Fig 1.
Spike inference applied to VTA DA neurons in vitro.
a. Example data trace of simultaneous GCaMP6f (top) and raw cell-attached electrophysiology trace (bottom). b. Generative model of calcium dynamics. Simulated data (left) from this model (right) shows the underlying calcium ct (purple line) decays at rate γ until at time s there is a spike, so that zs>0 (red vertical lines below). GCaMP observations yt (black dots) are noisy realizations of the underlying calcium concentration. c. Example of decay estimation using multiple spike-free segments (“Decay segs.”, light blue). Observed GCaMP (black) decays at the same rate between spikes (black vertical lines below). The rate of exponential decay, γ, was estimated by fitting an exponential decay model to the observed GCaMP in the spike-free segments. d. Estimated decay rate per timestep at 66.67 Hz for GCaMP6f at 30°C (median = 0.987, Q1 = 0.983, Q3 = 0.988), GCaMP6f at 37°C (median = 0.975, Q1 = 0.971, Q3 = 0.980), and GCaMP6m at 37°C (median = 0.986, Q1 = 0.983, Q3 = 0.991). e. Decay time (half life, t1/2) for GCaMP6f at 30°C (median = 0.772 s, Q1 = 0.619 s, Q3 = 0.866 s), GCaMP6f at 37°C (median = 0.412 s, Q1 = 0.359 s, Q3 = 0.523 s), and GCaMP6m at 37°C (median = 0.725 s, Q1 = 0.614 s, Q3 = 1.100 s). f. van Rossum distance for each experimental condition and for each recording, using the median decay rate across all recordings for that type of GCaMP and temperature, as a function of the difference between the average inferred firing rate and average observed firing rate. The distance is minimized when the average firing rate between the inferred and observed spikes is similar. g. van Rossum distance for each experimental condition and for each recording as a function of the decay rate γ when the tuning parameter is selected so that the inferred firing rate matches the observed firing rate of the recording. The shaded region and vertical dark grey line represent the estimated interquartile ranges and median values of the decay rate in d. h. Example of correspondence between observed GCaMP (black line, top) and estimated calcium (blue line, top), and observed spikes (black vertical dashes, bottom) and inferred spikes (blue vertical dashes, bottom) in a single cell expressing GCaMP6f measured at 30°C. i. Example of correspondence between observed GCaMP (black line, top) and estimated calcium (green line, top), and observed spikes (black vertical dashes, bottom) and inferred spikes (green vertical dashes, bottom) in a single cell expressing GCaMP6f measured at 37°C. j. Example of correspondence between observed GCaMP (black line, top) and estimated calcium (orange line, top), and observed spikes (black vertical dashes, bottom) and inferred spikes (orange vertical dashes, bottom) in a single cell expressing GCaMP6m measured at 37°C.
Fig 2.
Spike inference applied to in vivo calcium imaging in VTA DA during presentation of unexpected reward generates firing rate modulations comparable to those previously reported in other animals via electrophysiology.
a. Top: Schematic of the surgical strategy, where GCaMP6 is expressed in VTA DA neurons and a GRIN lens is implanted for imaging. Bottom: Schematic of recording setup, where the mouse is headfixed and VTA DA neurons are recorded via 2-photon calcium imaging. b. Schematic of spike estimation approach as applied to the in vivo data, where the observed ΔF/F is used to generate estimated calcium and inferred spikes using the decay parameter γ from in vitro experiment and a λ selected to target a 6 Hz average estimated firing rate. c. Left: Schematic showing unexpected reward is delivered after random inter-trial intervals. Middle: Example of an unexpected reward trial from a single cell, showing observed GCaMP (orange) and inferred spikes (black vertical lines). Right: All unexpected reward trials from an example cell, showing observed GCaMP (top) and inferred spikes (bottom). d. Mean observed GCaMP from population around presentation of unexpected reward (n = 65 cells). e. Mean population firing rate from inferred spikes from population around presentation of unexpected reward (n = 65 cells). f. Mean population firing rate from spikes recorded via electrophysiology (from Eshel et al. [39]; n = 40 cells) around presentation of unexpected reward. g-j. Comparison of inferred and electrophysiology spikes from Eshel et al. [39]. g. Unexpected reward response, where reward response is the mean firing rate over the first 600 ms following reward presentation, baseline subtracted using the mean firing rate over a 1 s period before reward presentation (inferred spikes over baseline median = 9.8 Hz, Q1 = 6.2 Hz, Q3 = 12.2 Hz; electrophysiology spikes over baseline median = 9.5 Hz, Q1 = 6.3 Hz, Q3 = 12.0 Hz). h. Peak reward response amplitude in inferred and electrophysiology spikes, where peak is maximum value of PSTH in the first 600 ms period following reward presentation (inferred spikes median = 29.0 Hz, Q1 = 23.7 Hz, Q3 = 38.6 Hz; electrophysiology median = 30.5 Hz, Q1 = 22.6 Hz, Q3 = 40.9 Hz). i. Full duration at half max of reward response peak in inferred and electrophysiology spikes (inferred spikes median full duration at half max = 183.9 ms, Q1 = 163.6 ms, Q3 = 235.9 ms; electrophysiology median full duration at half max = 153.5 ms, Q1 = 120.4 ms, Q3 = 272.5 ms). j. Mean tonic firing rates prior to presentations of unexpected reward, where tonic firing rates are calculated as the mean firing rate over a 1 s period prior to reward presentation (inferred spikes tonic firing rate median = 5.7 Hz, Q1 = 5.3 Hz, Q3 = 6.0 Hz; electrophysiology tonic firing rate median = 5.6 Hz, Q1 = 4.3 Hz, Q3 = 7.5 Hz). Vertical bars are interquartile range (Q1 and Q3). All data is from cells expressing GCaMP6f.
Fig 3.
Spike inference applied to Pavlovian conditioning in vivo imaging data recapitulates transient cue responses, pauses following reward omission, and previously reported relationships between expected and unexpected reward responses.
a. Top left: Schematic showing expected reward is delivered after a 2 s cue presentation. Bottom left: Example expected reward trial from a single cell, showing observed GCaMP (purple) and inferred spikes (black vertical lines). Right: All expected reward trials from the example cell, showing observed GCaMP (top) and inferred spikes (bottom). b. Mean observed GCaMP from population around presentation of expected reward. c. Mean inferred spikes from population around presentation of expected reward. d. Full duration at half max of cue response in inferred spikes (median = 189.3 ms, Q1 = 168.3 ms, Q3 = 205.6 ms). Vertical bars are interquartile range (Q1 and Q3). e. Top left: Schematic showing unexpected omission of reward, where reward is omitted after a 2 s cue presentation that previously predicted reward. Bottom left: Example unexpected omission trial from a single cell, showing observed GCaMP (green) and inferred spikes (black vertical lines). Right: All unexpected omission trials from example cell, showing observed GCaMP (top) and inferred spikes (bottom). f. Mean population observed GCaMP around unexpected omission of reward. g. Mean population firing rate from inferred spikes around unexpected omission of reward. Shaded areas are SEM. h. Mean omission response of inferred spikes, where omission response is mean firing rate over 1,300 ms following onset of reward omission, baseline subtracted using the mean firing rate over 1 s period before trial start (median = -1.3 Hz, Q1 = -2.3 Hz, Q3 = -0.1 Hz). Neurons that exhibited a significant decrease in firing following reward omission (22/65 neurons; 33.9% of population) are darker green. Vertical bars are interquartile range (Q1 and Q3). i. Scatterplot of expected reward response versus unexpected reward response, using inferred firing rate of each neuron, recapitulates correlations in Eshel et al. [39]. j. Scatterplot of omission response versus unexpected reward response, using inferred firing rate of each neuron, recapitulates correlations in Eshel et al. [39]. Responses in i and j are baseline subtracted. All data is from cells expressing GCaMP6f.
Fig 4.
Upward and downward ramps in inferred spikes during reward approach in a virtual reality environment, in agreement with recent reports from electrophysiology.
a. Left: Schematic of neural recording and behavioral setup. The mouse navigates a virtual reality (VR) environment and 2p microscope records neural data. Right: Schematic of VR T-maze paradigm. The mouse navigates a linear maze and at the end must turn to the side that featured more cues in the Cue Region to receive reward. b. Example of single trial data from an upward-ramping cell. Both observed GCaMP (green) and inferred spikes ramp upward over time as the mouse moves down the maze (position trace, yellow). c. Example of a single trial from a downward-ramping cell. Both observed GCaMP (green) and inferred spikes ramp downward over time as the mouse moves down the maze (position trace, yellow). d. All trials from the example upward-ramping cell in (b) of observed GCaMP by position. e. All trials from the example upward-ramping cell in (b) showing inferred spikes by position. f. Mean observed GCaMP by position for example upward-ramping cells. g. Mean inferred firing rate by position for example upward-ramping cells. h. All trials from the example downward-ramping cell in (c) of observed GCaMP by position. Heatmap color scales constrained to data between 1st and 99th percentile. i. All trials from the example downward-ramping cell in (c) showing inferred spikes by position. j. Mean observed GCaMP by position for example downward-ramping cells. k. Mean inferred firing rate by position for example downward-ramping cells. Shaded areas are SEM. l. Scatterplot showing how change in observed GCaMP from beginning to end of maze for each neuron relates to change in inferred firing rate for that neuron. Each data point represents a single neuron and its mean change in observed GCaMP and inferred firing rate. Red line is linear least-squares fit; shaded region is 95% confidence of the fit. m. Distribution of inferred spike slopes over time among neurons with significant position modulation with negative (red; mean slope = -0.12 spikes/s2; n = 72/303 neurons) or positive (blue; mean slope = 0.18 spikes/s2; n = 112/303 neurons) ramps, or with no significant position modulation (grey; mean slope = 0.02 spikes/s2; n = 119/303 neurons). Significant position modulation determined by a generalized linear model (GLM), where the inferred spikes were predicted by mouse position, with a factor for individual trials, where neurons were classified as significantly ramping if the coefficient associated with position was statistically significant at level alpha = 0.01. Data is from a mix of cells expressing either GCaMP6f or GCaMP6m.