Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior

# Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior This Zenodo record contains the MATLAB analysis code and processed data supporting: > Benjamin A. Filio, Amma Otchere, Subhiksha Srinivasan, Srijan Thota, Luke Drake, Lizmaylin Ramos, Philipp Maurus, and Mark J. Wagner. “Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior.” *Nature Neuroscience* (in press). The study examined how cerebellar granule cells (GrCs) and climbing fibers (CFs) represent rewarding outcomes that do not require physical consumption. Head-fixed mice learned to push a robotic handle for delayed optogenetic activation of ventral tegmental area dopamine neurons or electrical stimulation of the medial forebrain bundle. Additional experiments examined water reward, altered reward delays, chronic cell registration, passive stimulation, CF-mediated reinforcement, and optogenetic inhibition of GrCs during learning. The deposit includes processed, analysis-ready MATLAB data and scripts that reproduce the data-driven panels and statistical analyses. Raw two-photon movies, original behavioral videos, histology, schematics, and other non-data artwork are not included. No human-participant data or personally identifiable human information are included. ## Files in this Zenodo record ```text README.md filio-2026-main.zip mainData1.mat mainData2.mat mainLearn.mat CF_novice.mat IOChRmine.mat IOChRmine_learn.mat IOChRmine_imaging.mat passiveMFB.mat gtacr1.mat gtacr1_learn.mat ``` `filio-2026-main.zip` contains the nine top-level analysis scripts and the bundled MATLAB helper functions in `dependencies/`. The ten MAT files are supplied separately because of their size. ## Quick start 1. Download the code ZIP and all ten MAT files. 2. Extract the code ZIP while preserving its directory structure. 3. Either create a `data/` folder beside the top-level scripts and place all MAT files there, or leave the MAT files together in another folder and select that folder when MATLAB prompts you. 4. Open a top-level `.m` script in the MATLAB Editor. 5. Run the complete script. Each script locates its own directory, adds the adjacent `dependencies/` folder to the MATLAB path, and first looks for a local `data/` folder. If that folder is absent, MATLAB prompts for the directory containing the MAT files. Running a script from the beginning is the most reliable way to reproduce its analyses. For section-by-section execution, run the initialization section first, retain variables created by preceding sections, and keep the script active in the MATLAB Editor. Later sections can depend on variables calculated earlier in the same script. Figures are created as MATLAB figure windows, with statistical results displayed in figure titles or the Command Window. Publication-layout figure files are not saved automatically. ## Code structure After extraction, the code archive contains: ```text filio-2026-main/ |-- MainData.m |-- learningCurves_DA.m |-- CF_novice.m |-- IOChRmine.m |-- learningCurves_IOChRmine.m |-- IOChRmine_imaging.m |-- passiveMFB.m |-- gtacr1.m |-- learningCurves_gtacr1.m `-- dependencies/ `-- MATLAB analysis, plotting, statistics, filtering, and export helpers ``` The files in `dependencies/` are called automatically by the top-level scripts and do not need to be run independently. ## Top-level analysis scripts | Script | Required data | Principal analyses and manuscript figures | |---|---|---| | `MainData.m` | `mainData1.mat`, `mainData2.mat` | Primary behavioral, GrC, CF, pose-estimation, reward-delay, reward-type, and chronic-registration analyses for Figures 1–5 and Extended Data Figures 1–8 | | `learningCurves_DA.m` | `mainLearn.mat` | Dopamine-reinforced learning across training days; Extended Data Figure 1b–e | | `CF_novice.m` | `CF_novice.mat` | CF responses during the first dopamine-reinforced training session; Extended Data Figure 8g,h | | `IOChRmine.m` | `IOChRmine.mat` | Behavior reinforced by optogenetic CF activation and comparisons with controls and other reinforcers; Figure 6d–g and Extended Data Figure 9a–e | | `learningCurves_IOChRmine.m` | `IOChRmine_learn.mat` | Longitudinal learning with CF activation versus light-only controls; Figure 6h–j | | `IOChRmine_imaging.m` | `IOChRmine_imaging.mat` | GrC activity during CF-reinforced behavior; Figure 6k,l and Extended Data Figure 9f | | `passiveMFB.m` | `passiveMFB.mat` | GrC activity during passive medial forebrain bundle stimulation; Extended Data Figure 6b–e | | `gtacr1.m` | `gtacr1.mat` | Expert behavior following GrC-inhibition training, laser timing, control comparisons, and laser-off recovery; Figure 7d–i and Extended Data Figure 10b–d,h–j | | `learningCurves_gtacr1.m` | `gtacr1_learn.mat` | Longitudinal effects of GrC inhibition on learning; Extended Data Figure 10e–g | There is no required order across the nine top-level scripts. Each loads only the data file or files listed above. ## Data-file descriptions ### `mainData1.mat` and `mainData2.mat` These are two parts of the same primary dataset and are loaded together by `MainData.m`. They were split only to facilitate storage and distribution. - `micedats1` and `micedats2`: session-level cell arrays containing processed behavioral, imaging, stimulation, licking, reward, and pose-estimation data. `MainData.m` concatenates them into `micedats`. - `dirTable`: one row per dopamine-reward session, containing metadata used to select analysis cohorts, including reward type, nominal reward delay, imaging type, availability of GrC, CF, or pose data, water-training status, chronic-session relationships, stimulation duration, imaging day, sex, and animal/session identifiers. Together, these files contain 55 primary session records. Not every session contains every modality; optional nested fields are present only when the corresponding imaging, water-reward, chronic-registration, or video measurement was available. ### `mainLearn.mat` - `mouseData`: 16-element animal-level structure array containing dopamine-reinforced behavioral training data. - Common fields include `mousePath`, `rewardType`, `trainingDirs`, `normTime`, and `trainingDats`. - `trainingDats` contains one processed session structure per training day. `normTime` maps each animal’s observed training days onto the normalized learning axis used by `learningCurves_DA.m`. ### `CF_novice.mat` - `micedats`: three session structures from novice animals during the first day of dopamine-reinforced training. - Contains processed CF fluorescence/spike activity, reward-aligned activity, handle behavior, reward timing, and licking. - The dataset contains 329 CFs across three mice. ### `IOChRmine.mat` - `iodats`: 10 expert-session structures from 10 mice trained with optogenetic CF activation as the reinforcer. - `ctlDats`: eight expert-session structures from four light-only control mice. - `bsln`: summary behavioral arrays for comparisons with VTA, MFB, and water reinforcement. - Session structures include reach behavior, reward/laser timing, licking, aligned trajectories, cumulative successes, and expert behavioral metrics. ### `IOChRmine_learn.mat` - `mouseData`: 14-element animal-level structure array comprising 10 CF-reinforcement mice and four light-only controls. - Contains five to seven processed training sessions per animal, group labels, normalized training time, and session data used to calculate learning-related changes. ### `IOChRmine_imaging.mat` - `micedats`: 10 session structures containing GrC calcium activity during CF-reinforced behavior. - Contains 812 GrCs in total, including the activity used to identify 64 delay-ramping GrCs. - Includes continuous and event-aligned fluorescence, behavior, trial classifications, reward/laser signals, and time axes. ### `passiveMFB.mat` - `ccdats`: 14 passive-stimulation session structures from seven mice. - Contains 749 GrCs in total, together with stimulation-aligned calcium activity and behavioral, cue, and reward signals. ### `gtacr1.mat` - `micedats_opto`: a six-element cell array of experimental session groups used to analyze GrC stGtACR1 inhibition, genotype-matched controls, laser-off testing, and related conditions. Some groups contain empty placeholders where a corresponding test session was unavailable. - `bsln`: baseline VTA, MFB, and water behavioral summary arrays used for control and normative comparisons. - Session structures contain reach behavior, trial classifications, stimulation signals, laser timing, licking, and movement-start- and movement-end-aligned data. - The principal comparison pools 27 MFB sessions from 19 imaging mice with eight sessions from eight genotype-matched controls to form the 35-session normative group. The inhibition and first laser-off recovery groups each contain 11 sessions. ### `gtacr1_learn.mat` - `mouseData`: 33-element animal-level structure array containing 22 normative-baseline and 11 GrC-inhibition animals. - Contains three to eight processed training sessions per animal (190 animal-sessions in total), group labels, normalized training time, and behavioral records used to quantify changes from the first to last training day. ## Common MATLAB structures and variables Field availability varies by experiment and modality. ### Behavior and trials - `pos`: handle-position data. In aligned structures, dimensions are generally trial × time × coordinate. - `truestart`, `trueend`: detected movement-start and movement-end sample indices in the continuous behavioral recording. - `rewtimes`: reward or reinforcing-stimulation sample indices. - `rewarded`: trial-level indicator for reward delivery. - `goodMvmt` or `goodMvmts`: trial-quality masks used by the analyses. - `mvlen`: reach length, recomputed from aligned handle position where required. - `states`: behavioral-state signal. - `sol` or `laser`: stimulation command or detected illumination signal. - `lick`, `lickrate`: lick detector and derived lick-rate data. - `attmptRate`, `succRate`, `compltd`: attempts per minute, successes per minute, and completion proportion or percentage. Some scripts recompute these values from the underlying trials. - `dtb`: behavioral sampling interval in seconds. - `tmpxx`: behavioral time relative to the alignment event, in seconds. A successful reach is defined in the analysis scripts as a movement longer than 6 mm. See the article Methods and individual scripts for the complete trial-selection criteria. ### Neural activity - `sigFilt_right_green` and related green-channel fields: processed GrC fluorescence. - `sigFilt_right_red` and related red-channel fields: processed red-channel/CF fluorescence where applicable. - `rewAlgn`, `endAlgn`, `startAlgn`: structures containing data aligned to reward, movement end, or movement start. Neural arrays are generally trial × cell × aligned imaging time; behavioral arrays are generally trial × time × coordinate. - `tmpx_right` and related variables: imaging time relative to the alignment event, in seconds. CF datasets can include processed fluorescence, detected spiking, and spike-rate representations. The exact filtering, normalization, classification windows, and statistical criteria are implemented in the associated scripts and described in the paper. ### Markerless pose estimation - `dlc`: DeepLabCut-derived body-part labels and processed coordinate data. - `dlclikeli`: tracking likelihood values where available. - `tmpxDLC`: pose-estimation time relative to the alignment event, in seconds. - Derived body-part speeds are in pixels per second and include jaw, nose, right forelimb, and right hindlimb measurements where those landmarks passed tracking-quality criteria. Pose fields can be absent for sessions without usable video. The analysis excludes trials with poor lever/encoder agreement, prolonged low tracking confidence, missing coordinates, or implausibly high speeds. ### Chronic registration - `chronicGreen` and `chronicRed`: structures for GrCs or CFs registered across sessions. - These structures can include matched-cell matrices such as `finalMatchesMat`, session/day labels, and activity from paired experimental contexts. Cell counts after chronic matching are therefore smaller than the independently detected populations in the constituent sessions. ### Original paths Some fields retain original acquisition-directory strings as provenance or session identifiers. These paths are not expected to exist on another computer and do not need to be edited; the scripts load the distributed MAT files from the selected data directory. ## Missing data and filtering - `NaN` denotes missing, excluded, or unavailable numerical observations. - Empty cells and absent optional structure fields indicate that a modality or paired session was unavailable. - Top-level counts describe file contents, but individual panels apply the trial-, cell-, session-, registration-, and quality-filtering criteria specified in their scripts. - Consequently, the number of animals, sessions, cells, or trials can legitimately differ among panels derived from the same MAT file. - Statistical tests and multiple-comparison adjustments are implemented in the scripts and described in the corresponding figure captions and Methods. ## Software requirements The code was verified with: - MATLAB R2025a - Statistics and Machine Learning Toolbox - Signal Processing Toolbox Earlier MATLAB releases may work but were not tested. The MAT files are large; sufficient RAM is required, particularly for `MainData.m`, which loads both primary files. ## Source-data export Each top-level script defines: ```matlab exportSrcData = false; ``` Leave this set to `false` for ordinary figure reproduction. The numerical source-data workbooks accompanying the published article are not required to run the analyses. If regenerating source data from the code, set the parameter to `true` and ensure that a writable `excel/` directory containing `MainData.xlsx` is present beside the scripts. Close the workbook in Excel before running an export. ## Reproducibility notes - Some analyses use random shuffling, permutation controls, or bootstrap uncertainty estimates. Set the MATLAB random-number-generator seed before the relevant section when exact numerical reproduction is required. - Figure appearance can vary modestly with MATLAB version, operating system, fonts, and display settings. - The supplied analyses use processed and normalized signals stored in the MAT files. Re-running motion correction, neural source extraction, spike detection, or pose estimation requires the original imaging or video data, which are outside this deposit. ## Citation When using either the code or processed data, cite the associated *Nature Neuroscience* article and this Zenodo record. Use the version-specific DOI and formatted citation displayed on the Zenodo landing page. ## License Reuse is governed by the license specified for this Zenodo record. ## Contact Questions about the code or data may be directed to: Mark J. Wagner National Institute of Neurological Disorders and Stroke [e-mail]