Nature Protocols: doi: /nprot Supplementary Figure 1. Preparation of surgery tools for lens handling and bone removal.

Transcription

1 Supplementary Figure 1 Preparation of surgery tools for lens handling and bone removal. (a) Digital image depicting the tips of a pair of forceps that have been covered with ~10 mm of heat shrink. (b) Bulldog serrefine covered with ~5 mm of heat shrink. (c) Heat shrink covered bulldog serrefine tightly clamped in guide cannula holder. (d) Guide cannula holder with secured bulldog serrefine attached to a stereotaxic apparatus. (e) 27G needle bent at a 45 angle. Prepare needle before surgery and use to scoop out bone fragments after and peel away dura.

2 Supplementary Figure 2 Confirmation of cell type specific expression of GCaMP. (a) Confocal image stack of GCaMP expression within the VTA of a TH-Cre adult male mouse. (b) Confocal image stack of VTA neurons labeled with GCaMP6m (left) and stained for TH (center). Merged image shows co-labeling of GCaMP6m and TH indicating that GCaMP6m was successfully expressed within VTA-DA neurons (AAVdj-Ef1a-GCaMP6m; 1:2 dilution, 4.8 x ). All procedures were approved by UNC IACUC.

3 Supplementary Figure 3 Visualization of Ca 2+ transients through either a 0.5 or 1 mm lens implanted in the prefrontal cortex. Fluorescently encoded Ca 2+ dynamics were monitored over a 10-min acute recording session (15 Hz, down sampled to 5Hz) in adult wild type males. Either a (a) 0.5 or (b) 1 mm diameter lens was used to visualize fluorescently encoded Ca 2+ signals (AAV-DJ-CamkIIa- GCaMP6s, 5.3 x 10 12, 1:4 dilution, UNC vector core). Top row shows mean intensity projections of fluorescence signals across the entire 10-min session (spatial bin of 4). Notice the difference in contrast between anatomical structures that is conveyed by each of the lenses. PCA/ICA analysis was used to extract the spatial location (middle row) and activity traces (bottom row) of individual cells. 1.0 mm diameter lenses (right) have larger fields of view than 0.5 mm diameter lenses (left) and therefore have the potential to detect a larger number of cells. However, the quality of spatial and temporal filters is similar between lenses. Optical distortion of cellular signals along the edge of the lens occurs with both 0.5 and 1 mm diameter lenses. All procedures were approved by UNC IACUC.

4 Supplementary Figure 4 Position of injector needle relative to lens placement. (a) To reduce damage under the lens, it is recommended that viral injections are not made directly under the lens. Rather, injections should be placed 250 m lateral and ventral to the placement of the lens.

5 Supplementary Figure 5 Visualization of brain tissue and vasculature during lens implant. (a) A 1 mm diameter lens was implanted in the prefrontal cortex of an adult wild type male. Throughout the duration of the lens implant procedure, tissue and vasculature should be visible through the lens. These can be visualized by looking directly through the top of the lens with the naked eye or visualized by centering the surgical microscope above the top of the lens and looking through the microscope objectives. All procedures were approved by UNC IACUC.

6 Supplementary Figure 6 Cell death following long term expression of GCaMP in the somatosensory cortex. In vivo imaging of GCaMP encoded Ca 2+ dynamics (AAVdj-CamkII-GCaMP6f, Titer: 5.6 x 10 12, 1:4 dilution, UNC vector core) was conducted over a 10-min acute recording session (15 Hz, down sampled to 5Hz) 2 (top row) and 12 (bottom row) months following virus injection and lens implantation surgery. (a) Digital images of mean intensity pixel projections across the entire 10-min session (spatial bin of 4). Following 12 months of GCaMP expression (bottom), distinct biological structures are no longer visible. (b,c) PCA/ICA analysis was used to extract the spatial location and activity traces of individual cells (Mosaic Analysis Software, Inscopix). Because spatial or temporal characteristics of GCaMP expressing cells were not identifiable following 12 months of viral expression, independent components (ICs) extracted from this data set could not be sorted by these features. Thus, to compare ICs extracted at each time point, the first 125 ICs were selected from each data set. (b) Merged image of spatial filters for all extracted ICs 2 and 12 months following surgery. At 2 months (top), spatial filters of extracted ICs have morphological characteristics of cells. In contrast, defined anatomical structures are not visible in this same mouse 12 months following surgery (bottom). (c) Similarly, representative activity traces show temporal dynamics characteristic of Ca 2+ signals at 2, but not at 12 months following surgery. All procedures were approved by UNC IACUC.

7 Supplementary Figure 7 Practice baseplate attachment surgery. (a) Secure the dummy microscope in the microscope gripper connected to a micromanipulator. (b) Prepare a glass slide with a dummy head cap. Position the dummy microscope over a glass slide with the microscope placed about the same distance that would be necessary to appropriately attach a baseplate to the head cap of a mouse (~2 mm above the dummy head cap). (c) Attach the dummy microscope to the glass slide in a similar manner that would be used in an actual baseplate surgery.

8 Supplementary Figure 8 Habituation to microscope attachment. (a) Affix a wire cable to the dummy microscope with super glue. (b) Anesthetize mouse and attach dummy microscope to baseplate. When mouse is ambulatory, place the mouse in the test chamber and tether the wire cable to the ceiling. Continue habituation procedures until the mouse can move naturally through the environment while tethered to the dummy microscope.

9 Supplementary Figure 9 Repeated imaging of GCaMP encoded Ca 2+ signals in the somatosensory cortex. In vivo imaging of GCaMP encoded Ca 2+ dynamics (AAVdj-CamkII-GCaMP6f, titer: 5.6 x 10 12, 1:4 dilution, UNC vector core) within the somatosensory cortex of an adult wild type male was monitored over 10-min recording sessions (15 Hz, down sampled to 5Hz). One recording was made per day over a 5 day period. (a) Mean intensity projections across the entire 10-min session. Similar anatomical landmarks could be repeatedly visualized in the imaging data. (b) The spatial location of GCaMP expressing cells was extracted with PCA/ICA analysis (Mosaic Analysis Software, Inscopix). On each day, cells located within similar anatomical space could be visualized. However, the number of cells extracted with PCA/ICA analysis varied across sessions. (c) The extracted spatial maps were saved as individual TIFF files and converted to a TIFF stack in ImageJ. TurboReg motion correction algorithm (ImageJ) was applied to the maps for alignment. (d) Example of spatial maps from recording day 2 (red) and (5) merged on top of each other. Prior to motion correction, many similarly detected cells are misaligned. (e) Application of the TurboReg motion correction algorithm increases the number of aligned cells.

10 Supplementary Figure 10 Pre-processing of digitally acquired Ca 2+ imaging data. While multiple data analysis methods exist to extract independent cellular components, there are general data pre-processing steps that can be applied to in vivo Ca 2+ imaging data sets to facilitate future data analysis steps as well as increase the accuracy of data interpretation. These steps include (a,b) down sampling of the data to increase the speed of data processing as well as (b) the application of a motion correction algorithm, which increases signal detection accuracy. (a) If images contain a greater number of pixels than is required to visualize distinct biological structures, such as cells and vasculature, the data can be spatially binned to reduce the file size and facilitate future data analysis steps. We have found that a spatial bin of 4 works well for most neuron populations. Notice that spatial binning of the data by a factor of 4 does not interfere with the detection of distinct biological structures. (b) If more data samples were acquired than is necessary for the detection of GCaMP encoded Ca 2+ transients, the data can be further down sampled in the temporal domain. For data acquired at 15 Hz, we have found that a temporal bin of 3 (down sample to 5 Hz) adequately reduces the file size while still allowing the detection of individual Ca 2+ transients. (c) It is common for small frame shifts or drift in the focal plane to occur across a recording session. To realign images, and therefore the location of biological structures, a motion correction algorithm can be applied to the data set. The data shown in this paper was aligned via rigid body translation (Mosaic Analysis Software). In brief, this algorithm works by aligning all frames to a reference region (outlined in blue) that was selected from a reference frame in the data set. All frames in the data set (target images) are translated in the x,y plane to align each target image to the reference image on the basis of pixel intensity distribution within the reference region. All procedures were approved by UNC IACUC.

11 Supplementary Figure 11 Data processing information for supplemental video 4 and 6. A rigid body motion correction algorithm was applied to the spatially binned Ca 2+ imaging data shown in video 4 (Mosaic, Inscopix). (a) Reference image used for motion correction algorithm. (b) Region outlined in blue represents the area of the reference frame that was selected as the reference region. A large blood vessel was selected as its spatial location was stable throughout the imaging session and it exhibits a high contrast in pixel intensity compared to the surrounding neural tissue. Note to facilitate selection of the reference region, the contrast was enhanced and the pixel intensity was inverted by multiplying all pixels by a value of -1. (c) Distance in pixels that each frame was translated in the x plane. (d) Distance in pixels that each frame was translated in the y plane. (e) Mean intensity projection (F 0) used to convert the motion corrected and cropped video (Supplemental video 5) to a F/F format. All procedures were approved by UNC IACUC.

12 Supplementary Figure 12 Extraction of individual cellular Ca 2+ signals from high dimension data sets. (a) The photosensor (CMOS sensor) is composed of an array of photodetectors (pixels) that encode the spatial location and intensity of all detected emission signals that are captured during a defined exposure period. Photonic data collected during this exposure period are converted to a digital signal and stored as a TIFF file. (b) Each TIFF (left) file contains information regarding how physiological structures under the lens are distributed in anatomical space as well as the intensity of fluorescence emissions from these structures at a defined point in time. The sequential collection of these images (right) enables changes in fluorescence intensity, and therefore neural activity, to be extrapolated from changes in pixel intensity over time. However, this requires methods to extract meaningful biological signals from high dimensionality Ca 2+ imaging data sets. (c) Example of two analysis methods for the extraction of independent cellular signals. Briefly, region of interest analysis (left) requires the user to manually identify pixels associated with an individual cell. Pixels within this region of interest are then converted to DF/F activity traces. Analysis method on the left shows on example of an automated cell sorting algorithm (PCA/ICA) which extracts the spatial location and activity traces of individual cells based on statistical independence.

13 Supplementary Methods: Analysis of Ca 2+ imaging data with Thunder analysis software written in Python The following analysis script was written in an ipython notebook ( and uses the software package, Thunder 1. While Thunder is ideal for large-scale cluster computing, the script below is written for processing the data using a local machine. It can be modified to run on a remote Amazon EC2 cluster by following the instructions here: We first describe the individual cells in the analysis script/code. Here, the word cell refers to a cell of code in ipython notebook and should not be confused with brain cells. The script is presented after its description. Cell 1: First, we import the required modules for data operations and visualization: 1. matplotlib for plotting 2. seaborn is a statistical data visualization library based on matplotlib (optional) 3. numpy for data operations 4. time for calculating the processing time (optional) 5. mpld3 for enabling zooming in "inline" mode of matplotlib, such as zooming into fluorescence traces for manual sorting (optional) Cell 2: Load images to Thunder. The video recording of Ca 2+ transients needs to be broken down into frame-by-frame images (.tif here) labeled in order. Cell 3: Pre-process the data. First step is to normalize by the mean intensity across time for each pixel. Second step is to subtract the mean across all pixels for each frame/time-point. Cell 4: Run ICA. The example here shows ICA using 300 principal components and 200 independent components. This number should be estimated based on the data set to be analyzed. A rough rule of thumb is to guesstimate the number of cells (neurons) in the data and increase it by for the number of independent components. The number of principal components can be set to be more than the number of independent components. It is important to note that this is just a rule-of-thumb and not a principle rule. Thus, it is always recommended to verify the quality of signal extraction after automated cell sorting. Cell 5: Plot and save (plots) the normalized traces of each ICA as the z-score of the temporal filters after standardizing their sign. This is done since ICA is determined only up to a scaling factor and sign. Here, the sign of the maximum fluctuation is multiplied to the trace so that the maximum fluctuation (likely Ca 2+ transient) is plotted as positive.

14 Cell 6: Plot and save (plots) the spatial filters Cell 7: Save the raw temporal and spatial filters as numpy array files (*.npy) Cell 8: To visualize the spatial filters better, this cell normalizes them by the maximum weight and thresholds this ratio at a given threshold (here 0.5) Cell 9: The above cells (1-8) calculate and plot all the independent components, including their spatial and temporal filters, along with a thresholded image of their spatial filters. After this point, one has to manually select the ICs that are likely representative of Ca 2+ transients (See supplemental figure 8). This step requires visualizing each cell's temporal and spatial filters. We are currently working on a GUI based approach to cell selection. This cell is one way to visualize each IC's spatial and temporal filters in sequence and input whether the IC will be kept (enter 1) or discarded (enter -1). However, a GUI would speed up this selection process #Cell 1: #Set up plotting and import required modules %matplotlib inline import matplotlib.pyplot as plt import seaborn as sns sns.set_context('notebook') from thunder import Colorize #see Thunder documentation image = Colorize.image #This colorizes a matrix for plotting using matplotlib's imshow() import numpy as np import time import mpld3 mpld3.enable_notebook() #Cell 2: dirpath = #set your input directory path here rawdata=tsc.loadimages(dirpath,inputformat='tif',npartitions=1) #Set npartitions=1 for running Thunder on a local computer. If running on a cluster, you don't have to set its value. image(rawdata.first()[1]) #This visualizes the first image. This step is optional and is done just to make sure that the data loaded correctly #Cell 3: normdata = rawdata.totimeseries().normalize(baseline='mean').center(axis=1) #This is a key pre-processing step. The data for each pixel is first normalized by the mean fluorescence across all time-points for that pixel as normdata = (data-

15 mean)/mean. Then, for each time-point, the mean fluorescence across all pixels is subtracted from every pixel. This is the same pre-processing as in Mukamel et al normdata.cache() #Cache the data since it is going to be repeatedly queried during ICA #Cell 4: start_time = time.time() from thunder import ICA modelica = ICA(k=300,c=200).fit(normdata) # Run ICA on normalized data. k=#of principal components, c=#of ICs sns.set_style('darkgrid') #Seaborn plotting option plt.plot(modelica.a); #Plot all the temporal filters print("--- %s seconds ---" % (time.time() - start_time)) #Prints processing time #Cell 5: #Calculate the traces of each ICA. Standarize it by plotting a z-score of the ICA after standardizing the sign of the traces (see below) outputdir = #set your output directory path here for i in range(0,modelica.a.shape[1]): temp = np.sign(modelica.a[:,i].flat[abs(modelica.a[:,i]).argmax()])*modelica.a[:,i] # The above line multiplies each trace by the sign of the entry with maximum absolute magnitude. This is a way to display traces so that the maximum fluctuations (presumably Ca2+ transients) are positive. tempzscore = (temp-np.mean(temp))/np.std(temp) #Calculate z-score p=plt.plot(tempzscore) plt.savefig(outputdir+'ic'+str(i)+'trace_unthresholded.tif') plt.gca().cla() #Cell 6: #Plot absolute magnitude of the spatial filters. Instead of plotting the absolute magnitude, this can also be done by standardizing the sign as above. Here, absolute magnitude is used for ease. imgs = modelica.sigs.pack() #Packs the Thunder spatial filter RDD into a numpy array maps=colorize(cmap='gray',scale=1).transform(abs(imgs)) for i in range(0,modelica.a.shape[1]): image(maps[i,:,:]) plt.savefig(outputdir+'ic'+str(i)+'_unthresholded.tif')

16 #Cell 7: #Save spatial filter and the temporal filter as files in the output directory np.save(outputdir+'icspatialfilter',imgs) np.save(outputdir+'ictemporalfilter',modelica.a) #Cell 8: #This is a spatial filter visualization cell. Here, the ICs are postprocessed to #set a threshold on the spatial filter based on the maximum intensity thresholdtomax=0.5 #The ratio to maximum pixel intensity at which to threshold spatial filters for visualization thresholdedimgs = np.zeros_like(imgs) # Create thresholded images for the spatial filters for i in range(0,imgs.shape[0]): temp = abs(imgs[i,:,:]) tempmax = np.amax(temp) tempscaled = np.divide(temp,tempmax) tempscaled[tempscaled<thresholdtomax]=0 thresholdedimgs[i,:,:] = tempscaled #thresholdedimgs contains all the thresholded spatial filters. Colorizing this array will produce a plot maps=colorize(cmap='gray',scale=1).transform(abs(thresholdedimgs)) for i in range(0,thresholdedimgs.shape[0]): image(maps[i,:,:]) plt.savefig(outputdir+'ic'+str(i)+'_thresholded.tif') #save thresholded images #Cell 9: #The array ICkeepornot stores value 1 at the index corresponding to any selected IC and -1 at the index corresponding to an IC to be discarded. This array can be used to refer to the ICs saved above. temporalfilter = np.load(outputdir+'ictemporalfilter.npy') #Saved above spatialfilter = np.load(outputdir+'icspatialfilter.npy') #Saved above maps=colorize(cmap='gray',scale=1).transform(abs(spatialfilter)) ICkeepornot = np.zeros(temporalfilter.shape[1]) #Array storing whether an IC should be kept (=1) or discarded (=-1) for i in range(0,temporalfilter.shape[1]): temp = np.sign(temporalfilter[:,i].flat[abs(temporalfilter[:,i]).argmax()])*temporalfilter[:,i]

17 #standardize sign as above tempzscore = (temp-np.mean(temp))/np.std(temp) plt.subplot(2, 1, 2) plt.plot(tempzscore) #Plot z-score of temporal filter as subplot-2 plt.subplot(2, 1, 1) #Plot absolute values of unthresholded spatial filter as subplot-1 image(maps[i,:,:]) plt.show() ICkeepornot[i] = raw_input("press 1 to keep or -1 to discard") plt.gca().cla() Instructions for analyzing spatially defined Ca 2+ transients in relation to discrete behaviorally or environmental events in MATLAB Overview: The provided GUI is designed to run stably on MATLAB R2015a. The GUI offers an easy method of presenting Ca 2+ imaging data aligned to instances of session variables. Both raw and averaged data are presented, and if elected for, a cell map will be created in a separate third window that will spatially diagram each cell s location along with its response profile in color. In the following sections, we briefly describe what each MATLAB file accomplishes in the general order each function is called.

18 To begin an analysis session, all GUI related files provided here must be added to the MATLAB search path. This is done by command line addpath(<directory of all GUI files>) or MATLAB interface in the Current folder window in MATLAB, navigate to and right-click folder with all GUI files then select Add to path > Selected Folders and Subfolders. Run sdgui.m to begin type sdgui in the command window. Note: some of the functions described below may not work well with other versions of MATLAB so it is recommended that the suggested version of MATLAB is installed on a Windows computer before beginning data analysis. sdgui.m: creates the main interface for the user to load data into and define parameters for visualizing Ca 2+ imaging data. The first step is to import extracted s.d. traces from Ca 2+ imaging data (i.e., cellular responses) by selecting the Load button at the top left (Figure 1A). A check mark will appear once the data has finished loading. Data must be formatted such that the first column stores timestamps of imaging frames while subsequent columns represent fluorescence measures for an individual cell or ROI. Once complete, session variables (i.e., behavioral data) must be loaded. There are two options to import data. The Load button (Figure 1B) imports variables from a.mat file. Variables must be either numerical vectors or matrices. Matrices are split into individual vectors by columns and named according to matrix name with an index appended to the end. The Add button (Figure 1C) allows the user to input a MATLAB expression that will be evaluated in the base workspace. The expression must evaluate to a numerical vector. Note: If variable data is stored in another format such as.txt or.xls file convert the data to a.mat file before loading it into the GUI. This can be done by creating a new array in MATLAB labeled with the variable name and pasting in the timestamps associated with the variables of interest. Create an array for each variable and save all MATLAB variables into a single.mat file. The next step is to choose one of the added session variables as the variable to align Ca 2+ imaging data to. This is done by selecting the variable in the dropdown list (Figure 1D) at the top-right of the window. The time window around the align variable to visualize is defined in the pre and post window fields (Figure 1E). Data can also be smoothed (Figure 1F) using a sliding window average. The value in the text box defines the size of the sliding window, i.e., number of time points to average. To either create a map displaying the intensity of cellular responses organized by their spatial location (Figure 1G) or sort the cells from lowest to highest s.d. values centered around your variable of interest (Figure 1H), the start and end of a response window relative to the align variable must be defined in the two boxes provided (Figure 1I). The duration of the time window should be determined by the

19 duration of the behavioral variable or stimulus of interest as well as the binding kinetics of the Ca2+ indicator. Either the maximum or average response value (Figure 1J) will be defined with this response window. Cell map creation needs a user-defined directory (Figure 1G) with image files for each cell in the Ca 2+ imaging dataset (i.e., a.tif file indicating the spatial filter for each cell). Note: Spatial filters for individual cells can be obtained with the Python code provided here or with Mosaic analysis software. Data shown here is from spatial filters obtained with Mosaic analysis software. Finally, assign a label for the dataset and select Align (Figure 1K) when parameters are set. Either 2 or 3 figure windows will pop up displaying the average response of each cell aligned to the user defined variable, raw and average data for each cell aligned to the defined variable, and, if selected, a cell map indicating the mean or max s.d. value for each cell during the defined time window. Figure 1 sdalign.m: extracts and arranges data within the user-defined window around the align variable. Code here will arrange Ca 2+ imaging data into a three-dimensional array trials x time (window centered around align variable) x cells or ROIs. Average responses are calculated across trials to give a time x cells array with error calculated as SEM. The color axis for color plots generated in subsequent functions is determined by the range of responses so that the color axis covers the entire span of response values.

20 sdavg.m: creates figure to display average Ca 2+ imaging data for each cell aligned to chosen variable. The averaged response for each cell to the user-defined align variable is displayed in the color plot at the top of the figure. The response for a particular cell is calculated by averaging the aligned trace for each trial. Rows represent individual cells, and time relative to the align variable is represented across the x-axis (Figure 2). The bar graph at the bottom of the figure depicts the probability that the other session variables, i.e., variables other than the user-defined align variable, will occur within the windows before and after the occurrence of the user-defined align variable (Figure 2). Figure 2 sdbrowse.m: creates figure and interface to display raw data from each trial for each cell (Figure 3). This allows the user to browse through responses to the userdefined align variable for each trial. The user can interact with the scroll bars (Figure 3A & C) at the bottom to define specific cells and trials to focus data displayed in this figure. The first scroll bar (Figure 3A) selects the individual trial to focus, i.e., which instance of the userdefined align variable to focus. Alternatively, the trial number can be entered into the text box on the left (Figure 3B). Individual cells can similarly be chosen to focus (Figure 3C). One feature to note is the cell rank (Figure 3D) vs cell ID (Figure 3E).

21 Cell ID refers to the order within the original dataset loaded into the GUI. Cell rank refers to the order after the GUI sorts the data based on individual cells responses. If sorting was not chosen, cell rank and cell ID are identical. The scroll bar is ordered by cell rank. The top left color plot displays the response from each trial for the user-defined cell. The top right color plot displays the response from each cell for the user-defined trial. Color represents magnitude of response in both plots. The two traces at the bottom of the figure display response data from the userdefined cell. The top trace is the response for an individual cell to the user-defined trial. The bottom trace is the averaged response for an individual cell across all trials ± SEM. Figure 3 alignmap.m: This optional third function illustrates the spatial arrangement in the x,y-plane of all cells by constructing a visual map of cells with a response value for each cell coded in color. Individual cells are drawn based on files in the given directory with a median filter applied. A directory of image files for each cell must be defined. A response window must also be set. The number of files should equal the number of cells and ordered (by filename) exactly as in the data. Files should be named starting with ic followed by an index (e.g., ic.tif, ic2.tif ). Pixel-distance scale in the figure is identical to that of files provided in the user-defined directory.

22 The response profile is calculated as either the average or maximum s.d. response for each cell aligned to the user-defined variable across the user-defined time window. Figure 4 1. Freeman, J. et al. Mapping brain activity at scale with cluster computing. Nat. Methods 11, (2014).