, two, or three 233

events. 234
In addition, we examined the width of the peak which was measured as the full width of 235 the peak at half maximum (Fig. 5A). The peak (FWHM) of the Ca 2+ event is the length of time 236 that the Ca 2+ concentration stayed elevated for each defined peak. The fastest event occurred 237 for 120ms, to ensure that each event was captured we sampled with continuous 40ms 238 exposures. The fastest average event (Peak FWHM) for a Ca 2+ transient occurred in cells with 239 a Kir2.1 to NavRosD viral infectious unit ratio of 22:1 and generally lasted for ~400ms. Next, 240 we measured the inter-peak interval, which is the difference between two consecutive peaks. 241 The inter-peak interval is a measure of how frequently Ca 2+ transients occur in a given system 242 after a previous response. Again, the cells that exhibited the shortest average inter-peak 243 interval were cells that had the Kir2.1:NavD ratio of 22:1, although there was a great deal of 244 cell to cell variability. A single burst of blue light produced activity for an average of two-three 245

247
The behavior of HCN2 was quite different than CNG. Figure   Inter-Peak Interval time increased in concordance to increased Ca 2+ event widths. The 256 standard deviation of the mean is shown. 257 We found that the HCN2 produced longer inter-peak intervals, and longer Ca 2+ 258 transients, than cells expressing the CNG channel (Fig. 6). We found that the fastest events 259 occurred in cells with a viral unit ratio of Kir2.1 to NavD of 14:1 with an average of ~3 seconds 260 (Fig.6b). Compared to cells expressing CNG, the HCN2 cells in general had Ca 2+ transients 261 that lasted around seven seconds with the time between transients ranging between 30 to 60 262 seconds. Importantly, the HCN2 cells show continued Ca 2+ transients that lasted for up to 14 263 minutes after a single stimulus of blue light. 264

265
Although there are several well engineered Ca 2+ biosensors that are green, there 266 is still room for improvement, and continued engineering on green Ca 2+ sensors persists. 267 Further, if it is possible to image both red and green Ca 2+ sensors simultaneously, that can be 268 used to benchmark the other in terms of speed, signal amplitude, and linearity. G-GECO1 is 269 based on the GCaMP sensor and is based on the circularly permuted GFP with a fusion to the 270 M13 and calmodulin at the N and C termini (Zhao et al. 2011). It is pH-sensitive and is 271 relatively dim until it is excited, at which point it has a two fold increase in fluorescence 272 compared to GCaMP3 (Zhao et al. 2011). 273 Imaging G-GECO while continuously activating bPAC was generally straightforward. 274 We held the virus amount of Kir2.1 at 4x10^1 4 VG/ L , and NavD at 1.7x10^1 3 VG/ L, at a viral 275 ratio of Kir2.1:NavD at 22:1. The CNG channel viral titer was held at 2.5x10^1 4 VG/ L and we 276 expressed G-GECO1 heterologously (Fig. 7A). We varied the viral titer of bPAC from .5 -4 x 277 10^1 6 VG/ L . We found that decreasing the bPAC dilution to 1 x 10^1 6 VG/ L decreased the 278 width of the Ca 2+ event, as well as decreased the time between events (Supplemental Fig.1). 279 In principal an image splitter could be used to compare red sensors of unknown speed with a 280 benchmarked green sensor. This experiment shows that the system can be continuously 281 activated using very low power light. 282

290
Our goal was to create a simple, modular, reproducible system to screen Ca 2+ sensors. 291 The system can be optimized by varying the ratio of Kir2.1 to NavD channels, thereby 292 changing the type of Ca 2+ transients that occur and the frequency at which they occur. 293 Therefore, optimizing the ratio of the channels to the experimenters preference for cell type 294 and sensor can empower an automated screen for Ca 2+ sensors with different dynamic 295

properties. 296
The designing of the high throughput screen for Ca 2+ sensors can be modified. If a slow 297 sensor is preferable, the HCN2 channel with a greater Kir2.1 to NavD ratio is desirable. 298 Otherwise, to discover a fast sensor with increased linearity to Ca 2+ dynamics, the CNG 299 channel with a fast sodium channel would be favorable. For example, the Ca 2+ transients width 300 (length of time transient occurs) could potentially be greatly reduced by using a faster voltage 301 gated sodium channel. In addition, several papers have shown that including the connexin-43 302 protein increases the gap junction coupling of the HEK-293 cells (Nguyen, Kirkton, and Bursac 303 2016;Kirkton and Bursac 2011;Fahrenbach, Mejia-Alvarez, and Banach 2007). We did not 304 test how connexin-43 might affect this system but speculate that it could potentially stabilize 305 the activity between the cells allowing for more consistent oscillations. If the throughput needs 306 to be increased, the screening process could be broken into two steps. First, the screen could 307 use a HCN2 channel (Fig.6A) and the cells could be imaged slowly without creating large data 308 files. The best variants could then be screened in cells expressing the CNG channel and a fast 309 sodium channel. These prototypes will be expressed in cells that also express a benchmark 310 sensor such as GCAMP7f. Finally, the most promising variants would then be photo physically An ASI modular microscope

353
The experiments were conducted using wide field imaging on an ASI-XYZ stage fitted 354 with a modular infinity microscope. The objective imaged onto the detector chip of a 355 Hamamatsu ORCA-Flash 4.0 scientific-CMOS camera. The images were collected using 356 either matlab scripts that controlled the camera, ASI stage, SH1-Thorlabs shutter, and 357 ThorLabs DC4100 Four Channel-LED Driver, or by using μmanager (Edelstein et al. 2014). A 358 custom illumination system was arranged. Briefly, a dichroic mirror (XX) was positioned at the 359 entrance of the microscope to combine blue light LED illumination with yellow laser 360 illumination. An SH1-Thorlabs shutter was used to create brief illumination from a blue 361 ThorLABS LED (488nm) for rapid stimulation of bPAC. At 100% the blue light illumination was 362 70mW/cm 2 . The yellow illumination was provided with a Sapphire laser(561nm, 50mW, 363 Coherent). The laser beam was steered with two mirrors (arranged in a periscope) into an 364 entrance aperture of the ASI microscope. Before entering the microscope, the laser beam was 365 diffused with a 50 0 diffuser (ED1 C50 MD, Thorlabs) placed in the focus of an f = 20 mm 366 aspheric lens (ACL2520U, Thorlabs) that collimated the beam for further traveling into the 367 microscope. The illumination intensity at 100% of the laser power at the sample was 368 10mW/cm 2 . 369

370
Image data was stored in a Z-stack tiff file and loaded into the FIJI distribution of the 371 ImageJ software (Schindelin et al. 2012). The background fluorescence was corrected using 372 the photobleaching gui in FIJI. The cells were hand selected using a freehand ROI surrounding 373 the cell of interest. The average pixel value within the ROI for each frame was then loaded into 374 MatLab. Analysis was built in Matlab using the Find Peaks in the Signal Processing Toolbox. 375 The find peaks returns a vector with the local maxima (peaks) from the ROI trace data. In 376 addition, it returns the widths of the peak and the prominence of the peak. The inter-event 377 interval is the vector difference between each consecutive peak. 378 Brightness and contrast were the same for both images. C) 20 ms of blue light activates bPAC which produces bursts of cAMP over a period of 100 seconds. This is a stereotyped response that is repeatable with multiple doses of blue light.