Univerity of Wisconsin-Madison

SfN 2013 - Nanosymposium Session 507


Tuesday, November 12, 8:00 - 10:00AM, 4

Chair: Bradley Postle, University of Wisconsin - Madison

Speakers: D. E. Nee, D. Mendoza-Halliday, N. Myers, G. Wallis, A. C. Riggall, J. J. LaRocque, N. Rose, N. Zokaei

Neural correlates of shifting representational states in short-term memory

1Univ. of California, Berkeley, Berkeley, CA; 2Psychology, Univ. of Michigan, Ann Arbor, MI

Short-term memory (STM) - the maintenance of information in the absence of external stimulation - is at the core of higher-level cognition. Behavioral and neural data have demonstrated that information maintained in STM can be represented in qualitatively distinct states. These states include a single chunk held in the focus of attention (FA) available for immediate processing, a capacity-limited set of additional actively maintained items that the FA can access (direct access region; DAR), and passively maintained items held in the activated portion of long-term memory (aLTM). Little is known how information is shifted among these states. Here, we used fMRI to examine the neural correlates of shifting information among representational states. We employed a paradigm that has previously demonstrated dissociable performance costs associated with shifting the FA within the DAR and the DAR within the aLTM. Behavioral results fully replicated prior reports. Neural results indicated that areas known to be critically involved in top-down attention are involved in shifting information among representational states including the frontal eye fields (FEF), inferior frontal junction (IFJ), and intra-parietal sulcus (IPS). While univariate activations only partially dissociated different forms of shifting, task-related changes in FEF-hippocampal functional connectivity doubly dissociated shifting the FA and the DAR. Increased FEF-hippocampal connectivity was associated with shifting the FA, while decreased FEF-hippocampal connectivity was associated with shifting the DAR. These data indicate that interactions between top-down attention networks and the hippocampus mediate shifts of information among different representational states in STM. 

Transition from sensory to working memory representations along the primate dorsal visual pathway

McGill Univ., Montreal, QC, Canada

Visual working memory serves as a mental “sketch pad” that allows visual information from the past to be temporarily maintained. It is an essential component of abstract thinking and the planning of future actions, two central pillars of intelligence. It is currently established that in primates, the contents of visual working memory are encoded in the sustained activity of neurons in cortical areas located far downstream in the hierarchy of visual processing, such as the lateral prefrontal cortex (lPFC). Some studies have proposed that because neurons in these areas may lack sharp tuning for visual features, holding such features in working memory may also require the activity of feature-selective neurons in early visual cortex. However, this hypothesis raises the question as to how sensory and working memory representations would be distinguished if they were encoded by the same neurons. It remains unclear in which processing stage along the visual pathway working memory representations emerge. Here we show that when macaque monkeys maintained the direction of motion of a visual stimulus in working memory, neurons with sustained spiking activity representing the remembered direction were found both in lPFC and in the medial superior temporal area (MST), but not in early visual area middle temporal (MT), despite the well-established role of MT and MST in sensory coding of visual motion. Since MT and MST have direct reciprocal connections, our results indicate that working memory representations of motion direction emerge in area MST along the dorsal visual pathway. The strength of working memory representations was similar in MST and lPFC, but it was far more predictive of working memory performance in lPFC, an effect that depended on the similarity between the lPFC neuron’s preferred direction and the remembered direction. Moreover, local field potentials in MT were tuned to the remembered directions and, in the lower frequencies, were phase-coherent with spikes from lPFC. The latter suggests that the activity of lPFC neurons likely influences postsynaptic potentials in MT through feedback projections. Our results propose that in intermediate sensory processing stages, there is a transition in cortical architecture: from one that supports sensory processing in MT and upstream areas to one that supports working memory maintenance in MST and downstream areas such as lPFC. This partial segregation between sensory and working memory systems may be crucial for the brain to distinguish between the current sensory input and the contents of working memory.

Slow oscillations reflect top-down updating of the contents of visual working memory

1Exptl. Psychology, 2Dept. of Psychiatry, Oxford Univ., Oxford, United Kingdom

When we retain information in working memory (WM), we can update new information to the currently held set. This ability comes at the expense of increased forgetting of information encoded earlier. Cues can help us to update items selectively into WM by allowing us to encode new information only when it is likely to be behaviorally relevant. We recorded electroencephalographic data from 18 health young volunteers as they performed a visual working memory precision task. On each trial, subjects encoded orientation information from a first display into working memory. One second after array offset, a central cue indicated whether one of the previous items would be probed (protect condition), or whether the probed item would appear in the subsequent array (update condition). A second array appeared after the cue, followed by a forced-choice probe.
Valid cues increased the likelihood of remembering an item in WM, whether it was protected or updated. Invalid cues decreased this likelihood. In addition, we found that cues induced a desynchronization in the alpha band (8-14 Hz) for both cue types (starting ~500 ms after the cue), indicating that top-down attention modulates both the protection of current contents and anticipated updates in WM. Furthermore, alpha band desynchronization reappeared immediately before the onset of the memory probe. These results are in line with the proposal that the contents of WM guide anticipatory attention to facilitate behavior.

Low frequency oscillations recorded with MEG reflect both prospective and retrospective selection in a working memory task, and reveal cognitive control networks

1Oxford Univ., Oxford, United Kingdom; 2OHBA, Univ. of Oxford, Oxford, United Kingdom; 3Dept. of Psychiatry, Oxford Univ., Oxford, United Kingdom

We can select one of a number of items to encode into working memory if we are told in advance which item to remember. This improves performance for the cued item. There is a similar increase in performance if we are cued to select an item within working memory a short time after we have seen the working memory array, but before the probe to which it is to be compared is presented (retrocueing). This is not attributable to the iconic store (retrocues are still effective seconds after array presentation). We used a forced-choice 4-item precision/capacity visual working memory task for orientation with uncued, precued and retrocued trials. 48 healthy young volunteers performed the task whilst MEG was recorded using a 306 channel Elekta MEG system. We replicated the behavioural advantage previously found with precueing and retrocueing; guess rate is substantially reduced in both precue and retrocue trials relative to neutral trials.
Induced activity in the alpha (~10Hz) and beta (~20Hz) bands lateralizes in posterior cortex in response to a directional precue, desynchronizing in the hemisphere contralateral to the attended side. Using an LCMV beamformer to estimate induced responses in source space, and comparing cues directing attention to the left and right side of space, we replicate this lateralization of low frequency power with preparatory attention in dorsal stream visual and parietal regions, and show that retrocues give rise to a similar lateralization of low frequency power, of larger magnitude than for preparatory attention. Comparing cued and neutral trials, we show that modulations of low frequency power following informative retrocues are localized to a fronto-parietal network previously implicated in selective attention both for percepts and internal representations. Activation of this network ~300ms post-retrocue precedes the lateralization of alpha power, around 450ms post-cue. These findings are consistent with the proposal that low frequency oscillations coordinate the long-range interactions between brain regions that mediate cognitive control.

Understanding the representation and precision of transparent motion information during visual short-term memory with fMRI pattern classification

1Psychology, 2Psychiatry, Univ. of Wisconsin-Madison, Madison, WI

A number of recent results have begun to question long-standing beliefs in how information is represented temporarily during working memory (WM). The sensitivity of elevated delay-period BOLD activity to variation of memory load,typically observed in the intraparietal sulcus (IPS) and dorsolateral prefrontal cortex (dlPFC), has often been taken as evidence that these areas play a critical role in representation. However, using multivoxel pattern analyses (MVPA), we and others have shown that stimulus-specific representations can be recovered from posterior visual areas, but not parietal or frontal regions, during the delay periods of several visual WM tasks. In the present study we extended this work by examining the representation and precision of multiple items in memory for transparent motion stimuli. Transparent motion, where several directions of motion are briefly presented simultaneously in a single spatial location, provides a serious challenge for storage systems, as few additional clues (e.g., location, temporal order) can be used to help support the individual item representations. We tested the ability of participants to remember the direction of motion for multiple (1, 2, or 3) overlapping, coherently-moving random dot stimuli, each presented in a different color. After the delay, one of the motion stimuli was redisplayed and participants reported the direction that this probe had been rotated (clockwise or counter clockwise; magnitude of change 7, 20, or 45 degrees) compared to the remembered item. The responses of participants varied probabilistically with magnitude of the change. For each memory load participant's responses were fitted with a cumulative Gaussian distribution, providing a measure of precision (the reciprocal of the standard deviation of the response function fit) for each load. Participants showed decreased precision with increased memory load. Univariate GLM analyses of the BOLD data identified robust load-sensitive activation in IPS and dlPFC. MVPA in these regions, however, showed no evidence for stimulus representation. In contrast, in striate and extrastriate cortex, which did not show load-sensitive delay-period activity, MVPA indicated delay-period representation of the direction of motion. Within these regions the classifier evidence scaled with behavioral estimates of precision. These results support the view that frontal and parietal cortex represent high-order information about WM tasks, but are not involved in the direct representation of information.

Active representations of individual items in short-term memory: A matter of attention, not retention

1Neurosci. Training Program and Med. Sci. Training Program, 2Psychology, 3Psychology and Psychiatry, UW Madison, Madison, WI; 4Psychology, Brock Univ., St. Catharines, ON, Canada

For decades, a central assumption of cognitive neuroscience has been that short-term memory (STM) is accomplished by sustained, elevated neural activity. Theories of STM in recent years have expanded this view by suggesting that different levels of activation may correspond to an item being in one of various states, such as in activated long-term memory vs. in the focus of attention. This theoretical suggestion inspired several studies using multivariate pattern analysis (MVPA) to look for neural activity predictive of the category of items inside or outside the focus of attention. Instead of finding evidence for differing levels of above-baseline activity for item categories in different levels of STM, these studies failed to find any evidence at all for neural activity associated with the category of items retained outside the focus of attention, despite participants’ ability to later accurately retrieve those items. These results, obtained from decoding both fMRI and EEG data, suggested that only items in the focus of attention are accompanied by an active trace. However, these results were obtained by category-level decoding, which may be insensitive to some item-specific aspects of neural representations. Therefore, in the present study, participants performed a STM task modeled on the previous tasks, but in which they had to remember the directions of motion of coherently moving sets of dots. This would permit MVPA decoding of delay-period fMRI signal at the stimulus-item level, a more stringent procedure than stimulus category-level decoding. In the one-item version of the task, decoding in an ROI defined by its phasic response to the stimulus was well above chance (58% classification accuracy, where chance was 33%). In contrast, decoding in an ROI defined by elevated activity sustained throughout the delay period was less successful (38%). Participants also performed a two-item version of the task that employed retroactive cues indicating which of two items held in STM would be the target of an imminent memory probe. Evidence for the cued item was reliably high, whereas evidence for the uncued item dropped to baseline levels despite participants’ ability to subsequently respond to a memory probe for this uncued item. This result, in agreement with previous category-level decoding studies, suggests that an active neural representation is present for items in STM only when they are also in the focus of attention.

Levels of representations in verbal working memory

Rotman Res. Inst. at Baycrest Ctr., Toronto, ON, Canada

How does the brain maintain a to-be-remembered item ‘in working memory’ during a delay, particularly when the focus of attention is drawn to processing other information? On each trial in this fMRI study, participants encoded a single word in a deep or shallow level of processing and then either rehearsed the word, performed an easy-math task during which it was likely participants could think of or ‘refresh’ the word, or a hard-math task during which it was unlikely participants could refresh the word before attempting to recall the word. Multivariate pattern analysis was used to classify neural activity associated with deep versus shallow levels of processing during encoding, delay, and recall phases. The results revealed that information about the level of processing at encoding was present during the delay and at recall for math conditions, but not rehearsal, suggesting deep (semantic) cues were used for WM maintenance and recall when the focus of attention was drawn away from rehearsing the word.

Causal evidence for a focus of attention in working memory

1Oxford Univ., Oxford, United Kingdom; 2Univ. Col. London, London, United Kingdom; 3Reading Univ., Reading, United Kingdom

Some current neural models of visual working memory (WM) propose that sensory cortex maintains representations in an activated state across the short term, through sustained neuronal activity. Several theoretical cognitive models posit, however, that not all WM representations are retained in the same state. Through a focusing of attention, some items are represented in a more prioritised state. To test this, we applied the causal brain stimulation approach of transcranial magnetic stimulation (TMS) to motion sensitive MT+ in healthy humans, with the aim of disrupting maintenance of motion-related information held in the WM focus of attention (FOA).
In the first experiment, on each trial two differently coloured moving kinetograms were presented, requiring participants to remember the direction of motion of each patch. To bring one of the two motion directions into the FOA, we used an ‘incidental cueing’ technique, asking participants to report the colour of one of the two patches. The cue was never predictive of the probe. At the end of maintenance, they were probed to recall one of the two directions of motion. Incidental cueing resulted in the cued item being recalled with higher precision compared to the uncued one. A short train of TMS was applied after the incidental cue and before motion direction recall. Stimulation decreased the precision with which participants reported the incidentally cued motion direction, but increased recall precision of the uncued item. Thus TMS to MT+ disrupted only the memory for the item in the FOA, in an activated state.
In our second experiment, we aimed to repeat our findings but now by bringing an item into FOA implicitly, via the recency effect. We presented two moving dots patches sequentially, with the last item in the sequence being recalled with higher precision than the previous item. TMS applied during maintenance (i.e., after the two items had been presented, but prior to recall) now improved precision for motion direction of the first item in the sequence. Crucially, TMS disrupted recall precision of the last item.
Together these findings suggest that TMS to MT+ disrupts only memory of items prioritised within WM, regardless of how this activated state is instantiated, either by endogenous cueing or by virtue of recency. These findings provide the first direct evidence in humans for the role of MT+ in short-term retention of motion direction information, supporting models in which regions involved in perception also subserve WM retention. Moreover, they provide a strong line of evidence for the representation of items in WM in different states, with items in FOA being most vulnerable to sensory cortex disruption via TMS.

©2017 PostLab.