This readme file was generated on 2024-10-10 by Jiaxuan Qi ------------------- GENERAL INFORMATION ------------------- Title of Dataset: Data from: Dual Neuromodulatory Dynamics Underlie Birdsong Learning Author Contact Information (Name, Institution, Email, ORCID) Principal Investigator: Richard Mooney Institution: Duke University Email: mooney@neuro.duke.edu ORCID: https://orcid.org/0000-0002-3308-1367 Associate or Co-investigator: John Pearson Institution: Duke University Email: john.pearson@duke.edu ORCID: https://orcid.org/0000-0002-9876-7837 Alternate Contact(s): Jiaxuan Qi, Drew C Schreiner, Miles Martinez. *Date of data collection (single date, range, approximate date): 2019-2024 *Geographic location of data collection (if applicable): Durham, NC, USA *Funding and grant numbers (if applicable): NIH 5R01 NS099288; NIH RF1 NS118424; NIH F32 MH132152; NIH F31 NS132469 -------------------- DATA & FILE OVERVIEW -------------------- This dataset includes core data from the paper "Dual Neuromodulatory Dynamics Underlie Birdsong Learning". While learning in response to extrinsic reinforcement is theorized to be driven by dopamine signals that encode the difference between expected and experienced rewards, skills that enable verbal or musical expression can be learned without extrinsic reinforcement. Instead, spontaneous execution of these skills is thought to be intrinsically reinforcing. Whether dopamine signals similarly guide learning of these intrinsically reinforced behaviors is unknown. In juvenile zebra finches learning from an adult tutor, dopamine signaling in a song-specialized basal ganglia region is required for successful song copying, a spontaneous, intrinsically reinforced process. Here we show that dopamine dynamics in the song basal ganglia faithfully track the learned quality of juvenile song performance on a rendition-by-rendition basis. Furthermore, dopamine release in the basal ganglia is driven not only by inputs from midbrain dopamine neurons classically associated with reinforcement learning but also by song premotor inputs, which act via local cholinergic signaling to elevate dopamine during singing. While both cholinergic and dopaminergic signaling are necessary for juvenile song learning, only dopamine tracks the learned quality of song performance. Therefore, dopamine dynamics in the basal ganglia encode performance quality during self-directed, long-term learning of natural behaviors. File list (filenames, directory structure (for zipped files) and brief description of all data files): Microdialysis zipped files contains 3 subfolders indicating different drug groups(DHBE, Muscimol, SCH23390), each contains individual animals. Each of these contains song files (wav files) collected from each day. Files were in the format of: birdTag/age/songs. DA_FP and ACh_FP zipped files each contain subfolders of individual animals. Each of these contains fiber photometry (FP) data collected within each day, and each session. Each recording session was either in the format of birdTag/month_day_hour_minute_song_LaserPower_hemisphere, or BirdTag/dates/time. Within each session it contains photometry signal and timestamp signal (csv files), song data (wav files), and video data (avi files). Additional related data in the paper that was not included in the current data package: Singing-related DA and ACh dynamics in adult birds White noise experiments Optofluid cannula experiments Any additional data is available upon request. Please contact the corresponding author, Dr. Richard Mooney, at mooney@neuro.duke.edu, or the first author, Dr. Jiaxuan Qi, at jiaxuan.qi@duke.edu. -------------------------- METHODOLOGICAL INFORMATION -------------------------- Description of methods used for collection/generation of data: Microdialysis zipped file was collected by a custom sound analysis program, Sound Analysis Pro 2011 (SAP; http://soundanalysispro.com/) DA_FP and ACh_FP zipped files was collected by a photometry system (Neurophotometrics FP3001 or FP3002). In brief, light from 470 nm channel (for imaging in green fluorescence) and 415 nm channel (for isosbestic control signal to detect motion artifacts) LEDs was generated in turn at a total frame rate of 30 Hz (15 Hz per wavelength), bandpass filtered and directed down the patch cord via a 20× objective. Light was measured at the tip of the patch cord to make sure the power for each wavelength was ~50 μW. Emitted GRAB-DA/ACh fluorescence was collected through the same cannula and patch cord, split by a dichroic mirror, bandpass filtered, and focused onto opposite sides of an sCMOS camera sensor. Data was acquired using the open-source software Bonsai. Synchronized video and sound recordings were acquired using a webcam (Logitech). VAE, ANN, and ezTrack were used for song and motion analysis in the paper. Custom codes are available at https://github.com/pearsonlab/autoencoded-vocal-analysis, https://github.com/SamuelBrudner/juvenile_syllable_analysis, and https://github.com/denisecailab/ezTrack.. For detailed post-processing of the data, see Methods part of the paper. In case readers cannot access the publication, the post-processing procedures are reiterated here: Microdialysis: Individual song renditions were segmented into their component syllables using the Autoencoded Vocal Analysis (AVA) package in Python2. Briefly, spectrograms were taken as the log modulus of a short time Fourier transform (Hann windows, sample rate: 44.1 kHz, segment length: 512, overlap: 320). The total power of the spectrogram was calculated across channels. Syllables were detected when the total power exceeded a threshold, manually chosen for each bird. Detected syllables were then manually scaled and clipped to an appropriate range. They were interpolated to 128 target frequencies linearly spaced between 300 and 8000 Hz and 128 target times, resulting in a 128 x 128 spectrogram. After VAE training, syllables were labeled according to the UMAP projections of embedded spectrograms using custom software in MATLAB. Once the VAE was fully trained and syllables were segmented, we extracted latent embeddings for each syllable. We then trained a separate ANN with 3 linear layers to predict the age at which each syllable was produced. For drug treatment experiments, we restricted ANN training to a narrower developmental window that bracketed this drug treatment day (2-4 days before and after drug treatment day). Specially, both the drug day and the previous saline day were held out of the training set. A predicted age for each syllable was generated by the ANN to represent the maturity of each syllable. FP: For song analysis, songs were fed into the VAE-ANN to generate predicted age scores for each syllable. For FP signal analysis, photometry data were analyzed using custom-written Matlab scripts. In each imaging session, the blue channel was fitted to and subtracted from the green signal using the Matlab polyfit function. DA/ACh signals were z-scored in each imaging session and aligned to the audio recordings, for example, to syllable onset. -------------------------- DATA-SPECIFIC INFORMATION -------------------------- Microdialysis: Microdialysis zipped file contains wav files, each wav file contains audio during each recording session. A "ReadMe" files inside each animal folder indicates the age and drug treatment of that animal. FP: DA_FP and ACh_FP zipped files contains photometry signal and timestamp signal (csv files), song data (wav files), and video data (avi files). comebo_test_xxx.csv file contain the photometry signal. For blk658, blu688, grn671, red660: Timestamp column represented the time in seconds that frame was collected relatvie to the internal clock of photometry rig. Region0G column represented the fluorescence signal in arbitrary unit of that frame. LedState column: 6 and 1 represented either 415 or 470 channel. We post hoc identified the channel with high baseline fluorescence as the 470 channel. Columns of Stimulation, Output0, Output1, Input0, Input1 were of no use in our experiment. For pur568, pur569: Timestamp column represented the time in seconds that each frame was collected relatvie to the internal clock of photometry rig. Region0G column represented the fluorescence signal in arbitrary unit of each frame. LedState column: 2 and 1 represented either 415 or 470 channel. We post hoc identified the channel with high baseline fluorescence as the 470 channel. Columns of Stimulation, Output0, Output1, Input0, Input1 were of no use in our experiment. For blk633, red561, red562, red662: Timestamp column represented the time in seconds that frame was collected relatvie to the internal clock of photometry rig. Region0G column represented the fluorescence signal in arbitrary unit of that frame. LedState column: 2 and 1 represented 470 and 415 channel, respectively. Columns of Stimulation, Output0, Output1, Input0, Input1 were of no use in our experiment. frame_time_xxx.csv, test_audio_xxx.csv, test_video_xxx.csv represented the time in seconds that each photometry frame/audior frame/vidoe frame was collected relative to the internal clock of photometry rig. ------------------------- USE and ACCESS INFORMATION -------------------------- Data License: CC0 Other Rights Information: To cite the data: Mooney, R. & Pearson, J. (2024). Data from: Dual Neuromodulatory Dynamics Underlie Birdsong Learning. Duke Research Data Repository. https://doi.org/10.7924/r4s186852