We need insertion systems that can be rapidly reconfigured to target arbitrary combinations of a dozen or more brain regions day after day. Our modular insertion system is based on a set of arcs that can hold up to 24 Neuropixels at any angle relative to the brain surface. Each probe is mounted on a compact 3-axis manipulator (New Scale Technologies) and the entire system of arcs can be raised and lowered to facilitate rapid transitions between experiments. The complete system can be purchased from New Scale or manufactured from freely available design files. To complement the hardware, we are developing machine vision software, called Parallax, that can automatically detect the 3D location of individual probes and calibrate their motion to a global coordinate system.

Standard methods for electrode placement, based on skull landmarks, are too variable and imprecise to target small brain structures reliably. To make multi-probe electrophysiology experiments truly reproducible, probes must be placed within ~200 micrometers of their targets. We are developing procedures based on high-resolution MRI imaging to improve targeting accuracy (collaborating with Donghoon Lee in the UW Department of Radiology). Mice are implanted with a zirconia headframe that includes precisely machined fiducial holes. Using a 14T magnet and a manganese-based contrast agent, we routinely obtain scans with 100 micrometer isotropic resolution. We use scans from individual mice to guide Neuropixels probes to multiple connected brain regions.

In addition to the protocols for anatomy-guided electrophysiology, we are also developing open resources for aligning individual brains to the Allen Common Coordinate Framework (CCF) and vice versa. These include CCF-registered template volumes for light sheet imaging, a transform between the Paxinos mouse atlas and the CCF, and an extension to the Pinpoint trajectory planning app for visualizing individual brains in 3D.

Advancing our understanding of the function of neural circuits will depend critically on exploring how various cell types interact at the level of the whole brain during behavior. However, the vast majority of electrophysiology datasets lack information about the cell types that were recorded. To address this, we use “optotagging” to obtain the ground truth spiking activity of a variety of genetically defined cell types, then train machine learning models to classify cells based on their electrophysiological properties alone. The resulting “Cell Type Lookup Table” would dramatically improve our ability to connect electrophysiology experiments to the vast knowledge of cell types being generated by the Allen Institute for Brain Science and elsewhere.

Reproducible ephys requires rapid feedback about the quality of data in each experiment, along with a complete record of all code that was used for analysis. We are building a pipeline around the SpikeInterface Python package that is automatically triggered after each experiment, and which can be easily re-run with new spike sorting algorithms. Our pipeline was optimized for use within the Code Ocean computing platform, but it can also be deployed on a single machine or a local cluster. This pipeline has made it much easier to benchmark tools for preprocessing, spike sorting, and curation. For example, we compared the performance of a wide range of compression algorithms on Neuropixels data and found that audio codecs result in the smallest file sizes with no loss of information.

Our pipeline encompasses all steps of spike sorting. It starts by reading raw Neuropixels data stored in Open Ephys, SpikeGLX, or NWB formats followed by low-level signal processing. The output of each processing step can be checked via web-based visualizations hosted by Figurl. Next, spike sorting is performed by any algorithm supported by SpikeInterface. Switching sorters is as simple as changing one line of code; no additional installation is necessary. The outputs of the sorter are then curated using standard quality metrics and automated classifiers trained on human annotations. Outputs can be manually refined via sortingview, or exported to Phy format for offline exploration. Instructions for using the pipeline with your own data can be found on GitHub.

The reliable interpretation of spiking data depends on precisely registering each recorded neuron to an anatomical atlas. This involves coating each probe in dye and tracing the fluorescent tracks they leave behind in histological sections. A variety of tools exist for annotating probe tracks offline, but we needed a cloud-friendly method to complement our spike sorting pipeline. We based our workflow on two powerful open-source tools, the Ephys Alignment App (International Brain Lab) and Neuroglancer (Google). These tools can now be used to identify probe tracks and align them to electrophysiological landmarks without leaving a cloud computing environment.

Our team is supporting and extending a widely used open-source application for acquiring multichannel electrophysiology data. The software was designed around a plugin architecture that makes it easy for scientists to add new functionality. The Open Ephys GUI works seamlessly with all types of Neuropixels, and includes unique visualizations optimized for high-density linear probes. Current efforts are focused on building new plugins that make it easier to run experiments that combine electrophysiology with optical stimulation.