Welcome to the Stark Memory Research Group!

PI: Craig E. Stark, Ph.D., Director of the Center for the Neurobiology of Learning and Memory
Lab_May2014

Research in the Stark Memory Research Group is concerned with the mechanisms that underlie memory. The central question guiding research in the lab is how is it that we learn and remember information such that our past experiences influence our behavior? Using the techniques of functional magnetic resonance imaging (fMRI), traditional experimental psychology, neuropsychological studies of amnesic patients, and connectionist modeling, research in the Stark lab is focused on how the neural systems supporting these two types of memory operate and interact. 

Pattern Separation and Pattern Completion
In recent years, we have been interested in the concepts of pattern separation and pattern completion as underlying processes supported differentially within substructures of the hippocampus using the high-resolution fMRI techniques my lab has developed. Following our paper that came out in Science (Bakker et al.,2008, Science), we have completed behavioral and neuroimaging studies that test the hypothesis that a subregion of the hippocampus (the dentate gyrus) is performing pattern separation (a computation that leads it its role in episodic memory). Our previous work was able to show that activity in the dentate gyrus and the adjacent CA3 field of the hippocampus was sensitive to small changes in the input. Here, “lure items” that are similar to previously-seen items have activity consistent with the regions treating these as novel items (having “pattern separated” this similar input). In contrast, other regions in the hippocampus and adjacent MTL cortices have activity for these items that is more consistent with treating them as if they were repetitions of the previously seen item (referred to as “pattern completion”). While influential, a limitation of this work was that we only had three distinct levels of “mnemonic similarity”: new items, previously-seen items, and items similar to previously-seen items. Therefore, we could not measure the activity transfer function of the separate subregions with much resolution (how activity in each region responds to change in the input – a key prediction of the model we are working with).

To address this limitation, we developed a large set of stimuli with known levels of mnemonic similarity. By getting a strong handle on how readily participants (~150 tested) would confuse a similar lure item for a true repetition of an item, we have a manipulation much like one performed in a number of recent electrophysiological studies in rodents that can give us the x-axis in our investigations of the transfer functions. We have used this task now in both behavioral and neuroimaging studies using both young and aged (see below) volunteers to test our model’s predictions. In the most direct application (Lacy et al.,2011, Learning and Memory), the stimuli were used as a direct extension of the Bakker et al., (2008) paper. Consistent with our predictions (derived from computational models and both unit-recording and immediate early gene work in the rodent), we observed a very sharp transfer function in the CA3/dentate subregion (indicative of pattern separation) and a gradual (roughly linear) transfer function in the CA1 subregion. We hope this will help cast many of the current theories of human hippocampal function in a new light and allow us to make stronger connections between the human and animal literature. In addition to this task, we have developed a spatial pattern separation task to test the model’s predictions that is derived from another area of the rodent literature (Stark et al., 2010, Learning & Memory).

ScienceMethods.027ScienceData.09

Aging and Early Dementia
A second theme has been our rapid expansion into the study of aging and dementia. This is a very exciting area for us as a number of the theories and techniques we have been developing for our studies of pattern separation have clear applications in research on aging. We are attempting to unite our work on pattern separation with a rodent model of aging that highlights structural changes in the hippocampal network associated with healthy aging. In particular, the model highlights the age-related degradation of the perforant path – the input to the dentate gyrus and CA3. Thus, studying healthy aging provides us with a way of examining the effect of removing these structures (or at least impairing their function). This model allows us to test the hypothesis that these regions are performing pattern separation and it gives us a new perspective on age-related memory loss in humans – one that focuses on the anatomy and the computations performed by subregions rather than on more descriptive-level analyses.
We have now tested these predictions in several ways. Using the mnemonic-similarity task described above, we observed age-related behavioral changes in participants’ ability to accurately identify lure items as being similar to (but not the same as) previously-seen items (Yassa et al., 2010,
Hippocampus). This effect is behaviorally amplified in Mild Cognitive Impairment (MCI; Yassa et al., 2010, NeuroImage). The behavioral effect was paralleled by (and correlated with) activity changes in the CA3/DG consistent with an age-related change in the structures responsible for pattern separation. As noted above, we have developed a related task that assesses spatial pattern separation abilities to test these predictions as well. In a behavioral study (Stark et al., 2010, Learning & Memory), we observed a similar age-related shift away from pattern separation and towards pattern completion – again consistent with the model. Finally, using new diffusion tensor imaging (DTI) scanning and analysis techniques we have developed (see below), we have been able to visualize and quantify the integrity of the perforant path in humans (Yassa et al., 2010, PNAS). We have not only observed the predicted age-related loss in the perforant path, but have been able to correlate this with standardized neuropsychological test scores on a task that taxes pattern separation.

YassaNImg2010Data.055YassaPNAS2010Data.048

Interactions Across Memory Systems
A third theme we have been developing in the lab concerns the relationship between learning in the medial temporal lobe and learning in the striatum. Much of our work has involved tasks that can (and are) learned by both systems and our focus has previously been on the medial temporal lobe. Now, we are extending our scope to the striatum and to examining how these systems may interact (how reward-prediction errors in areas such as the nucleus accumbens may be interacting with the MTL). The literature has made much of the fact that these two memory system can be independent. These studies have moved our field forward considerably and provided some of the critical evidence to show that there are many different forms of memory and many different brain systems participating in memory. Memory is not one thing. That said, it is not clear that they should be viewed as entirely independent in normal operation.
We have recently (Mattfeld & Stark, 2010,
Cerebral Cortex) used fMRI to examine changes in activity in both systems over the course of arbitrary paired-associate learning (using a task my lab previously developed specifically to link fMRI with non-human primate neurophysiology; Law et al, 2005, Journal of Neuroscience). We observed multiple learning-related signals in the striatum (e.g., activity tracking probability correct versus the rate of learning) and were able to localize them with greater precision than was previously possible. Similar signals were also observed in the MTL. We also observed “functional connectivity” between these regions (correlation in the activity). What was most interesting, however, was not the presence or absence of these signals or the simple correlations in activity, but how the correlations changed over the course of learning. This form of analysis had never been applied to these issues but was able to show clear dissociations across striatal regions. The ventral striatum was robustly coupled with the MTL, but only during the peak periods of learning. The associative striatum showed the converse. Together, these are consistent with the hypothesis that the nucleus accumbens (ventral striatum) has a functional relationship with the MTL but that there is a functional dissociation between the caudate and the MTL.

MattfeldCC2010Data.061

Neuroimaging Methods
Fitting in with all three of the research areas mentioned so far has been our work in fMRI methods. In particular, we have recently finished an extensive analysis and quantification of the ability of over a dozen techniques to co-register MRI scans of structures in the MTL across individuals (Yassa & Stark, 2009, NeuroImage). The results of this study show conclusively that the approach we have been developing over the last several years (“ROI-AL”) is significantly more accurate than anything else currently in use. A chief problem with our approach is that in the past it took several hours to align each participant and it could only be run on the computers at the Center for Imaging Sciences at JHU. Thus, while we were able to use the technique, few others could, severely limiting its application. In our recent paper, we present a new implementation of the approach that takes only minutes to run and we have made the software available for use on Windows, Linux and Mac OS X platforms.
In addition to working on alignment methods (which are critical if one wishes to make claims about subregions of structures like the hippocampus – already not large), my lab has been active in pushing MRI’s resolution. Using techniques we developed (Kirwan et al., 2007,
Human Brain Mapping) at JHU and brought to UCI, we collect functional imaging from the hippocampus at one of if not the highest resolutions of any lab in the world. We developed this as this resolution was needed to address our scientific questions about the role of subregions of the hippocampus (and now striatum).
Diffusion Tensor Imaging (DTI) has been a popular scanning technique as it provides a non-invasive method for mapping white matter tracts in the brain. We have not used it though, as its resolution (2-3 mm) has been insufficient to resolve the fibers that would inform our work. Recently, however, the need to resolve these fibers grew as we wanted to be able to assess changes in the perforant path associated with age. As it is only a few millimeters wide and as it is buried inside other white matter tracts, we needed to develop numerous tools to measure and quantify it. In collaboration with Dr. Tugan Muftuler here at UCI, we were able to cut the in-plane resolution down to 0.66 mm. I then devised a technique that used anatomical constraints (derived from tract-tracing studies in the monkey) to identify and measure the integrity of the perforant path from this high-resolution DTI data. Our paper (Yassa et al., 2010,
PNAS) shows the predicted age-related change in the perforant path (described above) and details the new contribution made to DTI data analysis.

YassaPNAS2010Data.025YassaNrImg2009Data.045


Last Edited May 2014