Neural encoding of time: the striatum vs prefrontal cortex

The neural mechanisms of encoding timing are still controversial. According to one prominent hypothesis, time is encoded in local network dynamics – see a previous blog post dedicated to this issue. However, similar mechanisms (“population clocks”) have been linked to multiple areas across the brain, including the striatum, prefrontal and parietal cortices, and the hippocampus. Does this variety of brain regions reflect a specialisation of each area to track time at e.g. a different scale, or is time encoded in parallel in several regions?

To answer this question, Bakhurin et al. quantified and compared the degree of time encoding in two areas: the striatum and the orbitofrontal cortex (OFC). They acquired electrophysiological recordings in mice conditioned to receive a food reward (condensed milk) after a specific interval (2.5s) following an olfactory cue. Activity in both regions was measured simultaneously in 6 animals; in 5 further animals, only activity in the striatum was recorded. Spike sorting was used to isolate activity in single neurons (pyramidal cells in OFC and medium spiny neurons in the striatum). To quantify the degree to which each region tracks time, the authors used multivariate decoding – a multi-class support vector machine classifier, based on firing rates of multiple units – to estimate the elapsed time from neural activity. Ideally, feeding the data acquired e.g. 1s after the olfactory cue into the decoder would result in a correct estimation that 1s has elapsed since the cue. Using this technique, one can quantify whether neural activity in a given area is a better predictor of the actually elapsed time than neural activity in another area.

The results of this and several control analyses show that time can be decoded with higher fidelity from striatal activity than from prefrontal activity. This pattern of results – the striatum outperforming the OFC as a neural basis for decoding time – was robust and did not qualitatively change when using more or less neurons in each area; selecting units in the dorsal or ventral striatum, or in the medial or lateral OFC; or controlling for motor activity (animals licking in anticipation of the reward). These findings are interpreted by the authors in terms of the striatum providing a refined readout of upstream cortical activity. Thus, the striatum might outperform the OFC in encoding time per se. However, as the authors also note, neural activity in the OFC has a higher dimensionality than in the striatum (i.e., more principal components are needed to explain its variability). This might be due to the OFC encoding more task variables than the striatum, as suggested by the authors; however, it can also be explained by a higher anatomical or physiological variability, or a lower signal-to-noise ratio, in the OFC. Thus, it would have been beneficial for the study to include a task variable – perhaps reward accumulation over several trials – for which prefrontal activity would plausibly yield better decoding than striatal activity.

While the study shows differences in decoding performance between the two regions, it rarely addresses the question whether time encoding mechanisms are qualitatively similar or distinct between the two regions. The one finding that does suggest differences in how time is encoded by the two regions shows that motor responses distort time encoding more in the striatum than in the OFC. Specifically, training the decoder on trials in which animals displayed licking behaviour early on (first tercile) or relatively late (third tercile) induced systematic biases when the decoder was tested on the remaining trials (second tercile). Thus, in the striatum, motor responses seem to warp time encoding in the opposite directions: early motor response speed up in estimated time, while late motor responses induced delays in estimated time*. These effects were less pronounced in the OFC. In fact, early prefrontal activity seemed to be especially robust to any interference from motor responses.

Taken together, the paper shows that decoding elapsed time is overall more accurate based on striatal activity than on prefrontal activity – however, why this is the case remains an open question. On the other hand, striatal time-encoding activity might to some extent covary with motor-encoding activity. This co-dependency of time and motor encoding is weaker in the prefrontal cortex, suggesting intriguing qualitative dissociations between the neural mechanisms of time encoding in different regions. Previously, decoding based on different data modalities (MEG and fMRI) was used to find correlations and dissociations between decoding-enabling data features (e.g., early response latencies in the MEG and sensory regions in the fMRI). Perhaps future studies could use a similar approach to find whether time representation in one brain region can generalise to another region, suggesting shared mechanisms, or whether time encoding is subserved by neural mechanisms unique to each region.

Ryszard Auksztulewicz, Oxford Centre for Human Brain Activity 

Source article: Bakhurin KI, Goudar V, Shobe JL, Claar LD, Buonomano DV, Masmanidis SC (2016) Differential encoding of time by prefrontal and striatal network dynamics. J Neurosci, December 15, 1789-16. doi: 10.1523/JNEUROSCI.1789-16.2016

* In my original post, based on the published article, the sentence stated the opposite: “early motor response induce delays in estimated time, while late motor responses speed up estimated time”. However the Authors have asked me to correct this sentence according to their original intention, and have requested a correction in the journal article.