3 Spike sorting

We’ll cover how to spike sort using two programs: 1) Spike2 (written by Tony Lapsansky) and 2) Neuralynx (written by Eric Press).

The function of spike sorting is to isolate action potentials from the background voltage signal. These methods use the shape of the waveform to detect and distinguish the spiking activity of each neuron recorded by an electrode.

3.1 Spike2

Written by Tony Lapsansky, February 24, 2023

These instructions assume that you have been given a Spike2 recording file (extension .smrx) and asked to spike sort.

Spike2 includes a detailed description of the program, accessible by clicking Help → Index

Spike2

3.1.1 File naming conventions:

  • Use the name structure YEARMODA_sequence_investigator

  • Save data in the corresponding directory “C:\InvestigatorName\ephys\YEAR-MO-DA”

3.1.2 Spike sorting with Spike2

  1. Open the main Spike2 file for the recording. This file should have the extension .smrx.
  2. Apply a digital high pass filter, if needed. Note: if the data were collected with the high pass filter set at greater than 100 Hz (no LFP signal) then proceed to step 3.
    • Right click on the raw data channel (typically Ch1) and select FIR Digital Filters…. We want to use an FIR filter rather than an IIR filter as the latter can introduce a variable time lag in the resulting data (see Spike 2 Help → Index → Digital Filter for full explanation).
    • Under the pull down menu for Filter, change the filter from Example low pass filter to Example high pass filter.
    • Select the Show Details button in the bottom right.
    • Adjust blue slider change the filter length. Shift the slider until the coloured dots above the slider from red to yellow to green. This removes wobbles in the data. Use the minimum level (~1019) to achieve green. Fine adjustments can be made just under the slider. FIR Filtering
    • Hit Apply
    • Set Destination to the next available channel (typically Channel 4)
    • Click Okay
    • Close the filtering window. You are given the option to save the filter. This is unnecessary.
  3. Setting the threshold for spike identification
    • Right click on the filtered channel and select New WaveMark
    • Clear previous templates if any are present. To do so, select the trash can icon within each template. These may be present from a previous session.
    • Locate your cursor position, indicated by the vertical dashed line in the main window (typically found at time 0)
    • Slide the dashed line horizontally through the trace to observe potential spikes as determined by the default upper and lower thresholds.
    • Right click the upper bound marker (the upper horizontal dashed line in the WaveMark window) and select Move Away. We will rely on the lower bound to identify spikes for sorting, as the activity above baseline is typically closer in magnitude to the background.
    • Slide the dashed line horizontally through the trace to observe potential spikes as determined by the lower threshold alone.
    • Adjust the lower threshold to catch spikes of interest. This threshold will vary based based on the distance between the electrode and the neuron, the quality of the isolate, and the level of background noise. Values between 50 mV and 200 mV are typical.Set the lower bound so that spikes of interest are included and ambiguous spikes are excluded.
  4. Designing the spike template
    • Move the cursor to a characteristic spike. In the upper window, you will see the provisional template. Click and hold on the trace in the upper window and drag it to the first available spot in the lower, template window.
    • To set parameters for spike sorting, click on the button just to the left of the trash can icon (on the top half, upper right of the WaveMark window). This is the “parameters dialog” button. This opens a template settings window.
    • For the line Maximum amplitude change for a match enter a value between 10 and 20. This will allow a spike that fits a template to vary in maximum amplitude by up to 10-20%.
    • For the line Remove the DC offset before template matching, confirm that the box is checked. This means that Spike2 will account for abrupt shifts in the signal baseline before template matching. This is a stop-gap for any issues with the digital high pass filter.
    • Click OK. Spike template design
  5. Spike sorting
    • Back in the WaveMark window, make sure that the box Circular replay is unchecked. If checked, spike sorting will loop indefinitely.
    • Ensure that the vertical cursor on the main window is at time zero (or the first spike) so that no spikes are missed.
    • Back in the WaveMark window, make sure that the box Make templates is checked. If unchecked, only spikes corresponding to the provisional template will be identified. We want to let spike2 help us to identify potential multi-unit activity.
    • Hit the play button ▶️, which is called “run forward”. Spike sorting will proceed for several minutes. Each identified spike will appear briefly in the WaveMark window and will be assigned to a template. Spike sorting In this image, I have selected options for Overdraw and Show template limits
  6. Merge, delete, and save templates
    • After spike sorting has completed, select New Channel on the WaveMark window to place the spike sorted data in the next available channel (typically, Channel 5)
    • Close the existing WaveMark window.
    • Right click on the spike sorted channel and select Edit WaveMark.
    • Within the WaveMark window, go the pull down menu Analyse and select Principal components. Select OK. This opens a window containing a principal component analysis of all spikes colored by their assigned template.
    • Rotate around all three axes to determine if there is one, two, or more clusters. In theory, each cluster corresponds to a single neuron. Often, spikes are categorized into multiple templates, but realistically correspond to the activity of a single neuron.
    • Identify templates that should be deleted and those that should be merged. We will delete spikes corresponding to templates that are sparse and peripheral.
    • Delete the template(s) in the WaveMark window by selecting that template’s trash can icon.
    • Merge templates by dragging them into the same window
    • Hit the reclassify button in the WaveMark window to commit these changes to the data in the main window. Spike inspecting In this example, we have good evidence from the PCA to merge these five templates.
  7. Export the spike-sorted data
    • File → Export As
    • Select .mat (matlab data)
    • Use the same filename and location but with the .mat extension.
    • Hit Save
    • Select Add for All Channels
    • Click Export
    • Click OK (this will take several minutes)

Note: May need to select an earlier MATLAB file convention to work with R.

3.2 Neuralynx

Written by Eric Press, November 11, 2022

  1. Spike sorting database:
    1. Check the column labelled Sorting status to find days of recording that are cued meaning they are ready to be sorted. Recordings are cued for spike sorting once information about the recording has been added to the database. This includes observations from the day’s recording, whether the electrode position was moved from the previous recording, and the stimulus condition for each recording. The recordings are stored at the following location and are named/organized by date and time of recording:
      Computer/LaCie (D:)/Eric’s data/nlx_recordings
  2. Filtering the raw traces (CSCs):
    1. Use the NlxCSCFiltering tool on any Windows machine to run a band-pass filter on input CSC files.
    2. Choose all the CSC files for a given recording, change the PreAppend field to spfilt, which stands for spike-filtered and adjust the DSP filtering fields to match the image to the right. This selects for frequencies in the raw traces where spikes will be found, but removes low frequency (LFP) and high frequency components of the traces. Nix csc filter
  3. Examine the filtered traces:
    1. Take a closer look at the filtered traces (Open in Neuraview on any Windows machine) and determine which channels are likely to have isolatable spikes and how many distinct spikes there might be. It helps to keep Neuraview open when setting thresholds in the next step.
  4. Spike detection from filtered traces:
    1. Use the CSCSpikeExtractor tool on any Windows machine to detect spikes above or below a given µV) threshold. The units displayed in the program will be AdBitVolts which are simply 10.92x from the µV value.
    2. Based on the filtered traces, within CSCSpikeExtractor, set the spike extraction properties (Spike Extraction -> Properties OR Ctrl+P) as shown above. The Extraction Value is set to 10.92x the µV you chose by viewing the filtered traces.
    3. Press Ctrl+S to extract spikes from the selected file at the desired settings. The resulting file will be placed in the extracted spikes filter on the Desktop.
    4. Create subfolders in the recording folder for each threshold and move the extracted spikes at each threshold into the appropriate folder. These spike-detected files will be used for spike sorting in the next step.
    5. If it helps with detecting real spike waveforms while eliminating noise, run recordings through spike detection at multiple threshold (positive or negative) such that only all putative neurons are accounted for a minimal noise is detected. Spike extraction properties
  5. Spike sorting:
    1. Open the extracted spikes in Spikesort3D on either the Neuralynx machine or another Windows machine that has an active SpikeSort3D licence. You can also use TeamViewer to control the Neuralynx machine but this works much better with another Windows machine.
    2. Press OK when the feature selection window appears. If you want to select alternate features to display, select them from the list provided. Sometimes it can be helpful to use PCA1 – 3 in isolating neurons but often it makes things more challenging.
    3. Using the 3D Plot, examine the clustering of spikes. Follow the image below to aid in interacting with the 3D plot (MB = the scroll wheel button i.e. middle mouse button). You can change the features displayed on each axis with Q/W, A/S, and Z/X respectively. Also, Ctrl+P brings up a window that allows you to change the size and opacity of points on the plot (I find size = 2, alpha = 0.5 works well to improve visual definition of the clusters). If distinct clusters are difficult to see, find the combination of 3 features that produces the most noticeable clustering or the greatest spread of points in the space. The features displayed in the 3D plot are shown at the top left of the plot (i.e. X(3) Height # # # #). Use those features for the next step. 3D plot movement and interaction
    4. Run KlustaKwik (Cluster → Autocluster using KlustaKwik) and select the 3 features that generate the most clearly separable clusters on the 3D view – often, the first 3 (Peak, Valley, Energy) do a decent job. Change the MaxPossibleClusters to 10 before pressing Run. The remaining settings should match the image below. KlustaKwik interface
    5. Following calculations, use the Waveform window and the 3D plot to group the distinct clusters into what you believe are waveforms produced by distinct neurons. Use the number keys to highlight distinct clusters and Ctrl+M to merge clusters together. Ctrl+C copies the selected cluster and can be used to split a cluster into 2 if you believe portions of the cluster belong to distinct putative neurons. This step takes some practice. You can use Ctrl+Z to undo only one move. Otherwise, you may need to exit without saving and start again at step 4. Save with Ctrl+S often and click OK to overwrite the file.
    6. Once you are satisfied with the waveforms left, note how many there are, and whether it seems possible that some of the groups belong to the same neuron. Consider what you know about excitable membranes to make these decisions. Fill out the Spike Sorting Database with the information used to reach this point. This includes, the threshold(s), # of clusters, # of putative neurons (often 1 less than the # of clusters because it would be a stretch to include the smallest amplitude waveform as a distinct, separable neuron), and any else to note from performing sorting.
    7. Save each cluster to its own spike file (File → Save Multiple Spike Files)
    8. Open the separate spike files you just created, along with the original filtered trace in Neuraview. Scroll along the recording and examine if the sorting you performed seems believable. Do the spikes in different rows really seem like they’re different in the filtered trace? Do some spikes not seem like real spikes? If anything seems amiss, make the appropriate merges in SpikeSort3D before proceding.
    9. Export the relevant data from the sorting. Perform the following:
      1. File → Save ASCII Timestamp Files
      2. File → Save Multiple Spike Files
      3. File → Save ASCII Avg Waveforms
      4. Also, save the file itself with Ctrl+S
    10. Lastly, bring up all the waveforms together on the waveform plot. Take a screenshot and save it to the folder where the extracted spikes (and now timestamps files) are stored.
  6. Moving sorted files to other locations:
    1. Once a chunk of recordings have been sorted, copy/paste the entire recording file to Eric’s orange 1TB storage drive (Lacie). Place them in the following folder: Eric's data/sorted_recordings