Recording from the same cortical neurons over months with Neuropixels probes

For many studies, especially those focusing on learning and plasticity, it is important to track the activity of neurons over long periods. Prolonged recordings can be achieved with two-photon imaging, but this method is difficult in deep brain structures and has limited temporal resolution. Here we asked whether neurons could be reliably tracked across recordings with Neuropixels probes, which allow access not only to the cortex but also to deep brain structures and can resolve individual spikes.

We chronically implanted mice with prototype version 2.0 Neuropixels probes. They can have one shank or four shanks, and have tightly spaced recording sites aligned in two columns, as shown in the methods section. Neuropixels devices can be carried by mice for months without any observed problems. We implanted the probes vertically through primary visual cortex and cemented them to the cranium. For recordings, mice were head-fixed in front of computer screens. Recordings were performed at regular intervals for more than half a year. To characterize the neurons, during each recording we presented a library of natural images. Recordings yielded many neurons across more than 6 months as demonstrated in section 1, but there was substantial drift of the probe relative to the brain tissue across sessions. To compensate for it, we used a specially adapted version of Kilosort2 spike-sorting program that detects and corrects for such drifts in spliced recordings. Many V1 neurons responded to the presented images and showed a pattern of preferences that constituted a unique signature, as shown in section two. Figure 5 shows responses of all selective neurons on a day of recording and a consecutive day. They look similar, but how similar are they? In order to quantify that, we took pairs of recordings and spike sorted them together. For each signature on the first day, we find the signature from the second day that correlates with it the most. For most responsive neurons, it is the same neuron’s signature, because most dots on figure 6 are situated on the diagonal. This is true across days, weeks and even months. We are able to track significantly higher percentage of neurons than previous state-of-the-art methods. Moreover, our method, unlike previous ones, does not require continuous recording. To conclude, the new Neuropixels probe, whose geometry is particularly helpful for tracking drifts, and the adapted Kilosort 2 algorithm allow to track same cortical neurons over the course of days, weeks and months. The ability to follow large populations of individual neurons over extended periods opens new possibilities in the study of learning and plasticity.