Definition of cortical area A and redefinition of PPC based on visual and somatosensory responses.

A. Elevation (left) and azimuth (right) maps showing retinotopy that form the borders of V1 (primary visual cortex) and AM. The cortical surface was color-coded according to the location of the bar on the monitor at the time of the evoked visual response. B. Tactile stimuli of a tail (left, red) and a trunk (right, green) resulted in clear spots in S1. C. The center positions of S1 tail and S1 trunk regions relative to the center position of AM from 8 mice. The gap space between the S1 trunk region and AM was defined as A. D. A location of the cranial window and FOV of two-photon imaging was overlaid on the Allen CCF version 3 33. E. The average image of calcium imaging and ROI positions for panel F. F. Fluorescence traces (mean and s.e.m.) of ROI indicated in panel E when the mouse received visual stimulation (azimuth) and tactile stimuli (tail, trunk, and whisker). G,H. Single neuron resolution mapping of somatotopy (G) and retinotopy (H) revealed that A had neither retinotopy nor somatotopy, but had neurons activated by tactile stimuli and a small number of neurons activated by visual stimulation. On the other hand, AM had retinotopy and a small number of neurons activated by tactile stimuli. A was defined after the S1 and AM were determined by somatotopic and retinotopic organizations, respectively. Scale bar: 1mm. I. Redefined PPC in this study (A, AM and RL).

Vision and choice information are merged in AM.

A. Schematic illustration of a vision-guided choice task. B. The lick-ports were approached after a 4-s sampling period (SP). The mouse was given a reward (water) or a punishment (air-puff) based on the licking direction during response period (RP). ITI, Inter-trial-interval. C. Rules of vision and history task. D. Four example raw traces (black) and inferred spikes (green) during the task. E. Representative vision neurons (ROI 1-4 in I). F. Representative choice neuron (ROI 5-8 in i) and a non-selective neuron (ROI 9). G. Schematic of our encoding model. H. t-SNE result of single neuron activities (left). Task-related neurons were mapped onto the t-SNE (right). I. All active neurons in the vision task (left) and the t-values of video (red), blank (orange), left choice (cyan), and right choice (purple) are overlaid (middle). The ratio of vision and choice neurons are also shown (right). J. Map of Association index (harmonic mean of vision and choice neurons.) K. Decoding accuracy of choice by activity of 50 neurons from eight areas before and after the choice were plotted (left). The decoding accuracies 1s before (right bottom, gray shade; p=0.13; one-way ANOVA) and just after (right top, blue shade; p=3.6 × 10- 9; one-way ANOVA) the choice was shown. L. The activity of vision neurons was averaged and aligned by the order. M. The time after the onset of the video was precisely decoded with a population of vision neurons in panel L. N. Decoding accuracy of time during video presentation (top; p=1.1 × 10-15; one-way ANOVA) and black screen presentation (bottom; p=0.013; one-way ANOVA) were plotted. O. Decoding accuracies of time in video presentation and choice direction indicate that AM would be the best position for associating these two signals. Error bars, mean ± s.e.m. in E and F, 95% confidence interval in K and N. Scale bar: 1mm.

Robust representation of choice history in A.

A. The representative history neurons. Numbers correspond to that of panel B and C. B. The t-SNE result for the single neuron activity (left). The right panel shows that each cluster of t-SNE has the distinct functional types defined by the encoding model. C. Position of the single neurons imaged in history task (left) and the percentages of history, choice, and vision neurons. D. The percentage of history neuron and the association index were overlaid. Note that these two roughly corresponded to A and AM, respectively. E. The percentage of history neurons in A was higher in the history task than in the vision task (top; p=4.1 × 10-4; chi-squared test). The percentage of vision neurons in A was not different (bottom p=0.61; chi-squared test). F. Decoding accuracy of choice, outcome, and visual stimuli by the activity of 20 neurons from each area before and after the choice onset, reward delivery, and the end of the visual stimuli, respectively. Line colors corresponded to the areas shown in panel G. G. Decoding accuracy of the choice (p=2.2 × 10-13; one-way ANOVA), outcome (p=0.13; one-way ANOVA), and visual stimuli (p=0.094; one-way ANOVA) at the shaded time in panel F were compared for each area. h. Decoding accuracy of the previous choice (thick lines) and the next choice (thin lines) as a function of time relative to the next choice. Note that both A and S1t had larger information on the previous choice until 1 s before starting the next choice than the next choice. i. The decoding accuracy of the next choice from A activity was significantly larger when the next choice was correct than when the next choice was incorrect (Wilcoxon signed-rank test with Bonferroni correction). J. The activity of simultaneously imaged history neurons was aligned by their peak time. The continuous sequential activity can be observed in the two consecutive trials. K. jPCA plot of the history neurons shown in panel J (right), that of A neurons (middle), and that of AM neurons (right). L. The R2 of Mskew indicating how rotational dynamics was dominant was compared (p=5.1 × 10-33; one-way ANOVA). M. The R2 of Mskew was significantly positively correlated in A and S1t but not in AM (*** p<10-6; ** p<0.01; t-test). N. The decoded rate of time after the choice using neurons in A (top) and AM (bottom). Note that cross-choice mistakes (*) were rarer in A than in AM. O. Decoding accuracy of time with the correct choice (top; p=5.6 × 10-14; one-way ANOVA) and decoding rate of time with an incorrect choice (bottom; p=8.5 × 10-10; one-way ANOVA). Error bars, mean ± s.e.m. in i, 95% confidence interval in G. M, and O. Scale bar: 1mm.

Posture related neurons and robust representation of the history neurons.

A. Snapshots from the cameras. The green, magenta, and orange circles indicate that the estimated positions of the left forepaw, the right forepaw, and the tail, respectively, when each part crossed the edge of the body holder. Each line was drawn between each estimated body part and the center of the body holder. Blue circles indicate the edge of the body holder. B. Schematic of the regression model for estimation of posture/movement related activity. C. Pseudo-R2 values were mapped for the left forepaw, right forepaw, the tail and the pupil. D. The body holder was randomly rotated during the history task to prevent the mice from keeping choice history as posture. E. Representative holder angle-related neurons. Numbers correspond to their location in the left panel. Neuron 1 was holder-related but not to reward direction, whereas neuron 2 was related to both the holder and the reward. F. The t-values of history neurons and holder-related neurons were mapped. G. Decoding accuracy for the holder direction and the choice history in each area was compared. Note that the A and S1t had precise information for both. H. The t-values for the history-related activity and for the holder angle in each history neuron were plotted. No significant relationship was observed. I. The percentage of the history neurons in A (green) and AM (magenta) was not different between the history task and the task with random holder rotation (p=0.15 in A; p=0.38 in AM; chi-squared test). J. The decoding accuracy of choice history in A (green) and AM (magenta) was not significantly different regardless of the holder movements (p=0.87 in A; p=0.95 in AM; Wilcoxon’s rank sum test). Scale bar: 1mm.

Mesoscale correlation of PPC and surrounding areas.

A. Average of trial-to-trial correlation (rt_single) between a PPC neuron and a neuron in the other areas was plotted against the distance between the neuron pair. The correlation was small but significantly larger than zero regardless of the distances. B. Log-scaled histogram of the rt_single. The shaded area indicates a large positive correlation (> 0.3). C. The coupling neurons whose rt_single with seed neurons (large circle) in A (left) and AM (right) was larger than 0.3 were shown with the same color as the seed neuron. Most neurons coupling to the A neurons were in the S1t, while the neurons coupling with the AM neurons were spreading but the spreading area was specific to each seed neuron. D. Schematic of the K-means (k=6) clustering of neurons in A based on the ratio vector, R. This vector, given for each neuron, consists of the ratio of the number of coupling neurons (rt_single >0.3) in the seven areas other than A. In this case, for example, RiA, the vector of the i-th neuron in A, had the largest value of 0.5 in the fifth element, which indicates this neuron had half the number of coupling neurons in the RSP. K-means clustering of the ratio vectors can identify the clusters of A neurons sharing the distribution of the highly correlated neurons. The analysis was done for all neurons in AM instead of A as well. E. Polar plots of the preference vectors of A (left) and AM (right). The preference vector is mean ratio vector of each cluster. F. Angles between any pair of the preference vectors which were obtained 1,000 times with randomly selected neurons. G. The relationship between the preferred area of single neurons in A or AM and their task-relatedness was compared. For example, significantly large number of A neurons preferentially coupling with the S1t neurons were choice neurons or history neurons whereas those coupling with the AM neurons were vision neurons. H. CCt between pairs of eight areas during vision task and history task was highly correlated (r=0.95). I. The CCt between an area to the other areas were compared. A and AM had the two largest CCt (top; p=1.7 ×10-6; one-way ANOVA), which was not the case when coupling neurons were eliminated (bottom; p=6.7 × 10-7; one-way ANOVA). J. CCt as a function of its dimension. CCt from AM were uniform regardless of the interacting areas whereas A had large CCt with S1t. Error bars, 95% confidence interval.

Comparison of functional and anatomical correlation structures.

A. Trial-to-trial fluctuation of neuropil activity of the mouse 2. Sixteen trials presenting the natural video and when the left choice was made were shown. B. K-means clustering of each background pixels (k=8) from two mice resembling the functionally defined areas was shown. C. Correlation between background trial-to-trial activity (rt,neuropil) in pairs of areas was very similar regardless of the task types (r=0.95). D. Seed correlation analysis of neuropil activity for A and AM. E. Injection sites for cortico-cortical projections (blue) and thalamo-cortical projections (magenta). F. Two examples of projection sources (black circles) for a target site (red circle). G. A binary matrix Y (Npix × Ninj). Each column consists of the projection targets of each injection experiment. Two rows indicated by rectangles correspond to the sets of projection sources for the two target sites shown in panel F. H. ranatomy between each one of the pairs of 8 areas. I. ranatomy precisely predict rt,neuropil. J. Correlation between rt,neuropil and ranatomy with corico-cortical or thalamo-cortical projection was plotted against the number of injection sites used for the analysis. K. ranatomy map when the seed is in AI and in AM. Note that these maps resemble panel D. L. Moderate correlation between ranatomy and CCt (r=0.60). M. The first (largest) CCt component was the most inconsistent with the anatomical data. N. Information sharing index was plotted against the dimension of CCt. The first component had the largest information sharing index for all areas. O. Global information sharing index was plotted against the dimension of CCt. The first component had the largest global information sharing index for all areas. Error bars, mean ± s.e.m.

Intrinsic signal imaging and the individual variability of the position of the HVAs.

A. Activity maps of S1 for forelimb, hindlimb, whiskers, tail, trunk, and ear were shown. B. S1 regions obtained in panel A. A “homunculus” was shown for reference. C. Left, mouse skull and the bony landmarks for reference. We used the midline and a lambda on the skull to register the V1 and HVAs across different mice. Right, overlapped maps of the V1 and HVAs for 12 mice. Note that there is large variability of HVA locations across mice. D. Distribution of the center of mass of V1 and HVAs relative to the midline and lambda. E. Comparison of the distribution of the center of mass of V1 and HVAs between different ages (left) and between different sexes (right). No significant difference was found (p>0.05, Wilcoxon rank-sum test). The bar indicates ± standard deviation. F. Left, we used the midpoint of the center of V1 and the center of RL, and the angle between the midline and the straight line connecting the center of V1 and the center of RL to register the V1 and HVAs across different mice (internal registration). Right, the overlapped maps of the V1 and HVAs for 12 mice. Note that all the HVAs are better registered than the panel C, right. G. Location of V1 and HVAs before (left) and after (right) the internal registration. The gray lines connect the points from the same mice. The dot colors are the same as panel D. H. The standard deviation of the location of the center of mass along the anteroposterior and mediolateral axis before and after the internal registration. * p<0.05. ** p<0.01 (Two-sample F-test for equal variance). The dot colors are the same as panel D. I. The mean distance from the center position of each HVA in each mouse to the average center position across mice was compared before and after the registration.

A large field-of-view two-photon microscope and its scanning strategy.

A. A large field-of-view FOV) two-photon microscope, Diesel2p 16. The body holder, head-plate holder, display and four cameras are also shown. B. Left, the brain areas of the right cerebral cortex of the mouse from the Allen atlas (version 3) 33. Right, the scanning strategy to cover the 3 mm by 3 mm square area around the mouse PPC. Two 1.5 mm by 3 mm rectangle areas were sequentially scanned using two Galvo scanners and one resonant scanner. C. The scanning strategy in panel B resulted in 2459 x 1350 pixel2 for each square frame, which yields 29 pixels (>100 laser pulses) for a circle with a 10 μm diameter (the size of a single neuron). D. Representative activity of simultaneously detected neurons in the same FOV. Ten neurons in 8 areas are shown. Each color of trace corresponds to the area. E. Top. The image sequence was registered to the template within a session. Middle. The day-by-day distortion was corrected based on the blood vessels. Bottom. When overlapping the neurons across different mice, we registered images based on the boundaries of areas.

Training schedule and learning curves of the tasks for head-fixed mice.

A. A mouse was head-fixed and held inside a plastic body-holder. The mouse can enter the body-holder smoothly with its head plate into the aperture of the body holder. The display was located at a location covering the left visual field. B. A montage of 4-second natural videos (30 Hz, 24 out of 120 images) was shown. This photo was taken while moving a camera in a cage. C. The serial training schedule for multiple tasks. D. The correct rates of three example mice were plotted as a function of training sessions. Background color corresponds to the type of the task illustrated in panel C. The black, blue and orange dots indicate the imaging experiments of the vision task, history task and history task with a moving body-holder. E. The tongue extension movement during the task was monitored by the video put in front of the mouse (30 fps). Note that this does not necessarily indicate contact between the tongue and spout. We electronically measured contacts between the tongue and the spout and used the signal for the choice. F. Offline analysis captured the tongue extension to the left (left licking, magenta) and tongue extension to the right (right licking, green) during the response period. In both trials with the left choice (top panels) and the right choice (bottom panels), tongue extension movements during the sampling period were hardly observed.

Distribution of the task-related neurons in vision and history task.

A.The t-values for each type of task-related neurons were mapped in the vision task. B. The result of t-SNE analysis in the vision task. C and D. The same as A and B but in the history task, respectively.

Analysis of postures and movements during the task.

A. The position, the velocity and the speed of the three body parts were aligned by the initiation of the response period (RP) for the trials with left and right choice during the mice performing the history task (n=3 mice, total 6 sessions). Shaded areas indicate the sampling period (SP). B. Coordinates of the angle of the body parts. C. The pupil size increased and decreased after the black screen and the natural video started, respectively.

Canonical correlation analysis (CCA).

A. An example source neuron in AM (red) and neurons with rt ≥ 0.3 (blue) were shown. B. Trial-to-trial activity fluctuation of the highly correlated neurons (rt ≥ 0.3, panel b) was color coded in each trial block. C. Illustration of CCA pre-processing. Activity of the population of neurons was extracted in each area, and the PCA (principal component analysis) reduced the data dimensions to 10. Then, “fluctuation” activity was calculated similar to the single neuron case. The data during the sampling period of each trial block (LL:video/left choice. LR:video/right choice. RR: black/right choice. RL: black/left choice) was extracted and averaged. Then, the residual of the data after subtracting the trial-mean in each trial type was concatenated as a new matrix, Mt (one column vector per PC dimension, that is, 10 × Ntrial matrix. D. Illustration of the canonical correlation analysis and calculation of information sharing index. See methods for the precise definition. E. Illustration of the generalized canonical correlation analysis and calculation of global information sharing index. See methods for the precise definition.

The trial-to-trial fluctuation of the neuropil activity and its relation to the single neuron activity.

A. To compare the trial-to-trial fluctuation of the neuropil activity and single neuron activity, the single neurons were randomly selected (filled circle) within the distance, R, from the center (red square). Then, the trial-to-trial fluctuation of the neuropil activity at the center and the trial-to-trial fluctuation of the mean activity of selected neurons was compared. B. The z-scored neuropil activity in each trial was compared with the z-scored mean activity of N selected neurons in the corresponding trial was compared. C. The correlation coefficient obtained in the panel B was repeatedly calculated with a range of neurons averaged and the distance. The lines show the polynomial fitting curves.

Anatomical structure analyzed with anterograde tracer database and its relation to the correlation structure of the trial-to-trial variability in neuropil activity.

A. Using Allen connectivity atlas, the axonal density of corticocortical projection was analyzed. The nine cortical areas where A or AM receive the dense projection are shown. The density was obtained from multiple depths (0, 100, 200, 300, 400, 500, 600, 700 μm from pia), averaged and normalized with the injection volume, then averaged across the experiments. Only the projection from the S1t preferentially distributed in A rather than AM. Top-down axons from dACC, vACC, and OFC preferentially projected to AM rather than A, as well as visual cortices (AL, PM, and V1). Gray boundaries show A and AM in Allen brain atlas, and the white boundaries indicate A and AM in our definition. B. Top and second rows, the correlation between each cortical region and the seed points in each area (A, AM, RL, V1, PM, RSC, S1t, S1b) was color-mapped. The high correlation indicates that the region has highly shared input from the cortex (“CC”, top row) or the thalamus (“TC”, second row) with the seed region. A dotted circle indicates the imaging area. Second-bottom and bottom, the seed correlation of trial-to-trial activity fluctuation during the vision task (“M2, V”, second-bottom row) and during passively viewing the moving bar (“M2, P”, bottom row) of mouse 2. Note that all the correlation maps in each column are similar. C. A snapshot of the online interactive software, “BrainModules”, for exploring the anatomical seed correlation analysis. Link: https://web.ece.ucsb.edu/~riichirohira/TopBM-2.06/index.html

A. Randomly selected 20 active neurons (top) and randomly selected 20 highly active neurons (kurtosis > 5, bottom). B-J. The same analysis using only highly active neurons (B: Figure 5A; C: Figure 5B; D: Figure 5E (rt,single >0.25); E: Figure 5F; F: Figure 5H; G: Figure 5I; H: Figure 5J, I: Figure 6L; J: Figure 6M, K:, Figure 6N, L: Figure 6O).