|
Scientific Understanding of Consciousness |
|
Cortex Timescales of Population Coding
Nature 548, 92–96 (03 August 2017) Distinct timescales of population coding across cortex Caroline A. Runyan, et.al. Department of Neurobiology, Harvard Medical School, Boston, Massachusetts 02115, USA Neural Computation Laboratory, Istituto Italiano di Tecnologia, 38068 Rovereto, Italy [paraphrase] The cortex represents information across widely varying timescales. For instance, sensory cortex encodes stimuli that fluctuate over few tens of milliseconds, whereas in association cortex behavioural choices can require the maintenance of information over seconds. However, it remains poorly understood whether diverse timescales result mostly from features intrinsic to individual neurons or from neuronal population activity. This question remains unanswered, because the timescales of coding in populations of neurons have not been studied extensively, and population codes have not been compared systematically across cortical regions. Here we show that population codes can be essential to achieve long coding timescales. Furthermore, we find that the properties of population codes differ between sensory and association cortices. We compared coding for sensory stimuli and behavioural choices in auditory cortex and posterior parietal cortex as mice performed a sound localization task. Auditory stimulus information was stronger in auditory cortex than in posterior parietal cortex, and both regions contained choice information. Although auditory cortex and posterior parietal cortex coded information by tiling in time neurons that were transiently informative for approximately 200 milliseconds, the areas had major differences in functional coupling between neurons, measured as activity correlations that could not be explained by task events. Coupling among posterior parietal cortex neurons was strong and extended over long time lags, whereas coupling among auditory cortex neurons was weak and short-lived. Stronger coupling in posterior parietal cortex led to a population code with long timescales and a representation of choice that remained consistent for approximately 1 second. In contrast, auditory cortex had a code with rapid fluctuations in stimulus and choice information over hundreds of milliseconds. Our results reveal that population codes differ across cortex and that coupling is a variable property of cortical populations that affects the timescale of information coding and the accuracy of behaviour. The goal of this work was to compare coding across cortical regions for two key features of behavioural tasks: stimulus and choice. We developed a sound localization task in which mice reported perceptual decisions by navigating through a visual virtual reality T-maze. As mice ran down the T-stem, a sound cue was played from one of eight possible locations in head-centred, real-world coordinates. Mice reported whether the sound originated from their left or right by turning in that direction at the T-intersection. We focused on auditory cortex (AC), because it is necessary for sound localization tasks and on posterior parietal cortex (PPC), because it is involved in spatial auditory processing, receives inputs from AC, is a multisensory-motor interface, and is essential for virtual-navigation tasks. In each mouse, we imaged the activity of ~50 neurons simultaneously from AC and PPC on separate days. In both regions, neurons were transiently active at different time points, resulting in a population that tiled the trial. Activity in some AC neurons was selective for stimulus location; however, as a population, AC activity was heterogeneous and complex. In PPC, a substantial fraction of neurons had different activity on trials with opposite behavioural choices. The heterogeneity of activity patterns suggested that, in addition to stimulus and choice, multiple task-related variables, such as visual inputs, reward delivery, and the mouse’s running, might affect neuronal responses. To take these variables into account and to help isolate signals related to stimulus and choice, we developed an encoding model (generalized linear model, GLM). This model incorporated all measured task-related variables as predictors of each neuron’s activity. The model reliably predicted the time course and selectivity of single-neuron activity in AC and PPC. To determine whether stimulus and choice information were present in AC and PPC, we decoded the most likely stimulus category (left or right location) or choice (left or right turn) from neuronal activity by using Bayes’ rule to invert the prediction of the encoding model. Because stimulus locations and choices were related to one another by task design (for example, left stimuli required a left turn for reward), we decoupled these features and isolated information purely related to stimulus from information purely related to choice by selecting equal numbers of right and left choice trials for analysis in each stimulus condition (thus the same number of correct and error trials). Decoding performance was calculated as the mutual information between the decoded and actual stimulus category or choice. Pure information about the stimulus category was present in AC activity but was weak in PPC. Information purely about the choice was present in both AC and PPC populations. We investigated the codes for stimulus and choice information first by examining activity in single neurons. In AC, we considered both stimulus and choice information, whereas, in PPC, we focused on choice information only, because PPC contained little pure stimulus information. Stimulus and choice information were small but significant in individual neurons, on average, and only a minority of neurons had large stimulus or choice information. In both areas, individual neurons were briefly informative, with subsets of largely distinct neurons providing information at different time points in a triail. Single cells in AC and PPC were informative about the choice for ~100 and 300 ms, respectively, and individual AC cells were informative about the stimulus category for ~280 ms (calculated as a two-sided decay around the information peak of each cell; Therefore, in most individual neurons, information was weak and short-lived relative to the length of trials. Together our results reveal that, despite short coding timescales in individual neurons, long timescales can emerge in neuronal populations. However, coding timescales were variable across cortex and depended on the structure of the population code. AC had relatively weak coupling and a short timescale (hundreds of milliseconds), which might aid representations of rapidly fluctuating stimuli and high dimensional sensory features. Previous studies have proposed that noise correlations can have a detrimental, information-limiting effect and have thus suggested that sensory codes may benefit from weak coupling, which appears consistent with our findings in AC. However, in contrast, PPC had strong coupling and a long population timescale (~1 s), which appear to have a beneficial effect, because higher levels of coupling and temporal information consistency corresponded to more accurate task performance. In PPC, coupling timescales could be long enough to combine temporally separate inputs and could result in higher instantaneous information, because of information accumulation over time. We built a data-driven computational model that confirmed this effect of coupling. Further, from such a code, a downstream network could instantaneously read out a signal containing consistent and accumulated information about the recent estimate of the appropriate choice. Our model showed how such a temporally consistent choice signal could improve behavioural accuracy. We propose that codes underlying sensory representations and choice signals might differ substantially and that the structure of population codes may be a defining characteristic of cortical regions that contributes to the computations performed in each area. [end of paraphrase]
Return to — Reentry and Recursion |