Scanless two-photon voltage imaging

Parallel light-sculpting methods have been used to perform scanless two-photon photostimulation of multiple neurons simultaneously during all-optical neurophysiology experiments. We demonstrate that scanless two-photon excitation also enables high-resolution, high-contrast, voltage imaging by efficiently exciting fluorescence in a large fraction of the cellular soma. We present a thorough characterisation of scanless two-photon voltage imaging using existing parallel approaches and lasers with different repetition rates. We demonstrate voltage recordings of high frequency spike trains and sub-threshold depolarizations in intact brain tissue from neurons expressing the soma-targeted genetically encoded voltage indicator JEDI-2P-kv. Using a low repetition-rate laser, we perform recordings from up to ten neurons simultaneously. Finally, by co-expressing JEDI-2P-kv and the channelrhodopsin ChroME-ST in neurons of hippocampal organotypic slices, we perform single-beam, simultaneous, two-photon voltage imaging and photostimulation. This enables in-situ validation of the precise number and timing of light evoked action potentials and will pave the way for rapid and scalable identification of functional brain connections in intact neural circuits.


58
Voltage indicators, which generate optical signals whose magnitude varies as a function of membrane 59 potential, promise to address many of the aforementioned limitations of GECIs 11 . Following the first optical

152
This is likely the result of two factors. Firstly, responsive pixels imaged with high contrast are more likely to 153 be retained in the second segmentation step. These pixels are those where the cellular equator coincided 154 with the focal plane, where the excitation power density (and presumably photobleaching) is highest.

155
Secondly, the voltage responsive fluorophores are more likely to be tethered to the membrane, less mobile 156 and hence more susceptible to photobleaching.

178
The SNR of the responses to 100 mV steps (Protocol 2) was higher at all excitation power densities with

207
Overall, these results confirm that 2P-TF-GPC, 2P-TF-Gaussian and 2P-TF-CGH can successfully be 208 applied to scanless two-photon voltage imaging, albeit with different advantages and limitations. Since the 209 7 SNR of data acquired using 2P-TF-CGH was significantly higher than for 2P-TF-Gaussian or 2P-TF-GPC, 210 we consider it the optimal modality for imaging large numbers of cells simultaneously, for short periods, 211 with a given incident power. For prolonged recordings (continuous illumination for hundreds of milliseconds   212   or more) of neurons labelled with JEDI-2P-kv, we would recommend 2P-TF-GPC or 2P-TF-Gaussian, since   213 we observed lower photobleaching and higher photorecovery with these methods than with 2P-TF-CGH.

214
Although no significant performance differences were found between 2P-TF-Gaussian and 2P-TF-GPC 215 ( Figure 2c, Supplementary Figures 8 and 9), 2P-TF-Gaussian requires higher power at the laser output for

223
We set out to identify the imaging conditions (specifically the power densities and acquisition rates) required 224 to observe neural activity ranging from high-frequency spike trains to sub-threshold depolarizations in 225 densely labelled samples. We also aimed to determine whether the necessary imaging conditions perturb 226 neural activity or otherwise impact cellular physiology. We performed simultaneous 2P-TF-GPC imaging 227 and whole cell-patch clamp recordings of granule cells located in the dentate gyrus (DG) of organotypic 228 slices bulk-transduced with JEDI-2P-kv (see Methods). Expression of JEDI-2P-kv in the granule cells of the 229 DG was well localised to the plasma membrane, with no evidence of intracellular aggregation (Figure 3a).

230
Even though granule cells are extremely closely packed in DG, due to the optical sectioning conferred by 231 temporally focused, targeted illumination, we were able to image individual neurons with high-contrast and 232 high-resolution in this challenging preparation ( Figure 3b).

255
Having established that it was possible to record APs with high SNR in single trials at different acquisition 256 rates, we next tested whether we could also monitor individual spikes within high-frequency trains of action 257 potentials (such as bursts) under these conditions. We observed that an acquisition rate of 500 Hz was

270
We next examined whether these conditions (power density: 1.11 mW µm -2 , corresponding to 125 mW per 271 cell, 1 kHz acquisition rate) were also suitable for imaging sub-threshold changes in membrane potential.

272
To emulate excitatory PSPs, patched cells were clamped to -75 mV, while the membrane potential was 273 varied in 0.5 mV steps from 0 to 2.5 mV for 20 ms. This protocol was repeated 50 times. Since it was not 274 possible to detect these transients from individual recordings (Figure 4a   Next, we tested the capability of scanless two-photon voltage imaging to record spontaneous network 279 activity, a fundamental feature of developing neural circuits 54 . We performed simultaneous 280 electrophysiological (whole cell patch clamp, (current clamp)) and fluorescence recordings (2P-TF-GPC, 281 power density: 1.33 mW µm -2 (150 mW per cell), 1 kHz acquisition rate) of spontaneous activity from 282 neurons in hippocampal organotypic slices which exhibited a range of different resting potentials (n > 10 283 cells; 5 slices). We were able to observe several hallmarks of spontaneously generated activity, large slow

290
To test the robustness of our approach for long-term recordings, we repeated the same protocol (2P-TF-

299
The imaging period was primarily limited by axial sample drift which decreased the SNR (data not shown),

330
To test the capability of multiplexed 2P-TF-GPC to image multiple neurons simultaneously we used a 331 custom low-repetition rate source (940 nm, pulse duration 100 fs, repetition rate 250 kHz, 600 mW average 332 output power). Low-repetition rate sources, used for two-photon optogenetics, can provide higher peak 333 energies and lower average power, hence potentially minimize photoinduced thermal effects and scale up 334 the number of neurons that can be imaged simultaneously. Low-repetition rate lasers are particularly well 335 suited for techniques with long times, such as scanless two-photon imaging. We found that using the low 336 repetition rate laser, action potentials could be detected in single trials with power densities as small as  found to be highly synchronized, a characteristic feature of the immature hippocampus 54 . Control traces 351 recorded adjacent to targeted neurons confirmed that this was not an artefact due to crosstalk. We were 352 able to combine data from separate acquisitions to perform voltage imaging throughout a large region (200 353 x 150 µm²) as demonstrated in Figure 6. Since this data was acquired on a prototype system, the period 354 between sequential acquisitions was on the order of seconds. However, by optimizing the acquisition 355 pipeline, the period between sequential acquisitions could feasibly be reduced to milliseconds to enable 356 scanless two-photon voltage imaging of populations of neurons.

357
The number of achievable targets per acquisition and/or the imaging depth could be further increased, as 358 demonstrated for two-photon photostimulation of multiple cells 55 using high-power, low repetition rate,

389
However, for power densities above 0.02 mW µm -2 (2.5 mW per cell), we were able to detect action 390 potentials in single trials and measured similar latencies to those obtained using whole-cell patch clamp 391 recordings (4.3 ± 0.2 ms, mean ± s.e.m.) and jitter on the order of a millisecond (Supplementary Figure   392 20c-d).

393
The action potential probability decreased as a function of stimulation frequency, feasibly a result of

416
These results demonstrate that simultaneous, scanless two-photon voltage imaging and photostimulation 417 can be performed in multiple cells simultaneously using high-energy, low-repetition rate lasers, using 418 powers well-below the damage threshold.

420
In this work, we introduced scanless two-photon voltage imaging and performed high-contrast, high-421 resolution voltage imaging of single and multiple neurons expressing the newly developed GEVI JEDI-2P-422 kv. Due to the axial confinement conferred by temporal focusing we were able to perform high-contrast two-423 photon voltage imaging in densely labelled intact brain slices.

424
We performed a thorough characterisation of three, temporally focused, parallel excitation modalities (2P-

445
However, these imaging speeds reduce the accuracy of action potential detection for spike trains with 446 frequencies > 100 Hz. The optimal imaging conditions will also depend on the characteristics of the specific 447 GEVI used. In principle, a major advantage of voltage versus calcium imaging in neuroscience is the ability 448 to detect sub-threshold changes in somatic membrane potential. In our current configuration, we found that 449 imaging small sub-threshold signals required averaging data from up to 25 individual trials to reach a SNR 450 above 1 for depolarizations larger than 0.5 mV. Whilst these findings highlight the challenges of detecting

14
Since the current implementation of scanless two-photon voltage imaging requires continuous illumination,

455
we carefully investigated whether the illumination conditions required to observe different aspects of neural 456 activity induced any observable physiological perturbations. We found that single cells could be imaged 457 using 100 fs, 80 MHz sources, at lower average powers than those commonly used for existing kilohertz 458 scanning microscopes applied to two-photon voltage imaging 28,58,59 and did not observe any changes in AP 459 properties at these powers. However, we found that the latency of electrically induced APs was increased 460 slightly at all powers tested, and irreversible thermal damage was revealed with 2P-TF-CGH at the highest 461 powers tested using immunohistochemistry. In contrast, Caspase-3 staining did not reveal any non-linear 462 damage at the investigated powers. In fact, we found that one of the biggest impediments to long-term 463 voltage imaging was sample drift, a problem we imagine will be significantly more severe for scanless two-

469
We also demonstrated that scanless two-photon voltage imaging could be performed with a much lower 470 effective repetition rate (250 kHz) than used with existing kilohertz scanning microscopes. Even with this 471 low-repetition rate laser, fluorescence was excited with more than 50x the number of pulses per voxel than 472 for the scanning approaches. As a result, scanless two-photon voltage imaging is much more robust to 473 fluctuations in output laser power than the scanning approaches. We demonstrated AP detection in single 474 trials using 15-30 times lower average power than required using the high-repetition rate laser. Under these 475 conditions, we did not find any histological evidence of thermal stress or physiological perturbations with 476 whole-cell patch clamp recordings. As demonstrated in the case of two-photon optogenetics 55 , the major 477 advantage of low-repetition rate lasers is that multiple targets can be illuminated simultaneously, whilst the 478 average power delivered to the sample is kept below the thermal damage threshold, and much lower than

485
In this work, we demonstrated that electrically evoked action potentials and spontaneous activity in

507
All experiments in this study were performed at relatively superficial depths (<50 µm) in scattering tissue 508 where the expression pattern of JEDI-2P-kv was confined due to the approach used for viral delivery (bulk 509 transduction). The next step will be to monitor the membrane potential of multiple neurons simultaneously  16 neuronal activity will also be a major addition in connectivity mapping to non-invasively confirm the 527 successful optical induction of an action potential in the potential pre-synaptic cell population. We 528 demonstrated that scanless two-photon imaging can be performed with any of the existing modalities used 529 for parallel two-photon photostimulation. As seen in our all-optical recordings (Figure 7d), the excitability of 530 the targeted cells can vary between cells and over time, advising a confirmation of the pre-synaptic spike 531 in each instance. In contrast to current connectivity mapping approaches, which do not confirm pre-synaptic 532 spiking 71 or use GECIs 70 , an approach using a voltage indicator would additionally reveal the precise timing 533 of the pre-synaptic spike, which would facilitate correlation of the post-synaptic response and discrimination 534 from noise. We anticipate that improved stoichiometric co-expression of the GEVI and the

552
We anticipate that the description and thorough characterisation of scanless two-photon voltage imaging 553 presented in this manuscript will motivate its application to deciphering the logic and syntax of neural   of L11 (f = 500 mm, (Thorlabs, AC508-500-B)), L12 (f = 300 mm, (Thorlabs, AC508-300-B)), L7 (f = 300 633 mm, (Thorlabs, AC508-300-B)) and the objective lens (Nikon, CFI APO NIR, x40, 0.8 NA, f = 5 mm, water) 634 was used to de-magnify and relay the holographic spots on to the sample plane. In experiments where non-635 temporally focused spots were used (1030 nm excitation), the diffraction grating was replaced by a mirror.

636
Excitation paths 1 and 2 were combined prior to the tube lens (L7) using a polarising beam splitter (Thorlabs, 637 PBS253). The linear polarisation of the light exiting the objective was changed using a half-wave plate 638 following the PBS. For each modality, the rotation of the half-wave plate was set to that which was found 639 to maximise the two-photon excited JEDI-2P-kv fluorescence.    for all experiments. These coordinates were used to "stitch" data from sequential acquisitions into a single 679 dataset. In "dynamic range" mode, the field of view is inversely proportional to the exposure time, such that 680 we could acquire data from 532 rows (86 x 250 µm 2 FOV) at 500 Hz. The field of view could be increased 681 by a factor of 6 using "speed" mode, where data is read out at 8-bit. The camera was triggered using a 5

775
In current clamp configuration, neurons were injected with some current (less than 100 pA) if necessary to 776 maintain their resting membrane potential to -75 mV. In the latter configuration, bridge potential was 777 corrected (Bridge potential = 13.9 ± 4.2 MΩ; mean ± s.d.).

778
Neurons were first patched in whole-cell voltage clamp configuration. Protocol 1 was then performed to 779 confirm that the fluorescence of the patched cell was voltage responsive (Supplementary Figure 12).

780
The ability to record single action potentials in neurons (Figure 3c

805
The axial resolution of JEDI-2P-kv (Supplementary Figure 12) was measured by electrically triggering an 806 action potential and measuring the fluorescence response while displacing the objective in the z axis (from 807 +50 to -50 µm, in 5 µm steps). The lateral resolution was measured (from +20 to -20 µm, in 2 µm steps) by 808 mechanically moving the sample in the x-y axis.

809
To measure the performances of JEDI-2P-kv under 1030 nm illumination, hippocampal organotypic slices 810 were infected with a mixture of AAV1_EF1a_DIO_JEDI-2P_Kv2.1_WPRE and 811 AAV9_hSyn_Cre_WPRE_hGH (see Table 2) at DIV 3 in order to get a sparser expression. Isolated

25
Various titrations were tested to achieve sufficient levels of expression of both sensor and actuator. When 833 slices were transduced with both viruses on the same day, we observed a reduction in the expression of 834 JEDI-2P-kv. Furthermore, overexpression-mediated apoptosis was observed in some cases when slices 835 were transduced with both viruses simultaneously. The best results were obtained by transducing slices 836 with JEDI-2P-kv first, followed a week later by ChroME-ST which resulted in strong co-expression of both 837 proteins (Figure 8a). However, in general we found that the expression levels of both proteins were more 838 variable when the two constructs were co-expressed than when either construct was expressed 839 independently.

854
For each cell, the power density was increased until a spike was detected optically in at least one of five 855 repeats. The final set of power densities used was between 0.02 -0.08 mW µm -2 (2.5 -9 mW) per cell.

860
Immunostaining was performed on hippocampal organotypic slices to assess the potential non-linear 861 photodamage induced by two different laser sources (A and C) during our experiments.

862
In the case of laser A, slices expressing JEDI-2P-kv were illuminated with a holographic spot (12 µm

878
Slices were incubated with primary antibodies diluted in a solution of BSA 5 % in PBS (Table 3)