script for deeplabcut analysis of behavior  (MathWorks Inc)

Journal: bioRxiv

Article Title: VTA dopamine neurons drive spatiotemporally heterogeneous striatal dopamine signals during learning

doi: 10.1101/2023.07.01.547331

Figure Lengend Snippet: A) Behavior data was extracted from video recordings using the DeepLabCut pipeline. B) Video frames containing experimenter labeled rat body parts (ears, nose, tail base, etc) and static features of the environment (cue, chamber corners), were used to train a deep learning network to estimate frame by frame x,y coordinates for the entire video data set. This data was used to calculate movement and position information across optogenetic Pavlovian conditioning. C) Head direction was interpolated by calculating the angle between the vector projecting from the cranial implant to the nose with the vector projecting from the cranial implant to the cue. D) Example trials showing linear cue-evoked movement. Each dot represents the position of the rat’s nose on an individual video frame, on 5 different trials early in training. E) Example trials showing rotational movement during 5 trials late in training. F) Polar plots showing distribution of head direction angles for an example rat across training.

Article Snippet: Markerless tracking of animal body parts was conducted using version 2.2.1.1 of the DeepLabCut (DLC) Toolbox 37,41 and analysis of movement features based on these tracked coordinates was conducted in Matlab R2020b (Mathworks).

Techniques: Labeling, Plasmid Preparation

Journal: bioRxiv

Article Title: Long-term optical imaging of the spinal cord in awake, behaving animals

doi: 10.1101/2023.05.22.541477

Figure Lengend Snippet: a. 3D model of the recording chamber implanted at the T12-L1 vertebrae. Skeleton created using microCT data from a naive mouse . Insets: a’ , optical window (cover glass) provides access to the dorsal spinal cord post-T13 laminectomy and a” , coronal view slightly off-axis shows a slice through side bars and highlights side bar placement adjacent to the articular processes and location of Teflon AF (green film). b. Diagram of chamber components that permit optical access to the lumbar (L4-L5) or lumbosacral (L5-S1) spinal cord. Super glue fixes the stabilizing plate to the two side bars. c. Two Teflon materials inhibit post-laminectomy fibrosis: opaque PRECLUDE is applied immediately after laminectomy (as in d ), then removed and replaced subsequently by transparent Teflon AF fluoropolymer (as in e ). d. Workflow for long-term spinal cord imaging includes three surgical steps. e. Overhead view of the spinal chamber one week after window placement. f. Using only silicone to cover the spinal cord leads to fibrosis and disappearance of the dorsal vein (black arrow) within a month (Mouse #1). Sequencing PRECLUDE and Teflon AF inhibits fibrosis, allowing visualization of the dorsal vein and ascending venules (black arrows) for months. Fluorescent imaging of mouse #2 and #3 in and , respectively. g. Whole body microCT sagittal max projection after chamber implantation. Chamber (green dashed line) fashioned using a 3D printed radiotransparent material (BioMed Clear). Scale bar, 5.0 mm. h. Coronal slice (yellow arrow and dashed line in g ) from microCT, post laminectomy, shows intact surrounding bone in relation to the glass window. A surface layer of radiotransparent dental cement secures the glass window to the side bars. Scale bar, 1.0 mm. i. Multi-vertebral 3D reconstruction of microCT data in g-h confirms T13 window placement and bone integrity. j. Two mice exhibiting normal behaviors after chamber implant (see ). k. Body weight of chamber-implanted mice (n = 16) pre- and post-surgery compared to age-matched controls (n = 2). l. Tracking of open field locomotion (30 min) after chamber implant. Inset: DeepLabCut markers of individual body parts used for openfield tracking. Scale bar, 10 cm. m. Locomotor speed of the same mouse during 30-min sessions, pre- and post-surgery (Stage 1, side bar). n. Mean open field locomotor speed comparing naïve (n = 7) and post-surgery mice at different stages (n = 7, 2, 18, and 19, respectively) with “Window late” indicating beyond 30 days post window procedure. Most mice in n-p are used for imaging. Bar plot and error bars in n - p are mean ± SD, gray lines indicate animals measured across multiple stages. o. Mean latency to fall in final (3rd) trial on an accelerating rotarod comparing naïve (n = 14) and post-surgery mice at different stages (n = 12, 2, 10, 5, 5, 5, 5, respectively). p. Von Frey mechanical thresholds comparing naïve (n = 9) and post-surgery mice at different stages (n = 7, 6, 13, and 19, respectively) with “Window late” indicating beyond 30 days post window procedure. q. Microglia (CX3CR1-EYFP) and astrocyte (α-GFAP) immunofluorescence of 100-μm thick L4 sections before and after chamber implant. Scale bars, 300 μm and 50 μm (zoom).

Article Snippet: We imported feature locations from DeepLabCut CSV files into MATLAB using a custom CIAtah function.

Techniques: Imaging, Sequencing, Immunofluorescence

Journal: bioRxiv

Article Title: Long-term optical imaging of the spinal cord in awake, behaving animals

doi: 10.1101/2023.05.22.541477

Figure Lengend Snippet: a. 3D printed phantom, used for microCT validation studies, representing skull and spinal column. A 3D printed spinal chamber (Surgical Guide) is implanted with radio-transparent and -opaque dental cement along with miniature brass and steel screws (to evaluate impact on microCT scans). b. Reconstructed horizontal view from microCT scan of phantom in a of 3D printed side bars showing radiotransparent properties of the materials. Yellow bars indicate acquisition planes with reconstruction artifacts; red arrows highlight reduced reconstruction of spinal chamber and column. Scale bars, 2 mm. c. Coronal view of scan as in b shows details of metal screws along with artifact scan lines aligned with areas of higher and lower density of the metal screws. Scale bars, 2 mm. d. Coronal view through sections of the phantom without (left) and with (right) metal screws in the acquisition plane shows the artifacts introduced by miniature steel screws. Scale bars, 2 mm. e. Coronal section from microCT scan (20 μm resolution) of a dissected mouse spinal column, with tissue and muscle attached, placed inside a 3D printed test piece, using the same material (BioMED Clear) as for the 3D printed spinal chamber. Note the ability to reconstruct details and internal geometry of the spinal column, along with surrounding soft tissue. Scale bars, 2 mm. f. Off-axis and sagittal views of 3D reconstructed microCT scan as in e . Note the detailed reconstruction of the spinal column, soft tissue, and geometry of the test piece, confirming that BioMed Clear is microCT compatible. g. Pipeline for 3D reconstruction of microCT scans; see for details. h. Coronal view of mouse with 3D printed spinal chamber (see – ) showing an acquisition plane at the T13 laminectomy location. Scale bars, 2 mm. i. 3D reconstruction of the mouse in – and h with bone (gray), spinal chamber (blue), and circular glass coverslip window (red). Inset: zoomed in view highlights the T13 laminectomy and placement of the circular coverglass. j. Model error (sum of score map cross-entropy and body part location L1-distance losses) as a function of DeepLabCut iterations for model trained using data from 3 mice. Model is trained for 600,000 iterations until convergence. k. Weights of individual animals after bar implant. l. Mean (per animal) latency to fall in all three trials on an accelerating rotarod, comparing naïve (n = 14) and post-surgery mice, at different stages (n = 12, 2, 10, 5, 5, 5, 5, respectively).

Article Snippet: We imported feature locations from DeepLabCut CSV files into MATLAB using a custom CIAtah function.

Techniques: Biomarker Discovery

Journal: bioRxiv

Article Title: Long-term optical imaging of the spinal cord in awake, behaving animals

doi: 10.1101/2023.05.22.541477

Figure Lengend Snippet: a. Imaging a large spinal cord field is subject to several types of motion artifacts. b. Modular motion correction pipeline that addresses each of the issues outlined in a : features identified using deep learning followed by control point and rigid registration (large displacement motion correction method, LD-MCM ), deformation correction using displacement fields followed by rigid registration (non-rigid motion correction method, NR-MCM ), and manual or automated cross-session motion correction ( CS-MCM ). c. LD-MCM utilizes deep learning, here DeepLabCut (DLC), to identify vascular features after manual annotation and training. LD-MCM uses these features to transform and register frames to a reference frame’s features. Point clouds overlapping each feature in the reference frame show rostrocaudal and mediolateral extent of motion during each frame of an entire imaging session (2.31 mins, 20 Hz). Inset, zoomed in field of view shows overlap of markers with distinct vasculature features. Scale bar, 300 μm. d. Mean projection image across all movie frames for the raw movie and after TurboReg or LD-MCM demonstrates reduced motion with LD-MCM. Arrows indicate features seen only in LD-MCM (yellow) and others that are barely visible after TurboReg (white). Scale bar, 300 μm. e. Point clouds for rostrocaudal and mediolateral movement of feature #3 (as in c ) after motion correction with TurboReg, NoRMCorre, and LD-MCM. Each dot represents the location of that feature on an individual frame during an imaging session (2.31 min, 20 Hz). Arrow indicates location of LD-MCM points, showing negligible post-procedure motion. f. Boxplots show rostrocaudal displacement of the spinal cord relative to the mean location in the raw movie and after motion correction with TurboReg, NoRMCorre, and LD-MCM over all features, for each of 3 movies (n = 2 mice). Features for all methods identified with DLC. Arrow highlights negligible post-LD-MCM motion (as in e ). g. Synthetic image (116 × 77 px) before and after image alignment with NR-MCM. Vectors (red) indicate the displacement field orientation and magnitude at a given pixel location; the vector field is sub-sampled (5x) and magnitude scaled for display purposes. h. NR-MCM (as in g ) on an example frame from one-photon fluorescence imaging of spinal cord GCaMP-expressing neurons (same mouse as ). Yellow arrows highlight features that are aligned after registration. Scale bar, 300 μm. i. Mean projection images of the first 5,000 frames in a movie (~12.5 min, 20 Hz) show improved motion correction with NR-MCM compared to TurboReg and NoRMCorre. Yellow arrows highlight stable features in NR-MCM movies. Scale bar, 300 μm. j. 2D correlation coefficient of all frames to the mean frame of the movie (as in i ) for NR-MCM compared to raw, TurboReg, and NoRMCorre. All movies were spatially filtered to remove large magnitude, low-frequency changes in fluorescence, which artificially enhances correlations. Right inset: histogram of correlation coefficients across all frames; vertical axis is aligned to that in the main graph. k. Boxplots summarize results, as in j, over 3 movies (n = 2 mice). Boxplots in all figures display the 1st, 2nd, and 3rd quartiles with whiskers indicating 1.5*IQR; outliers are omitted.

Article Snippet: We imported feature locations from DeepLabCut CSV files into MATLAB using a custom CIAtah function.

Techniques: Imaging, Control, Plasmid Preparation, Fluorescence, Expressing

Journal: bioRxiv

Article Title: Long-term optical imaging of the spinal cord in awake, behaving animals

doi: 10.1101/2023.05.22.541477

Figure Lengend Snippet: a. Comparison of reference frame 42 to movement frame 804 before and after control point motion correction. Scale bar, 300 μm. b. Example from a Phox2a-Cre; Ai162 mouse of DLC-identified vascular features used for cross-session registration with day 41 used for training the DLC model. Scale bar, 300 μm. c. Model error (sum of score map cross-entropy and body part location L1-distance losses) as a function of DeepLabCut iterations shows that performance approaches convergence by 500,000 iterations. d. Spearman’s correlation of each feature to other features in a movie from a Phox2a-Cre; Ai162 mouse. Green arrow indicates a feature that has reduced correlation with all other features and can thus be thrown out, improving overall motion correction. e. Point clouds for rostrocaudal and mediolateral movement of several features (from mouse in a ) before motion correction. Each dot (2001 frames) represents the location of that feature on an individual frame during an imaging session (~6 min, 13.9 Hz). f. DLC identification is consistent across both (i) large mediolateral shifts in the field of view not present in the training set and (ii) camera errors that result in a split of the field of view. These shifts can be used to rapidly exclude frames or movies in downstream analysis. Scale bar, 300 μm. g. Consistent DeepLapCut labeling of vascular features in a Phox2a-Cre; Ai162 (GCaMP6s) mouse across 52 neural activity imaging sessions, spanning nearly 5 months. DeepLabCut was trained using manual annotation from 20 frames from a single imaging session (day 75) demonstrating the robustness of annotation and consistency of imaging. Scale bar, 300 μm.

Article Snippet: We imported feature locations from DeepLabCut CSV files into MATLAB using a custom CIAtah function.

Techniques: Comparison, Control, Imaging, Labeling, Activity Assay

Journal: bioRxiv

Article Title: Long-term optical imaging of the spinal cord in awake, behaving animals

doi: 10.1101/2023.05.22.541477

Figure Lengend Snippet: a. Visibly opaque (black) infrared acrylic allows imaging of animal behavior using near-IR light sources and cameras, while blocking animal observation of experimenters (e.g. during stimulus delivery). b. Model error (sum of score map cross-entropy and body part location L1-distance losses) as a function of DeepLabCut iterations for model trained using data from one mouse. Model training is terminated after 500,000 iterations, when the loss asymptotes. c. Part affinity fields for DeepLabCut networks across multiple cameras. d. Speed of individual body parts across shows correlation of body part movement across cameras (#1–4). The mean speed across all cameras for each body part is used for display in . Camera locations correspond to 1, left side of the body; 2, right side of the body; 3, right face; and 4, below the animal. Letters below each black arrow indicate the stimulus presented (C: cold; P: pinch; H; heat; A: air puff; S: sound); black bar denotes duration of the sound stimuli. e. 3D CAD of miniature microscope positioning above spinal implant chamber. f. Image of miniature microscope mounting. g. View of dorsal vein after procedure in f . h. Ambulating mouse after mounting procedure. i. General locomotion of a mouse in an open field during freely moving spinal cord imaging. Scale bar, 10 cm. j. Locomotor trace during the open field session in i (3.68 min, 10 Hz).

Article Snippet: We imported feature locations from DeepLabCut CSV files into MATLAB using a custom CIAtah function.

Techniques: Imaging, Blocking Assay, Microscopy