Interspersed Distribution of Selectivity to Kinematic Stimulus Features in Supragranular Layers of Mouse Barrel Cortex

Neurons in the primary sensory regions of neocortex have heterogeneous response properties. The spatial arrangement of neurons with particular response properties is a key aspect of population representations and can shed light on how local circuits are wired. Here, we investigated how neurons with sensitivity to different kinematic features of whisker stimuli are distributed across local circuits in supragranular layers of the barrel cortex. Using 2-photon calcium population imaging in anesthetized mice, we found that nearby neurons represent diverse kinematic features, providing a rich population representation at the local scale. Neurons interspersed in space therefore responded differently to a common stimulus kinematic feature. Conversely, neurons with similar feature selectivity were located no closer to each other than predicted by a random distribution null hypothesis. This ﬁ nding relied on de ﬁ ning a null hypothesis that was speci ﬁ c for testing the spatial distribution of tuning across neurons. We also measured how neurons sensitive to speci ﬁ c features were distributed relative to barrel boundaries, and found no systematic organization. Our results are compatible with randomly distributed selectivity to kinematic features, with no systematic ordering superimposed upon the whisker map.


Introduction
In the primary somatosensory "barrel" cortex, neurons responsive to the same whisker play different roles in representing the identity and location of objects contacted by the whisker (von Heimendahl et al. 2007;Jadhav et al. 2009;O'Connor et al. 2010;Petreanu et al. 2012;Safaai et al. 2013;Chen et al. 2013a;Clancy et al. 2015;Peron et al. 2015;Sofroniew et al. 2015).Barrel cortex neurons are tuned to diverse stimulus properties (Hires et al. 2012;Petreanu et al. 2012;Yamashita et al. 2013;Chen et al. 2013a;Sofroniew et al. 2015).Specifically, the dynamical or temporal stimulus features to which neurons are sensitive vary from cell to cell, providing a rich representation of stimulus dynamics at the population level (Estebanez et al. 2012).How are neurons with heterogeneous response properties distributed in space?Barrel cortex neurons with different selectivity to spatial stimulus characteristics (strength of tuning to principal whisker, tuning to direction of whisker deflection, or correlated motion of multiple whiskers) are found within each barrel column (Andermann and Moore 2006;Kerr et al. 2007;Sato et al. 2007;Kremer et al. 2011;Clancy et al. 2015;Estebanez et al. 2016).
A recent study of selectivity to different textures found that neurons preferring the same texture tend to cluster together across rat barrel cortex (Garion et al. 2014).Here, we examined the spatial distribution of neuronal selectivity to kinematic features of whisker motion using 2-photon calcium population imaging in mice, seeking to uncover systematic ordering principles.

Animal Preparation
All procedures complied with Society for Neuroscience, European, Spanish, and institutional policies for the care and use of animals in research.The protocol was approved by the local (Instituto de Neurociencias) bioethics and biosafety committee and by the institutional (CSIC) bioethics subcommittee.All experiments were performed under ketamine-xylazine anesthesia, and every effort was made to minimize suffering.
Female mice (CD1) at postnatal day ~30 were anesthetized using ketamine-xylazine (120 and 16 mg/kg body weight).The animal skull was exposed and cleaned and a metal plate attached with dental acrylic cement.A small craniotomy (~2 mm diameter) was made above barrel cortex.The location of the craniotomy was determined stereotactically (1.5 mm from Bregma, 3.4 mm from midline).Post hoc cytochromeoxidase staining of tangential slices (see below) confirmed that the acquisition fields fell within the area corresponding to the caudal edge of the whisker pad.The exposed dura was covered with agarose.Eyelid and hindpaw reflexes were monitored throughout the experiments, and refresher ketamine-xylazine doses (20% of initial) added if necessary.

Stimulus Design
Although population calcium imaging permits analysis of tuning properties of neurons in terms of their spatial relationships, because of limitations in effective temporal resolution it is not ideal for an unbiased reverse correlation computation of receptive fields (Sharpee 2013).Our strategy was to construct a stimulus set that would capture features that neurons might selectively respond to (Jones et al. 2004;Arabzadeh et al. 2005;Petersen et al. 2008;Jadhav et al. 2009;Estebanez et al. 2012), while remaining small enough in size to permit adequate sampling during the course of an experiment.The set included idealized "position," "velocity," and "acceleration" filters, constructed by convolving the corresponding impulse functions with a Gaussian filter (Fig. 1B).Each waveform type was presented with 3 possible amplitudes, in the ratio 1:3:5.Stimuli were prepared in Matlab (Mathworks).

Whisker Stimulation
We inserted 10-15 contralateral vibrissae into a glass capillary tube glued to a piezoelectric bender (Physik Instrumente), placed 2-3 mm from the skin.This form of stimulation selected for neurons sensitive to correlated whisker motion, which are readily found in the barrel cortex and, in layer 2/3, are more numerous than those sensitive to uncorrelated motion (Estebanez et al. 2012(Estebanez et al. , 2016)).Since our aim was to examine whether different neurons have similar or diverse feature selectivity when interrogated with a common stimulus, motion of the piezoelectric actuator in the main set of experiments was always in the rostro-caudal direction.
Maximum deflection amplitude was 400 μm.Each deflection achieved a maximum speed of approximately 400 mm/s; speed only briefly remained close to maximum (median speed ~40 mm/s; for other parameters, see Pitas et al. 2016).These values are at the higher end of those used for passive stimulation in previous studies or recorded during free whisking in air (Kwegyir-Afful et al. 2008;Khatri et al. 2009), but in the range reached during natural whisker motion (Bagdasarian et al. 2013).
In a recording, each stimulus was presented 10 times; 4 recordings were done for each field of view, for a total of 40 repetitions per field.Consecutive deflections were separated by 1 s, an interval long enough to distinguish calcium transients (checked using electrophysiology; data not shown).Mechanical stimulation artifacts (resonances) were possible given the frequency range necessary to reproduce the deflection waveforms (~100 Hz).To rule them out, we used a custom-built optoelectronic device to check that the mechanical waveform described by the piezoelectric bender reliably followed the electrical input (Fig. 1B).

2-Photon Calcium Imaging
Calcium indicator in acetoxymethyl ester form (OGB1-AM, Life Technologies) was prepared by dissolving 50 μg of dye in 4 μL of 20% pluronic acid in DMSO (Life Technologies) and diluting (1:11) in artificial cerebrospinal fluid containing 100 μM Alexa Fluor 594 (for visualization; Life Technologies).Patch pipettes (tip diameter 2-4 μm) were pulled (Narishige), filled with dye solution and introduced into the cortex.The pipette tip was visualized under 2-photon scanning mode and gradually advanced to 100-150 μm below the cortical surface.Dye was pressure injected at 2-10 psi (Picospritzer) over 5-10 min.After injection, the pipette was withdrawn and the craniotomy covered with warm agarose (Sigma) and coverslipped.Imaging of whisker-evoked Ca 2+ transients was performed using a 2photon microscope (Leica TCS SP5 MP) with a Spectra Physics Mai Tai HP laser.Excitation wavelength was 830 nm.Cells were imaged using a 25× water-immersion objective (NA 0.95, Leica) at depths of 100-350 μm from the cortical surface.Full-frame images of 512 by 128 pixels were acquired at a spatial resolution of 1.48 pixel/μm; frame sampling rate was 8.8 Hz.

2-Photon Image Analysis
Images were registered with a modified StackReg ImageJ (NIH) plugin.For each movie, rigid transformation was performed using the first frame as reference.Experiments with excessive x-, y-, or z-axis fluctuations were discarded.Calcium responses were extracted using Caltracer 2.5, a Matlab (Mathworks) software package (Rafael Yuste Lab, Columbia University).Briefly, cells were visualized using the mean image of all frames and cell contours outlined automatically to define neuronal regions of interest (ROIs).Detected ROIs were supervised and adjusted manually if necessary.The mean raw fluorescence of each cell was estimated for all frames and background from unstained blood vessels subtracted (Greenberg et al. 2008).A perisomatic halo was drawn automatically in order to correct for neuropil contamination: halo fluorescence was scaled (optimal factor = 0.7) and subtracted from the background-subtracted mean fluorescence (Fig. 1A,C) (Kerlin et al. 2010;Chen et al. 2013b;Feinberg and Meister 2015).Finally, corrected fluorescence values were converted into ΔF/F 0 .Typically, responsive neurons had a skewed raw fluorescence distribution.F 0 was set to the eighth percentile of corrected fluorescence within a symmetric 6 s sliding window.Response amplitude was calculated as the difference between the mean ΔF/F 0 value of the first 5 frames after a stimulus and the mean value of the last 3 frames before the stimulus.

Measurements of Response Properties
To assess whether to score neurons as tuned to kinematic features, we first computed the linear regression between magnitude of all calcium responses and amplitude of the corresponding stimulus waveforms for all neurons in a field of view.To correct for false discoveries of significantly tuned neurons in the simultaneously recorded population, we then applied the Benjamini-Hochberg-Yekutieli procedure for controlling the false discovery rate (corrected significance level: P < 0.05; fdr_bh Matlab function written by David Groppe) (Benjamini and Hochberg 1995;Benjamini and Yekutieli 2001).For neurons scored as tuned according to this procedure, and whose responses thus changed as a function of one or more stimulus parameters, we then defined tuning strengths for position, velocity, and acceleration as the respective coefficients of regression between response magnitude and stimulus waveform amplitude.Coefficients that did not reach the significance limit were set to zero.Where a neuron was significantly tuned to several features, we scored it as having mixed selectivity.Linear correlation and regression do not take into account the nonlinearities inherent to neuronal tuning and the conversion of a sensory stimulus into a change in calcium-dependent fluorescence (Peron et al. 2015).However, our approach provided a principled way to compare between the strength of responsiveness to different stimulus parameters defined over a similar time course.We also assessed tuning significance via direct shuffling of responses with respect to stimulus magnitude for each neuron (1000 repeats), with no qualitative change in results.Finally, we tested neurons for consistency of tuning properties across stimulus repetitions.Each ROI was recorded over 4 repeats of the stimulus set (see above), and the results given in the figures correspond to data collected over the entire set of 4 repeats.To check for consistency, data were split into 2 subsets consisting of the first and last pairs of recordings.95% of neurons displayed identical tuning to at least one feature over the 2 subsets, and 81% displayed identical tuning to all 3 features, with no qualitative change in overall conclusions arising from splitting the data into subsets.
To measure the extent to which pairs of neurons in a field of view were similarly tuned we calculated the similarity index (SI).First, for each neuron in the field of view we constructed a tuning vector whose components were the cell's tuning strength (regression coefficient) for position, velocity, and acceleration (with tuning strengths not statistically significant entered as zero).Next, we computed the SI for each pair of neurons by taking the dot product of the 2 neurons' tuning vectors.
The slope of the linear fit between SI and the distance separating each pair of neurons was compared against that for shuffled values.Shuffled data were generated by randomly reassigning feature selectivity (regression coefficients) across neurons in the field of view, calculating the slope of the linear fit and then obtaining a mean slope for the field of view; shuffling was repeated 100 times.Pairwise correlations were calculated using the Spearman correlation coefficient computed for the entire corrected calcium fluorescence time series of the 2 cells, over the duration of the concatenated 4 recordings.We analyzed the dependence of neuronal correlations on distance by computing the slope of the linear fit between the 2 variables.Shuffled data for each experiment were constructed by randomly permuting the temporal order of responses to different stimulus presentations for each neuron in the field of view.

Histology and Barrel Field Reconstruction
In some experiments, brains were removed and fixed in 4% paraformaldehyde after imaging.The cortex was cut tangentially in 150 μm vibratome sections and stained for cytochrome oxidase.The barrel field was reconstructed and the acquired field localized in the reconstructed tissue.To place acquired fields in their original position, the pial vasculature of a fixed brightfield image was aligned using Photoshop (Adobe) with that from an in vivo bright-field picture of the same cortical surface.Then, using a single image of the cortical surface taken before starting the 2-photon acquisition, the fluorescent somata of layer 2/3 neurons in the field were aligned with the bright-field images.Distances to the closest barrel border were measured manually in Canvas (ACD Systems).

Results
To measure tuning to kinematic stimulus features, we imaged the activity of layer 2/3 neurons in ketamine-xylazine anesthetized mice (n = 10) while stimulating whiskers with controlled patterns (Fig. 1).To generate the stimulus set, we exploited the fact that neurons in the whisker pathway are tuned to kinematic stimulus features such as velocity or acceleration (Jones et al. 2004;Arabzadeh et al. 2005;Petersen et al. 2008;Jadhav et al. 2009;Estebanez et al. 2012).We created a set of whisker deflection waveforms consisting of idealized "position," "velocity," and "acceleration" filters (Fig. 1B), each with 3 possible amplitudes.We reasoned that a neuron selectively sensitive to one of these features would exhibit responses whose size would covary preferentially with the amplitude of the corresponding waveform type.Different deflection waveforms and amplitudes were randomly interleaved in time.This stimulation produced clear calcium responses that allowed us to determine differential tuning of individual neurons to particular stimulus features (Fig. 1C,D).
In general, we considered that a neuron was tuned to a specific feature (position, velocity, acceleration) when the size of the calcium response displayed a significant relationship with the amplitude of the corresponding stimulus waveform (see Materials and Methods).Traces from an example neuron can be seen in Fig. 2A.For this neuron, calcium responses clearly grew with the amplitude of the velocity waveform, with increases in acceleration or position waveforms having a smaller effect (Fig. 2A,B).Reflecting this, the value of the regression coefficient between stimulus amplitude and neuronal response was greatest for velocity [r p = 0.0069 for position (5-95% CI 0.0032-0.011);r v = 0.0169 for velocity (CI 0.0132-0.021);r a = 0.0106 for acceleration (CI 0.0073-0.0151);Fig. 2B].
Where a neuron was significantly tuned to several features, we labeled it as having mixed selectivity (see Materials and Methods).Neurons throughout the whisker pathway, particularly in the barrel cortex, do not act as pure encoders of a single stimulus physical parameter or dimension; rather, their preferred features tile a space defined by multiple dimensions (Maravall et al. 2007;Petersen et al. 2008;Estebanez et al. 2012;Bale et al. 2013;Chagas et al. 2013;Maravall et al. 2013;Campagner et al. 2016).Thus, a characterization that allows for mixed selectivity better captures biological diversity.To display the tuning of each neuron visually, we used an RGB color map (Fig. 2C).We translated significant tuning strengths to color intensity levels by representing tuning to position as a red color intensity value, to velocity as green intensity, and to acceleration as blue intensity.The outcome for each neuron was a tone that mixed the appropriate intensities of red, green, and blue, reflecting tuning strength to position, velocity, and acceleration, respectively.Neurons with no significant tuning were depicted as outlines; neurons with significant tuning to a single feature were pure red, green, or blue (Fig. 2C,F; Supplementary Figs. 1 and 2 show identical data depicted in separate panels for the red, green, and blue channels).As the example neuron was tuned to velocity and (more weakly) to acceleration, it appears as bluish green in this representation (Fig. 2C, black arrow).Overall, 50.2% of neurons in the data set responded to whisker stimulation.The analysis evidenced both tuned and non-tuned neurons, with an overall majority of non-tuned cells (71.8% of all neurons were non-tuned; n = 10 mice, 49 fields of view, 1054 neurons; Fig. 2D,E).Within tuned neurons, all categories of selectivity were represented (Fig. 2E).
Neighboring neurons in primary sensory cortices share synaptic input (Harris and Mrsic-Flogel 2013).To assess whether this is reflected in similarities in the behavior of neurons at the population level, for each acquired field of view we computed pairwise correlations across all neuron pairs over the full duration of the recorded calcium time series (i.e.including whisker stimulation).An example raster plot capturing the activity of neurons in a field of view suggests that correlations across neurons in that region were higher than expected by chance (Fig. 3A).This is borne out by the distribution of correlation coefficients for that field of view (Fig. 3B); mean correlation coefficient was 0.14 ± 0.0062 compared with 0.0022 ± 0.0013 obtained by shuffling responses of different neurons across stimulus presentations (n = 31 neurons; see Materials and Methods).Taking all experiments into account, the correlation coefficient for the data set was 0.1004 ± 0.008 (mean ± SEM across fields of view), compared with 0.011 ± 0.0024 for shuffled data (n = 49 fields of view including n = 1054 neurons; P = 1.11 × 10 −9 ; Wilcoxon signed-rank test; Fig. 3C).We also compared neuronal and neuropil response correlations by choosing ROIs within the neuropil in each field of view (10 ROIs per field) and repeating the analysis above.Neuropil responses were more correlated with each other than neuronal responses (for neuropil, mean ± SEM coefficient was 0.1792 ± 0.0097; n = 49 fields including n = 490 ROIs; vs. neurons, P = 2.13 × 10 −7 ; Wilcoxon signed-rank test).
Previous publications in barrel cortex have shown correlated spontaneous and evoked activity between neurons, falling-off over distances ~160-200 μm (Kerr et al. 2007;Sato et al. 2007;  Clancy et al. 2015).We examined the effect of distance on correlations over up to 300 μm.For the neurons in the raster plot of Fig. 3A, correlations decreased with distance, with a mean spatial gradient of −0.15 mm −1 (0.0025 mm −1 for shuffled data; Fig. 3D).Most of the fall-off in correlation occurred within the first 100 μm of distance between neurons.For the complete data set, the spatial gradient was −0.29 ± 0.062 mm −1 (median ± standard error of median [SEM]), significantly different than for shuffled data, −0.0062 ± 0.014 mm −1 (n = 49 fields of view; P = 1.31 × 10 −8 ; Wilcoxon signed-rank test; Fig. 3E).Yet for experiments with pairs of neurons separated by over 200 μm, the spatial gradient decreased to −0.13 ± 0.17 mm −1 , compared with −0.109 ± 0.085 mm −1 for shuffled data (median, SEM; n = 30 fields of view with >4 pairs extending >200 μm; P = 0.106; Wilcoxon signed-rank test).Therefore, while nearby neurons showed correlated activity during stimulation, this correlation decreased with distance over a range ~200 μm, comparable to the diameter of a barrel column.
Because correlations in neuronal activity during stimulation were higher in nearby neurons, probably because of partially shared synaptic input, we wondered whether functional tuning would evidence spatial organization within comparable distances (~200 μm).To test this, we analyzed the similarity of response tuning as a function of distance between cells.We created an index (SI) to measure the extent to which the tuning of a pair of neurons was alike (Materials and Methods).We then plotted SI as a function of distance between the neurons in the pair (Fig. 3F).In some experiments, there appeared to be a relationship between SI and distance (for the example in Fig. 3F, spatial gradient = −1.29 mm −1 ; compared with shuffled, 0.133 mm −1 ; n = 29 neurons, n = 465 pairs).This was consistent with a visually apparent tendency for neurons to cluster in some of the fields of view (e.g.nearby green neurons in Fig. 2F, left panel).However, pooling across experiments indicated no consistent spatial organization of SI (median slope −0.467 ± 0.000675 mm −1 ; for shuffled data, −0.0169 ± 0.119 mm −1 ; n = 26 fields of view; P = 0.114; Wilcoxon signed-rank test; Fig. 3G).These results confirm a visual intuition from Fig. 2F: Functional responses to kinematic features are heterogeneous on a local scale, with neurons tuned to different features interspersed within the circuit.
Finally, we wondered if tuned neurons were systematically arranged relative to the barrel structure, for example, by asymmetric distribution between barrel-and septum-related territories.Thus, we examined the localization of tuned neurons relative to the histologically reconstructed barrel field map (n = 3 mice; respectively, n = 167, 102, and 117 imaged neurons, of which n = 59, 39, and 47 were tuned; imaged at depths ranging between respectively;Fig. 4A,B;Supplementary Fig. 3 shows identical data in separate panels for the red, green, and blue channels).Tuned neurons were evenly distributed between barrel-and septum-related areas (38.8% and 35.8%, respectively were tuned; P = 0.58; Fisher exact test; Fig. 4C).We also tested whether neurons tuned to specific features were located preferentially relative to barrel or septum areas.To this end, we measured the horizontal distance of each significantly tuned neuron to the nearest barrel border.Neurons tuned to different features were not located differently relative to barrel borders, that is, relative to the columnar structure of barrel cortex (position vs. velocity: P = 0.77; position vs. acceleration: P = 0.70; velocity vs. acceleration: P = 0.68; Kolmogorov-Smirnov 2-sample test; Fig. 4D).Thus, we did not find evidence for spatial organization of kinematic feature tuning with respect to the barrel map in layer 2/3 of mouse barrel cortex.

Discussion
Neurons in the barrel cortex have strikingly heterogeneous response properties (Maravall and Diamond 2014).In layers 2/3 of the barrel field, neurons represent a wide range of spatial and temporal properties of whisker motion (Andermann and Arrows point to values for the color-coded connected pairs in the inset panel.Linear fit for true (black solid line) and shuffled data (dashed line).(G) Slope of the linear fits between SI and interneuronal distance for true and shuffled data for all fields; average in black.
Moore 2006; Kerr et al. 2007;Sato et al. 2007;Jacob et al. 2008;Kremer et al. 2011;Estebanez et al. 2012;Garion et al. 2014;Peron et al. 2015;Estebanez et al. 2016).Here, we explored how neurons with sensitivity to different kinematic features are arranged.We found no ordering principle for feature selectivity-no systematically mapped representation.Instead, tuning follows an intermingled "salt-and-pepper" arrangement, with different neurons encoding for different features.This is reminiscent of the organization of orientation selectivity in rodent visual cortex (Ohki et al. 2005;Mrsic-Flogel et al. 2007).The overall correlation between responses of neuron pairs did fall-off with distance.This apparently counterintuitive difference between the ordered dependence of correlations on distance and the disordered distribution of feature selectivity is again similar to mouse visual cortex (Denman and Contreras 2014;Montijn et al. 2014).Diversity across differentially tuned subnetworks of neurons appears to be a common principle governing the local connectivity of primary sensory cortical areas in rodents (Bandyopadhyay et al. 2010;Rothschild et al. 2010).
We chose a simple stimulus set that exploits known feature selectivity in barrel cortex neurons, to provide adequate sampling of neuronal responses (Fig. 1B).These stimuli covered a limited region of the possible space to which neurons can be responsive (Jacob et al. 2008;Estebanez et al. 2012;Maravall and Diamond 2014).For example, neurons can be selectively driven by the amount of correlated motion across multiple whiskers; furthermore, layer 2/3 neurons are systematically arranged in relation to barrels and septa according to their sensitivity to correlated motion (Estebanez et al. 2016).Thus, a more extensive stimulation protocol could have identified feature selectivity in additional neurons.Our aim in using the present design was to bring out potential differences in selectivity across neighboring neurons, rather than to identify all possible stimulus features evoking responses.
Our findings differ from a recent study that found spatial clustering of barrel cortex neurons tuned to the same texture (Garion et al. 2014).Several factors may have contributed to this difference.Important experimental differences include species (rat in the earlier study vs. mouse here), anesthesia (urethane vs. ketamine-xylazine) and form of stimulation ("electrical whisking" vs. passive).The former study also examined the build-up of responses during repeated stimulation, while we focused on tuning of responses temporally locked to a single, brief stimulus.In addition, two critical differences lie in the analysis.First, to build maps in the previous study, neurons preferring a particular texture, but potentially responding to others as well, were assigned that texture only.Instead, we chose to allow for mixed selectivity, since sensitivity to diverse, intermediate kinematic features is a hallmark of neurons in the whisker pathway (Petersen et al. 2008;Estebanez et al. 2012;Bale et al. 2013;Maravall et al. 2013).Second, the earlier study compared the true distribution of distances between neurons to a null hypothesis constructed by randomly and uniformly distributing an identical number of neurons across the entire field of view.This comparison did not provide a specific test for clustering of feature selectivity: A positive clustering result could arise simply from the imaged cell bodies being closer than expected had they been randomly scattered across the entire field of view.In contrast, our null hypothesis shuffled feature selectivity across neurons sited at their true locations, an approach designed to test for clustering specifically.

Figure 1 .
Figure 1.(A) Top, average projection of a movie showing cell bodies loaded with OGB.Bottom, same projection with regions of interest (blue) and neuropil halos (purple).Scale bar, 25 μm.(B) Waveforms of stimuli constructed as ideal position, velocity, and acceleration filters.Left, electrical input delivered to the piezoelectric actuator.Center, mechanical waveform recorded with an optical sensor.Right, superposition of input and mechanical output.(C) OGB fluorescence (blue) from 2 example neurons with corresponding perisomatic neuropil signal (halo, in purple) and corrected trace (black).Bottom trace, stimulus times.(D) Five examples of single-sweep calcium responses to different kinematic features in an example neuron.Each feature (P: position, V: velocity, A: acceleration) was presented in 3 different amplitudes (P1, P2, and P3 for position, for example).Different waveforms and amplitudes were in randomly interspersed order.

Figure 2 .
Figure 2. (A) All (gray) and average (colored thick trace) calcium responses to position (top), velocity (middle), and acceleration (bottom) stimuli in an example neuron.Arrowheads mark stimulus onset.(B) Plot of the median calcium responses and filter amplitudes for position (red circles), velocity (green triangles), and acceleration (blue squares) stimuli, for the same neuron.Stimulus amplitude is normalized to the smallest value.Tuning is strongest to velocity, as reflected in the steepest slope.(C) Mask of cell bodies used to quantify calcium changes from movies in an example field of view.Cell bodies are color-coded according to feature tuning (mixed selectivity allowed).The neuron in A-B is marked with a black arrow and its color depicted as a sum of red, green, and blue intensities proportional to the strength of its tuning to position, velocity, and acceleration, respectively.Outlined cell bodies do not respond or have no significant tuning.Scale bar, 25 μm.(D) Number of neurons with and without significant tuning to stimulus features in each field acquired.Thick black line, average.(E) Mean percentage of neurons with significant tuning to a single feature (Pos, Vel, Acc), mixed features or none.(F) Two further examples of the local distribution of kinematic feature selectivity within fields of view.Scale bars, 25 μm.

Figure 3 .
Figure 3. (A) Raster plot of calcium response activity for all neurons in an example field during a 90 s period of whisker stimulation.Feature tuning selectivity for each neuron is indicated in color code bar on right.(B) Distribution of correlation coefficients for pairs of neurons in the example field in A (gray).Shuffled data are shown in black.Average represented by arrowheads on top.(C) Mean correlation coefficient for each field acquired and for corresponding shuffled data; average in black.(D) Correlation coefficients as a function of distance between neurons for the example field in A. Linear fit for true (black solid line) and shuffled data (dashed line).(E) Slope of the linear fit between correlation coefficients and interneuronal distance for true and shuffled data for all fields; average in black.(F) Tuning SI as a function of distance between pairs of neurons for the field of view in A. Each dot corresponds to one pair of neurons.