hide
Free keywords:
-
Abstract:
Understanding the mechanisms of learning and memory consolidation requires characterizing how the response properties of individual neurons and interactions across populations of neurons change over time, during periods spanning multiple days.
We used multiple chronically implanted tetrodes to record single unit activity from area V1 of the awake, behaving macaque and developed a method to quantitatively determine recording stability. Our method is based on a statistical framework which uses similarity of action potential waveforms to detect stable recordings given a pre-defined type I error rate. The similarity measure that was used takes into account both the shape of the action potential waveform and the amplitude ratio across channels, which depends on the location of the neuron relative to the tetrode.
271 well-isolated single units were recorded from 7 tetrodes during two periods of up to 23 days. We computed the distribution of pairwise similarities of average waveforms recorded on consecutive recording sessions during the first 34 days after implantation of the chronic drive. During this period, there was no recording stability due to regular adjustments of the tetrodes. We used this distribution as an empirical null distribution for hypothesis testing.
Using this statistical procedure and a type I error rate of alpha = 0.05, we find that of all single units recorded on a given day, 51 could be recorded for at least 2 days, 40 for at least 3 days, and 25 for at least 7 days.
In addition, we adapted a recently proposed multivariate statistical test (Gretton et al., 2007) to test whether the waveforms obtained at consecutive days come from the same underlying probability distribution. Using this test we obtained qualitatively similar results.
To validate these results, we compared orientation tuning functions of neurons that were tracked across days. Consistent with the claim that the same neurons were recorded across days and the fact that the monkey was not performing a learning task, the distribution of tuning differences of stable and orientation-tuned neurons across days was highly significantly different (Wilcoxon rank sum test, n1 = 79, n2 = 582, p < 10^-34) from the distribution of tuning differences across different neurons.
Our results show that using only waveform information it is possible to reliably track stable neurons across days with a limited type I error probability. This statistical approach is particularly important since, in a learning experiment, properties of neurons such as orientation tuning are potentially changed and therefore cannot be used to evaluate stability.
