My name is Andrew Yang, I'm one of the residents in Neurosurgery at Penn, and I'll be speaking about some of our results looking into the mechanisms of working memory in humans. There's a wealth of evidence from behavioral psychology to suggest that on average we can only hold up to four chunks of information in our short-term memory. In fact, this capacity is fundamental to cognition, as individual variability is highly correlated with intelligence. However, which aspects of our neural hardware or software determines our working memory capacity is not well understood. There's been much interest in neural oscillations, which can be accorded with techniques such as EEG, electrodes on the surface of the brain, and depth electrodes. These oscillations demonstrate specific temporal relations with neuronal spiking, and are thought to have a functional role in coordinating activity of neuronal populations or ensembles. In particular, an influential class of working memory models propose that the four-item working memory capacity may be determined by the nature of interactions between two specific classes of oscillations. A slower rhythm in the theta frequency range, and a faster oscillation in the gamma frequency range. In these models, each item of information, labeled A, B, C, and D in this cartoon, is represented within a gamma cycle. And these gamma cycles aren't in turn associated with specific positions or phases of the underlying theta oscillations. Hence, the model predicts that working memory capacity is in part determined by how many gamma cycles can be reliably nested into a theta oscillation. There are two assumptions of this model that are worth going over. The first assumption is that the functional unit within the brain for representation of items, whether that item is a sensory perception or a concept, are populations of neurons. There is much empiric evidence for this point. The second assumption, also illustrated in this cartoon, is that neuronal ensembles representing each item is active during each gamma cycle, which on average is on the order of 20-40 milliseconds. There is some evidence for this second assumption. Here are recordings obtained from the rodent hippocampus while the animal is moving around in an arena. Each row shows a spiking activity in recorded neurons. The circles indicate groups of neurons that represent specific locations within the arena. And you can see that each of those neuronal ensembles representing each of those locations fired together within a time span of 20-40 milliseconds. This phenomena in which neuronal ensembles representing distinct items are active at distinct phases of theta is referred to as theta phase coding. So the objective in our study was to test that theta phase coding is a physiological mechanism that allows us to hold multiple items of working memory using gamma oscillations as a marker of neuronal ensembles. To test our hypothesis, we designed our experiment such that in each block, subjects are presented with three unique cue odors in succession. Following a short retention interval of a few seconds, subjects experienced a probe odor along with up to two questions to test their memory for the original odor sequence. Hence, each block consists of the three stages of working memory, encoding of information into working memory, maintenance or retention of this information over a short delay, and retrieval of working memory content. The first question was whether the probe was part of the original three-item odor sequence or whether it was novel. A second question was provided if the subject answered that the probe was just encountered and asked which cue odor in the original sequence the probe corresponded to, first, second or third. This first example block shows a correct behavioral response. This second example block shows an incorrect behavioral response. We obtained intracranial EEG from eight patients, focusing on two regions of interest, the primary olfactory sensory cortex, which is the piriforme cortex, and the hippocampus. Each subject performed 50 blocks while odors were delivered to the patient's nose with high temporal precision using a custom olfactometer, while the patients followed instructions and answered questions on a computer monitor. This is an example of the type of signals we obtained with intracranial EEG. In red is the raw oscillatory activity, and in black is the respiratory trace of the subject as they are instructed to take a deep breath in response to each odor delivery. We next take a closer look at the neural response to the first cue odor in this block, where now the raw trace has been decomposed into its frequency components. In blue is oscillatory activity in the theta frequency band, and in red is activity in the gamma band. Even at the level of individual odor responses, we can see that odor-induced gamma activity is more prominent at specific phase ranges of the concurrent theta oscillations. In this example, there is enhanced gamma activity at the downslope of theta, as emphasized in gray. For each cue odor trial, we computed a histogram of the amount of gamma activity across the phases of theta, averaging across each consecutive theta cycle as shown by the arrows. For each subject, we then averaged the trial-level phase histograms across blocks based on the sequence position the cue odor was presented in. Note that the subset of odors presented in each of the three cue positions included many different odor types. Here is a trial-average phase histogram in one subject, in which neural responses to cue odors presented first in the sequence of three was averaged across all 50 trials. The bars show average gamma activity across each phase bin of theta. We see that gamma activity induced by cue odors in position 1, position 2, and position 3 form distinct clusters along the theta cycle. This exemplifies theta phase coding, in which odor-induced gamma occurs at distinct phases of theta depending on its sequence position. In other words, the preferred theta phase at which gamma occurs tells us something about the sequence position of the odor, and this information is represented independently from the identity of the odor. Importantly, theta phase coding was only seen in blocks where subjects successfully remembered the odor sequence. Here we show the phase histogram in the paraform cortex and in the hippocampus of one subject, showing distinct theta phase preferences for odors presented in each of the three sequence positions, as indicated by the three colors. In contrast, for odor sequences that were not successfully retained in working memory, clusters based on sequence position were not as reliably observed. This correlation with behavior suggests that theta phase coding may be a mechanism that supports working memory for odor sequences. Across all subjects included in this study, theta phase coding was more frequently observed during successful memory in both structures. In summary, this is the first demonstration in humans of theta phase coding across sensory cortex and hippocampus in the service of sequence memory. Theta phase coding is a possible mechanism by which neuronal populations can keep multiple items online in working memory. Finally, these findings suggest that working memory capacity may in part be due to neural dynamics during the initial sensory perception and not just as a retention failure. Thank you for your attention.

Andrew Yang

Neurosurgery

working memory

neural oscillations

theta phase coding