Representing and remembering naturalistic events
My work is currently focusing on relating existing theories of memory to research paradigms that tap more directly into real-world experiences. Though we know that memories can be dissociated based on content - such as items and contexts (Ranganath & Ritchey, 2012; Ritchey, Libby, & Ranganath, 2015) much of the evidence for these dissociations comes from simple paradigms with objects against backgrounds. How does this work in real life, where we extract 'event representations' from a continuous flow of incoming sensory input? How do we represent information as unique, or categorize it based on overlap with other experiences?
I designed a study to answer these questions. In this research, I filmed 8 video clips, each involving a central character in a single context. In total, I filmed two people in four different contexts, and importantly, there were two classes of contexts (i.e., two cafes and two grocery stores). I then used representational similarity analysis over patterns of fMRI data to characterize the contents peoples' brains extracted from these video clips. Though analyses are ongoing, I am finding that anterior-temporal brain areas extract stable representations of specific people across multiple dynamic episodes. Conversely, posterior-medial brain areas extract stable representations of particular spatial contexts (e.g., this cafe). Here's where it gets super cool. Medial prefrontal cortex extracts representations of classes of contexts (e.g., any cafe) despite major perceptual differences and different characters, and hippocampal coding seems to be highly specific to a single event (e.g., any one video versus all others). This offers some major clues about the way networks of brain areas may extract particular bits of information from ongoing experience, and scaffold them together into a holistic representation of an event.
Naturalistic memory in the aging brain
In recent years, studies have demonstrated that the brain's posterior-medial cortico-hippocampal network is heavily involved in the segmentation and representation of continuous, naturalistic experiences (Chen et al., 2017; Baldassano et al., 2017). A very interesting and potentially critical component of this network of brain areas is that it is the very brain network most susceptible to amyloid plaque deposition in the brains of elderly people at risk for Alzheimer's disease. Though amyloid beta is not always strongly associated with memory deficits, it is possible that our current tests are not sensitive enough to posterior-medial network function. In fact, most memory tests used in aging may not be sensitive to the way memory works in the real world at all!
I am currently tackling this question by having older participants with and without significant amyloid deposition lie in the scanner while they view and recall an episode of a television show. My hypothesis: amyloid beta deposition will disrupt the posterior-medial network's ability to segment continuous experiences into meaningful chunks, which will disrupt representations of particular events, and will also thus lead to deficits in memory for those events. Data from this line of research may give us key insights about the way early Alzheimer's disease affects the way we understand and remember events in our daily lives, and could lead to new and more sensitive biomarkers!
Domain selectivity in the human MTL
Memory isn't a single entity. Rather, there are different types of memories - for instance, detailed recollection versus a general sense of familiarity - that seem to rely on different brain regions. Evidence has accumulated suggesting that dissociable regions in the medial temporal lobe (MTL) support memory for objects/items versus space/context (Deshmukh & Knierim, 2011; Ranganath & Ritchey, 2012). This view has been partially corroborated in the human brain: whereas peririhinal cortex (PRC) selectively responds to objects (Taylor et al., 2006), parahippocampal cortex (PHC) selectively responds to spatial or contextual infromation (Davachi, Mitchell, & Wagner, 2003). However, the full extents of these mnemonic processing pathways have remained elusive in the human given no evidence for functional dissociations between lateral and medial entorhinal cortex (LEC and MEC).
A major focus of my research has been elucidating the functional roles of LEC and MEC in the human brain. I developed a task in which subjects study objects in particular spatial locations, and during a surprise test, are presented with similar but not identical objects and spatial configurations (object and spatial lures). The logic here is that nonoverlapping information should interfere with information in memory (mnemonic interference), and different regions of the MTL should be involved in respectively overcoming interference for objects or spatial locations. Crucially, this task was used in conjunction with high-resolution (1.5mm isotropic) functional MRI (fMRI). This allowed me not only to parcellate subfields of the hippocampus, but also LEC versus MEC in the human brain.
This vein of research (Reagh et al., 2013; Reagh & Yassa, 2014) has yielded three main discoveries: (1) Human LEC and MEC are in fact dissociable based on information domain. (2) The dentate gyrus and CA3 subfields of the hippocampus (DG/CA3), which are involved in pattern separation of overlapping information, are similarly engaged during discrimination of object and spatial lures. Thus, the hippocampus appears to be capable of flexibly resolving interference across information domains. (3) A correlational structure emerged between PRC-LEC-DG/CA3 and PHC-MEC-DG/CA3 during object and spatial discrimination, respectively. This suggests that cortical regions in the MTL may work in conjunction with hippocampal DG/CA3 during pattern separation.
Selective age-related vulnerability in MTL structures
It is commonly known that aging adversely affects memory. However, certain brain areas exhibit age-related changes earlier and to a greater extent than others (Jack et al., 1997). The "transentorhinal region" made up of PRC and LEC are especially vulnerable (Braak et al., 1993; Raz et al., 2004), and at very early stages of diseases associated with advanced age, such as Alzheimer's disease (AD) (Burke et al., 2011; Khan et al., 2014; Stranahan et al., 2010). Thus, PRC and LEC may be a kind of "ground zero" for age-related memory decline (Yassa, 2014).
Conveniently, the task described above was able to dissociate LEC and MEC functional signals in the human brain. I thus applied this task to examining the effects of aging on task performance, hypothesizing that aging would affect object discrimination (PRC & LEC) more so than spatial discrimination (PHC & MEC). I examined general aging such that all of our older adults were free of an overt diagnosis of dementia. Consistent with our hypothesis, older adults were impaired at resolving object interference to a far greater extent than spatial interference, whereas both abilities were equal in younger adults (Reagh et al., 2016). Interestingly, splitting the healthy older adults into high (aged unimpaired; AU) and low (aged impaired; AI) performers on the basis of neuropsychological testing revealed that object discrimination deficits are widespread, whereas spatial discrimination deficits are only seen in the low performing AI individuals.
Although this behavioral evidence is consistent with early age-associated PRC and LEC dysfunction, we need more direct evidence about what's going on in the brain. In a recent fMRI study (Reagh et al., 2018), we replicated our behavioral findings, and furthermore found that asymptomatic (i.e., without memory impairment) older participants show hypoactivity in the anterolateral EC (a more modernized segmentation of human LEC), and hyperactivity in DG/CA3. Thus, in addition to what we previously knew about DG/CA3 hyperactivity in aging (e.g., Yassa et al., 2010; Bakker et al., 2012), our findings suggest that the alEC may play a critical role in age-related dysfunction in the MTL. Finally, I found that the ratio of activity between alEC and DG/CA3 was skewed toward DG/CA3 activity in older adults, and that the extent of this skew predicted behavioral impairments. Thus, interplay between alEC and DG/CA3 may comprise an important mechanism for early changes to the MTL memory system, even in ostensibly "typical" aging.
Modifying memories through repeated learning
At this point, researchers have been studying how the brain learns and remembers for decades. However, there are so many factors that play into this process that we know little about even very basic questions. One very relevant example is that of repeated learning. Consider the following scenarios: (1) You meet someone for the first time at a party, and then see them again the next day. They are "familiar" to you in the sense that you have seen them the day before. (2) You meet a friend at a party, with whom you have a long history, and then see them again the next day. They are "familiar" to you both in that you saw them the previous day, and also many times prior to that. How does the memory machinery of the MTL represent this distinction?
One way I went at this is to prefamiliarize people with well-known landmarks (e.g., the Eiffel Tower) in a study phase, and test recognition memory the next day. Importantly, during the recognition phase, items were presented twice to induce the "hey, I just saw that" sort of familiarity. It turns out that, within the MTL, an interesting and previously undiscovered dissociation emerged between the anterior and posterior portions of the hippocampal DG/CA3. Whereas anterior DG/CA3 was sensitive to within-session familiarity, posterior DG/CA3 signalied well-learned items that were novel to the test session (Reagh et al., 2014). I am very interested in further exploring dissociations along the hippocampal longitudinal axis.
Another fun way I have gone at this is to have participants implicitly study items either once (1Enc) or three times (3Enc), and then test their ability to recognize identical targets and reject similar lures as a function of repetition. The motivation for this is based on mine and Mike's Competitive Trace Theory, which you can read about in detail here. Briefly, I put forward the idea that the hippocampal DG/CA3 circuit is so good at creating novel representations that it may attempt to do so if identical information is presented multiple times. This could induce nonoverlapping representations that may compete for representation, which strengthens the gist of the experience (e.g., "I saw a black dog") at the expense of details (e.g., "It was a labrador retriever whose name was Marty McFloof"). Although this is a largely theoretical pitch, I'll note that it does nicely account for how episodic memory for events becomes semantic over time.
Back to the experiment! I found that repeated study enhanced target recognition but reduced discrimination of similar lures at test (Reagh & Yassa, 2014). This is a highly counterintuitive result, but fits well with the logic described above: perhaps multiple encounters with the same object led the hippocampus to create similar but not identical representations for each object, causing central features to be strengthened while peripheral details fell by the wayside.
A followup high-resolution fMRI experiment provided some clues as to the mechanism (Reagh et al., 2017). We found that the hippocampal DG/CA3, known for its involvement in orthogonalizing similar inputs, responded strongly to lure objects studied only once (in line with plenty of prior work). However, for items studied three times, activity actually ramped down. The opposite trend was found for target recognition in regions of the brain including perirhinal cortex, anterior cingulate/ventromedial prefrontal cortex, and an anterior-medial portion of the thalamus corresponding to nucleus reuniens (the first evidence of its involvement in memory processes in the human!). Conversely, hippocampal CA1 rampted its activity down during target hits. Thus, we get an overall picture of hippocampal activity decreasing and neocortical activity increasing with repeated study. Finally, we found a connection among these processes in anterior cingulate/ventromedial prefrontal cortex, anterior CA1, and nucleus reuniens. For target recognition, we found that the nucleus reuniens statistically mediated correlations between the hippocampus and prefrontal cortex, in line with the anatomical connectivity among these regions. Thus, the thalamus may play a key role in adjudicating between hippocampal and neocortical involvement for well-studied information.
