Volitional modulation of optically recorded calcium signals during neuroprosthetic learning.

All animal procedures were performed in accordance with UC Berkeley Animal Care and Use Committee regulations. 6 C57BL/6J and 4 CD1 male wild-type mice were used in these experiments, ranging in age from postnatal day 30-45. Animals were housed with a 12h dark – 12h light reversed light cycle. All behavioral tests were performed in the same cohort of animals.

Surgery

Mice were anesthetized using isoflurane (2% vol isoflurane/vol O2) and placed in a stereotaxic apparatus. Body temperature was maintained at 37°C using a feedback-controlled heating pad (FHC, 40-90-8D) and a small incision was made in the scalp. The skull was cleaned and a steel headplate was affixed over M1 (1 mm rostral, 1 mm lateral to Bregma) or S1 (1 mm caudal, 3 mm lateral to Bregma) using Metabond dental cement (Parkell, S380). A 3 mm craniotomy was opened over M1 or S1, and 200 nL of AAV2.9 Syn.GCamp6f.WPRE.SV40 20 (University of Pennsylvania Vector Core) was injected 250 μm below the pia using a Nanoliter 2000 injector (World Precision Instruments). The tracer was delivered using a pulled glass pipette (tip diameter 40-60 μm) at a rate of 40 nL/minute. The pipette was left in the brain for 10 minutes after completion of the injection to prevent backflow. After the pipette was removed, the brain was covered with silicone oil (Sigma product # 181138) and a glass coverslip was affixed to the skull with dental cement, as previously described 21 . We allowed 2 weeks for recovery and gCaMP6f expression.

Two-Photon Imaging

In vivo imaging was performed with a Moveable Objective Microscope (Sutter) using a Chameleon Ultra Ti:Sapphire mode-locked laser (Coherent, Santa Clara CA) tuned to 900 nm. Photons were collected with a Hamamatsu photomultiplier tube (H10770PA-40) using a Nikon objective (16x, 0.8 NA). Animals were head-fixed on a custom-made spring mounted imaging platform and placed under the 2p microscope. This setup allowed them to run freely, and their movements were recorded by an accelerometer fixed to the underside of the platform. Frames of 128×512 pixels (~160 × 160 μm) were collected at 7.23 Hz using ScanImage software 22 at 130-180 μm below the pia. The same imaging fields were used every day, localized by landmarks in the surface blood vessels. Imaged fields were stable over the course of training, and because the cortex was stabilized by a snugly fitting coverslip, only severe movements caused motion artifacts. Motion correction for slow drift in the imaging field was performed manually. Any period of gross movement during the task that caused cells to move out of their ROIs resulted in poor task performance, as ΔF/F of E1 was reduced. In this sense, mice were punished for excessive movement and seem to have learned to remain still during the task ( Supplementary Figure 2c ). Behavioral Task Two ensembles of 1-11 single cells each were chosen for inclusion in the “output” population. Cells with bright nuclei, indicating over-expression, were excluded, as were cells with many, poorly separable calcium events, an activity pattern indicative of fast-spiking interneurons. No other selection criteria were used to partition the recorded cells into each ensemble. We also ensured that many cells with good signal were included in the indirect population to enable a proper comparison. The cells assigned to the output population were changed on some days. Ensemble activity was measured as mean ΔF/F for all component neurons. Fluorescence values from these ensembles were binned in 200 millisecond bins and entered into an online transform algorithm that related neural activity to the pitch of an auditory cursor. By modulating activity in these ensembles, rodents controlled the pitch of the cursor. The modulations that we required of the mice were calibrated daily based on a baseline recording session of roughly 2 minutes. Next, 10-15 minutes of spontaneous baseline activity was recorded to assess chance levels of performance and spontaneous levels of activity. Fluorescence values were smoothed by a moving average of the past 3 time points. Changes in the frequency of the auditory cursor were binned in quarter-octave intervals to match rodent psychophysical discrimination thresholds 23 . Mice then had to modulate calcium dynamics in these neuronal ensembles to move the cursor to a high-pitched target tone that was associated with a 10% sucrose solution reward. A trial was marked incorrect if a target was not achieved within 30 seconds of trial initiation. A trial was self-initiated when E1 and E2 activity returned to baseline levels (either by decreased activity in E1 or increased activity in E2), which reset the tone to its starting pitch. Regions of interest (ROIs) were extracted from recorded neural data in real time. These ROIs were entered into custom routines in MATLAB (Mathworks, Natick, MA) that translated fluorescence levels into the appropriate feedback pitch and played the pitch on speakers mounted on 2 sides of the imaging platform. Frequencies used for auditory feedback ranged from 1-24 kHz in quarter-octave increments. When a target was hit, a MATLAB-controlled Data Acquisition board (National Instruments, Austin, TX) triggered the operant box to supply the appropriate reward to rodents. Each daily training session lasted 48±2 min (71±4 trials).

Data Analysis

All analyses were performed with custom written routines in MATLAB (Mathworks, Natick, MA). Recorded movies were spatially aligned using the dftregistration routine in MATLAB 24 . Regions of interest were manually selected to include the soma of neurons that appeared consistently throughout all recorded movies. Fluorescence traces were extracted from each ROI and data is presented as the relative change in fluorescence, ΔF/F. No statistical methods were used to pre-determine sample sizes but our sample sizes are similar to those generally employed in the field. For analyses of behavioral performance during the contingency degradation and reinstatement, the first 10 trials of a session were removed before calculating performance. For all sliding window analyses, sessions were divided into an equal number of bins to determine the window size, and the step size was a fraction of this window size. For the cross-correlation histograms, fluorescence traces from output cells were z-scored and values above 3 standard deviations were considered an event. The first time point in which fluorescence values crossed this threshold during each event was used for time-locking. Fluorescence values in other populations of cells were then averaged around these indices. In all cases, multiple comparisons were controlled for using the Bonferroni correction. Differences between groups were tested with T-tests and trends over time were tested with T-tests on the linear regression coefficients. All statistical tests were two-tailed. Data distributions were assumed to be normal, but this was not formally tested. Data collection and analysis were not performed blind to the experimental conditions. Randomization was not performed, as the experiment primarily involved within-animal comparisons and there were not multiple experimental cohorts.

Supplementary Material Supp Supp Movie 1 Supp Movie 2