Skip to main content
Loading Explore guides…
Memory Networks in the Brain · Acedra
Home
Explore
Medicine
Memory Networks in the Brain
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
157 segments
Copy
Hide video
0:04
Hello everyone. My name is Dr. K.C. Hewitt, and for the summer, Anny Reyes and I are your moderators. We are pleased that you could join us for this week’s lecture in our first 7-week KnowNeuroanatomy mini-series that brings you lectures from experts in the field, covering different neuroanatomy topics each week. This series was created by trainees and early-career neuropsychologists to provide free, high-quality didactic opportunities. We also would like to
0:29
recognize the KnowNeuroanatomy planning group for their hard work for pulling this mini-series off.
0:34
We would also like to thank our sponsors for their financial support for the series.
0:39
Now before we start, here are disclaimers for the series. This training is not meant to replace formal education in neuropsychology and the views of the speakers are their own.
0:49
Questions can be submitted via the Q&A box in the lower left side of your screen, and a recording of today's lecture will be provided on our website later this week. And Dr. McAndrews has also been providing her slides for you as well. Questions will be asked at the end and not during the lecture. Today’s Q&A is led by Anny Reyes. Now, it is my pleasure to introduce Dr. Mary Pat McAndrews for today’s lecture titled,
1:13
Memory Networks in the Brain. Dr. McAndrews is the Director of the Neuropsychology Clinic and is the Senior Scientist and Head of the Division of Clinical and Computational Neuroscience at the Krembil Institute at the University Health Network, and Professor in the Department of Psychology, University of Toronto. She completed her PhD in Psychology at the University of Toronto and her post-doctoral studies in Neuropsychology at the Montreal
1:37
Neurological Institute. Her research focuses on memory networks, using functional neuroimaging, studies of patients with medial temporal lobe dysfunction, and brain stimulation techniques to understand how these networks are organized and modified by focal damage or therapies.
2:03
Okay, thank you very much, K.C., for that introduction. Hopefully, everyone can see my screen, and unless I get a nod otherwise, I’ll just continue on. So, thanks for inviting me to talk about my very favorite topic, which is memory networks in the brain. Instead of a formal outline, I’m just going to tell you that, primarily, I’ll be speaking much more at a macro level, not at tiny brain circuitry. Also, a lot of the research that I’ll be talking
2:32
about—the information comes from fMRI, which is our current window into networks in the brain.
2:41
Okay, and that's,
2:48
there we go! Okay, so every talk about memory in the brain has to start with the medial temporal lobe, in cartoon here on the left and on the right. Keeping in mind that while we focus a great deal on the hippocampus per se, there are a number of other important players within the medial temporal lobe that provide their own important sustenance and contributions to memory, including the peri-hippocampal, entorhinal, and perirhinal cortices.
3:15
Of course, no talk about memory in the brain or the medial temporal lobe can forego speaking about the famous case of H.M., who suffered amnesia following bilateral medial temporal lobe surgery. Brenda Milner was able to use data from his decades of research with him to help us understand much better the primary role of the hippocampus and surrounding structures on memory. A lovely
3:41
book, for those of you who haven't yet read it, is from Suzanne Corkin, who worked with H.M. for many decades before her and his death. This slide just shows you what we know from MRI of H.M.'s lesion, that in fact, compared to the healthy brain that we see here, the removal of the entorhinal cortex, amygdala, and hippocampus was about two centimeters of the hippocampus, with the remaining two centimeters largely atrophic. So, we knew from that point in time, as well as from
4:10
earlier studies of animals, that this was really kind of solidifying the field that the hippocampus was a really important structure for memory, and without it, devastating consequences could follow.
4:22
So, I'm going to talk really about two different patient populations today, both in terms of lesion work, but more so on the neuroimaging side of things. And one patient population that we've learned a lot more about the primary role of the hippocampus and medial temporal structures is individuals with medial temporal sclerosis or medial temporal lobe epilepsy. So, in this case, we have kind of the cardinal lesion that's present in about half of the individuals with temporal
4:49
epilepsy, is medial temporal sclerosis, when you have loss of neurons and invagination of glial tissue. And if we take this cartoon here to represent memory performance, from good to bad, we know that preoperatively, when we assess patients, those with medial temporal sclerosis will show poorer memory performance than those without. But if we look for changes from pre- to post-op, we actually see that it's the ones who had the healthier hippocampus functioning better at the
5:16
outset that showed the greatest decline. So again, telling us that that structure was important. If we looked at a slide here, just showing a simple decline in hippocampal volume over time, we can easily see that year to year, everybody who's an older individual loses some degree of volume, but there's a greater degree of volume loss in MCI and AD as there's progressive memory impairment across the board. So, the hippocampus is quite important. We know from lots of lesion studies.
5:48
Given that I'm going to be talking a lot about functional imaging, I think it's also really important to know that does this hippocampal activation that we're going to talk about, correspond to memory capacity? Is it telling us something important? And there are a plethora of studies that show that the hippocampus is activated when you're performing encoding and retrieval tasks. That's, you know, tens of thousands by
6:10
this point, quite frankly. But is it meaningfully related to performance? And I'm just going to show a couple of examples of that. So, for example, here's an older study showing that if you look at individuals who are showing hippocampal activation during a retrieval task, and in fact for healthy individuals, healthy aging individuals, the degree of activation in the hippocampus is correlated across subjects to successful memory retrieval. So, it seems to
6:35
suggest that it's telling us something important about memory because it's correlating there. And conversely, over here, we're looking at hippocampal response during encoding. So this kind of subsequent memory effect, how active it is during encoding, does that actually relate in any way to subsequent memory performance? And here are two sketches showing that activation. These beta weights, which is basically a kind of conditional activation,
7:02
are actually strongly correlated with both high confidence hit rates as well as a recollection estimate. So, it does seem to be correlating with at least the kinds of information about memory performance that we value in terms of telling us something about how the integrity of that system.
7:20
So, then we can also say, okay, does it help us out with patients at all? These are interesting studies to look at healthy brains, but does it tell us something that's going to be useful in a patient situation? And first of all, let's just take a quick look at the mechemical activation as an index of integrity in temporal lobe epilepsy. So, in this case, what we see is that kind of again, the cartoon of what we would expect to see is that
7:44
a high degree of activation should, in fact, tell us that memory performance is good versus poor or correlate with memory performance being good versus poor, and it should also show that same kind of decline following surgery. So, let's look at a couple of examples here. One example from an earlier study in my lab was really just looking at the asymmetry of activation between the healthy and the epileptogenic hippocampus. And so, in this case, what we could say is,
8:12
is the difference between the two meaningful? And in this case, the difference between the two is actually strongly correlated with the difference between memory that was preferentially encoded and retrieved by the healthy versus the impaired side. So, that did seem to suggest that this idea of preoperatively, that strength of activation correlating with memory performance.
8:34
Here's another earlier study from the Queen Square group. And in this case, they were looking at the degree of activation before surgery, and did that tell us anything about the change in memory performance following surgery? Again, looking at this kind of pre- to post-operative decline, but in this case, not in terms of activation, but in terms of memory. So here, they were looking at
8:56
a word encoding task, hippocampal activation in the left hemisphere. And what you can see here is a reliable relationship between the amount that you lost in terms of that same verbal learning test that you were showing the activation on and the degree of activation at the left side of the hippocampus. So, the more it was activated during the encoding task, the more you stood to lose. So, both of these are examples where it does seem to suggest that there's some
9:22
correspondence between activation in this region and integrity of the memory system, but it doesn't turn out always to be the case. And I'll show you some examples right now. And the kind of examples I'm going to be showing you in this case have more to do with what's going on in a different patient population. In this case, MCI or mild cognitive impairment or Alzheimer's disease. And again, here's an early study showing individuals,
9:49
in this case, what we're seeing here are activations that are stronger in healthy controls than in an amnestic MCI population. And this is during when they're doing an encoding and retrieval task, either doing a very shallow task, is it color of black and white, versus a deep retrieval task. And in terms of these are the items that were encoded in a very deep, meaningful way. And the patterns down here are showing in this case, the spatial extent of
10:17
activation and the hit rate during these tasks. And what you can see is that for then normal controls, there does seem to be this nice relationship between better memory performance and greater activation in this region of the parahippocampal gyrus for the shallow task and in this region of hippocampus for the deep task. However, for the MCI population, in fact, there was no reliable relationship. So, more wasn't necessarily, more activation or more
10:44
spatial extent of activation wasn't necessarily correlated with better memory performance.
10:49
And a number of different studies have come out of Ray Sperling's group. I'm just going to show you a couple of examples that have left people really questioning this relationship between hippocampal activation and performance, or degree, or integrity of that system. So here, what we can see is that looking at patients with…these are both groups of individuals with MCI, mild cognitive impairment. They're not in the AD stage yet in terms of functional deficits
11:19
in daily life, but they do qualify in terms of their psychometric properties being lower than healthy controls that are age-matched. On the left side, we have people with very mild cognitive impairment, and on the right side, people with more severe, i.e., their scores were lower on standardized tests. Importantly, in each of these maps, what we see is a comparison between that group and healthy control group, okay? And so in this case, what we see is actually in the very
11:47
mild case, you actually have greater activation in the medial temporal lobe bilaterally, in fact, than you do in the healthy controls, whereas in the very severe cases, in fact, that activation has dropped off entirely. So now, it's no longer stronger than healthy controls, in fact, it's about the same or in some instances weaker. So what's going on here? The only thing that's changed is in fact the degree to which their cognitive memory test performance is falling below
12:19
the healthy control norm. They haven't changed a category. And down below, this is a kind of pastiche of a number of different studies from Sperling's lab, but again, just to kind of highlight the important point here that I want to show you. So in this case, up in this figure here, we have what happens in a standard encoding paradigm. What you see in general in healthy individuals is you see, and this is a visual encoding paradigm,
12:45
you see a lot of activation relative to a control task in the hippocampus, in occipital regions, and in medial prefrontal, medial frontal regions. And you see deactivation in these regions, what we know of as the hubs of the default mode network. So the medial precuneus and medial posterior cingulate are deactivated. So that's the standard pattern. Down below, we actually see the difference in this case between individuals who are cognitively normal
13:15
but who either have an APOE4 allele or not. And in this case, you actually see that the E4s are not showing that same degree of deactivation during the encoding task. So, in fact, not only are they maybe not turning up their hippocampus to the same degree, in fact, we don’t know that in this case, but they’re not deactivating other parts of the network that actually should be deactivated.
13:37
And on the right panel here, again only healthy controls, just splitting them up into high and low performing older adults versus high and low performing young adults, and this is all about deactivation of that medial posterior parietal cortex. And you see that both young adult groups are actually showing a good amount of deactivation, but only the high performing older adults are doing that. The lower performing ones, presumably potentially on their way to
14:04
MCI diagnosis, are in fact not showing that deactivation at all. And again, the degree of deactivation in the older adults is telling us something about what’s going on in their memory. So the main point of this slide is really to tell you that a magnitude of activation isn’t necessarily always telling you how healthy—more isn’t always better. And in fact, in the Alzheimer’s spectrum, it may be equally important to look at other parts of the network
14:30
to know what’s going on or to be able to better characterize what’s happening in terms of memory performance. We’ll come back to network ideas again a little bit later on in the talk. But right now I want to focus back in on the hippocampus a little bit. Just again, to tell us a little bit more about the hippocampus in general in terms of what we know both from behavioral studies but from imaging studies as well, what are the key kinds of elements that really
14:56
are the ones that the hippocampus is mainly sensitive to, or responsive to, or engaged on? And so this is a kind of the code that Charan Ranganath has used in much of his work: this idea that the hippocampus is really important for binding of items in context. That is, it takes information, as we’ll see from a wide array of inputs from the neocortex, and it puts those together in particular ways between the particular item as well as its contextual
15:24
features and associations. So, what are the kinds of stimuli or tasks that are the ones that are most likely to show preferential engagement or, in fact, decline relative to hippocampus? And these are the kinds of elements or materials that are often used: elements within complex scenes, word or picture associations, recollection of context, remembering what you thought about when you saw something, or what the picture was paired with versus item familiarity—just saying, "Yeah,
15:55
I remember that item, but I don’t really remember anything about it, what I studied it with, etc." Autobiographical retrieval, remembering events from your own personal past, seems to be, as we’ll see, something that’s really preferentially engaging hippocampus. And a nod to Eleanor Maguire’s work, where she packages much of this kind of information by saying, actually, the hippocampus, because of its primary role in spatial information processing as well,
16:21
is really much more critical for what we might call scene construction. And all of these other things are merely taking, you know, riding along with scene construction, where they involve scene construction to a very high extent. So, these are the things that are mostly engaging hippocampus. We know this from lots of studies of healthy individuals. We’ll go through just one tiny slide, a pastiche of this, and this is an old slide
16:46
from Lila Devachi, and bringing forward this idea: when we’re encountering individual information in the world, we take in the information in our sensory analysis, like this bowl of fruit that we see before us. We know that in the ventral visual processing stream, we actually end up with some information proceeding on through the perirhinal or entorhinal cortices, where we get object encoding that includes some features of that object—the shape, its color, etc.
17:20
We know that other aspects go forward into a more dorsal route through posterior parahippocampal gyrus, where we get the spatial layout of information. And as those things proceed up into the hippocampus, they’re bound together in terms of now those elements are all part of a complex of information. But it doesn’t become that bound item in context until it reaches the hippocampus.
17:45
And there’s some very elegant little studies showing this very long ago, back in the day, probably before many of you were born, showing these kinds of effects. So, for example, we’ll show both an encoding and retrieval in this case, Davachi one, where we’re looking at encoding effects as a function of subsequent memory recognition. So people are given items in context, and there we then break down their original encoding activations
18:17
depending on whether they remembered the item and their source (or the context), the item alone (that’s they’ve forgotten the context but they recall the item), or they’ve forgotten everything about the original experience. And you can see at least for this part, a little area in the hippocampus, it seems to discriminate between item and source. And all other things, in fact, it’s not even that responsive to remembering the item versus
18:41
forgetting the item. So, it really seems to be the importance of its engagement at the time of learning was important for being able to recall those bound bits of information together. And here’s a similar kind of study, but in this case, looking at what happens during recognition of that material and then looking at either activations in the perirhinal cortex or in the hippocampus. And again, the hippocampus, very much the same as up here.
19:07
Now, when we’re recognizing the information, it is preferentially activated when we’re able to recognize both item and source together, and not so much when you’re just recognizing either the item alone or you’ve forgotten it entirely. Whereas in the perirhinal cortex, it seems to be item information that’s most important—whether you recall the source or not isn’t that relevant to the degree of activation that we see there.
19:30
So, we see that even within the kind of the hippocampus and the medial temporal lobe, I should say, there are some segregation of types of information processing and types of representations that are there. But it’s at the kind of apex of the system in the hippocampus where that information all comes together. And we can also see this, here’s a nice little older study—there are many other examples of this showing the same kind of deficit in
19:56
individuals with hippocampal damage. So, it’s always important in my mind not to just rely on functional imaging because there are lots of ways you can go wrong there, but also to pair it up to be able to see, are we seeing the same kind of thing if we’re looking for the kind of nature’s experiment, which is a selective deficit. So, in this case, with Mieke Verfaellie’s lab showing that there’s a disproportionate deficit
20:19
in recognition for conjunctions of objects and location. So, these are individuals who have selective hippocampal damage—they’re not either Alzheimer’s patients or epilepsy patients—they’re individuals with either encephalitic lesions or hypoxic damage to the hippocampus. And in this very simple experiment, you get an object that’s put in a little grid, you’re showing one every second, and then after a very brief delay—we’re not talking days or weeks
20:48
or months or even minutes, talking one second to eight seconds—you’re either given an opportunity to remember the conjunction, so for example, did you see the alligator in that square, or did you see the alligator in that square? Or you can be probed with a location cue—did you see anything in this square? Or an object lure—did you see this item at all? So, what happens under those circumstances? Well, if we look at the difference between
21:12
controls and these amnesic patients in terms of accuracy, they’re doing fairly well at the feature detection, so they’re able to say whether or not they saw the anchor or not. But it’s really the where they’re—and they’re able to say, did I see this square filled or not at that very, very brief delay? But even at one second, they’re starting to show this deficit in terms of being able to do that kind of binding of items in context. So, again, nice information
21:39
showing us a nice example showing us how critical that hippocampus is for that kind of information.
21:45
Okay, give one more example of this kind of binding of item in recollection, and this has to do with autobiographical memory. So, we know from many studies now of both patients and functional imaging that the hippocampus is invariably activated when we’re recalling personal past memories. When given a cue like “first kiss” and asked to think about that while you’re in the magnet, or to describe it to an individual so they can write down your
22:11
transcript and see how detailed your recollections are, the hippocampus is a really key player here. We know that its activation is related to the vividness of the amount of detail that is in that memory recollection. And we also know that patients who are AD or at risk for AD also have less detailed memories. So, for example, in this study, this is individuals with at risk for AD, and we see, or sorry, individuals who have AD, and we see that their autobiographical events,
22:40
for how rich and detailed their memory was, was actually correlated with the degree—the volume of tissue in the right hippocampus. And here, a nice study by Matt Grilli really showing that individuals who are carriers of the APOE4—these are all healthy individuals, healthy older adults—are also showing less detailed memory than individuals who are non-carriers of that. So, even very early in the process, by which we assume that there’s some
23:08
degradation of these memory circuits, we’re seeing that that autobiographical memory seems to be really hit very strikingly. We’ll come back to that in a minute.
23:19
First, I want to raise a couple of complexities in terms of the hippocampus, because so far I’ve really been talking about it as one structure that does the same thing all throughout its length, and all throughout its circuitry, and that’s not exactly the case. So, we’ll talk a little bit about different properties that might be assigned to other subfields within the hippocampus, as in here, as we see that there are the classic projections from entorhinal cortex into dentate
23:48
gyrus, CA3, CA1—we won’t talk terribly much about hippocampal anatomy, but it’s important to know that these different subfields may have slightly different properties in terms of the nature of the
23:59
computations that are carried out there. And so, one thing that you probably have heard about is this notion of pattern separation and pattern completion. These we believe are two intrinsic properties of hippocampus, but they may be geared towards slightly different subfields, or slightly different subfields may take more of the bulk of the work in terms of these two different processes. So, in terms of pattern separation, this is the idea where we
24:22
see two somewhat similar inputs, and yet the hippocampus is able to actually generate two orthogonal or completely separate outputs, so that one can really distinguish A and A prime. In pattern completion, it’s kind of the converse property, where again there’s overlapping inputs, and in this case, you’re actually able to use either A or A prime to get back to essentially the same representation. And Craig Stark's group has done quite a bit of work showing that the
24:52
relationship amongst the hippocampal subfields and how they're tuned to one process or the other. So, in this little graph here, what you can see is, for example, a minor change in input can evoke a very big change in output, which essentially is what this is, right? Showing that this area of dentate gyrus seems to be really important for laying down those orthogonal memory traces, whereas other regions of the brain, such as the CA fields,
25:22
are a little bit more tuned to pattern completion, although CA3, as you see, may take slightly different characteristics depending on the tuning curve. And how do we actually understand that in terms of fMRI? Well, you actually take advantage of a little trick that we know about in terms of hippocampal activation. So, the hippocampus is generally, we've found, not that it's very interested in novel things, okay, in
25:49
a new learning experiment. So, you can easily show that if you repeat presentations, or you repeat a certain stimulus in a very short window, you can actually find that the amount of activation in the hippocampus is reduced. It's less interested in re-encoding the same periscope because it's just done it already. Whereas if you present maybe a slightly different kind of stimulus, you might
26:13
get a resurgence of hippocampal activation. So, in theory, you could have two different kinds of patterns of activation. If you're looking for a system that does pattern completion, what you might find is that a lot of activation on the first presentation, but both of these are treated as essentially the same by the hippocampus, and so activation goes down. If you're looking for a system that's tuned toward pattern separation, you might find, conversely,
26:38
that you find the repeated presentation goes down, but the other lure is now treated as a novel item because it's different enough, right? So, a separation would look like, would not show the same kind of activity for a completely different lure versus the repeated presentation. And some very nice work out of that group has actually shown that exact pattern. So, within, right CA1, it essentially works as a pattern completer,
27:12
so it's not at all discriminating between the exact repetition and a lure that's somewhat similar. Whereas in dentate gyrus and CA3, they're hard to resolve right now with our current neural imaging, you actually see that that pattern separation is taking place. So, that's one kind of important aspect to keep in mind about the hippocampus: it's probably not doing exactly the same thing in each of its different subfields, and that may be important for
27:40
diseases or disorders where different subfields are preferentially attacked. We’ll Complexity number two is that we're very aware now that it's actually also not homogeneous along its anterior to posterior axis, that there seem to be differences in terms of the representations that are there, in terms of the processes that take place, and part of that has to do with its connectivity. So, here we can actually see that if we can, we can essentially
28:07
distinguish between the connectivity of the anterior hippocampus and posterior, in blue, and they're going essentially preferentially to different regions of the brain, with anterior more to anterior temporal and orbital frontal cortex, and posterior more to these posterior regions of the brain, including that posterior cingulate, precuneus area that we talked about before.
28:29
And nice kind of theoretical paper by Jessica Robin and Morris Moscovitch, talks about this in terms of a retrieval of complex events, like autobiographical memory, that there may be differences in terms of the kind of nature of the representation in the anterior versus posterior hippocampus. The anterior may be geared more towards the gist of information, the posterior towards more detailed information. So, that when we get a kind of cue to allow
28:56
us to remember a particular event, the anterior hippocampus may give us the kind of, or may allow us to retrieve that gist of information, but not the more fine-grained details if there's damage to the posterior region. And I'll show you some examples of that in just the next slide or two. In fact, the next slide. Okay, so are there meat on the bones of this kind of idea? Okay, yes, they do talk to different regions, but do they really have different properties? Do they have
29:22
different representations? And I think in one of your earlier lectures, you heard a little bit about the idea of place cells and grid cells, so kind of different granularity of representations within the medial temporal lobe, and this is another way of looking at granularity. In this case, this is essentially just looking at the correlation either spatially or temporally at a voxel level within the hippocampus. So, in inter-voxel similarity, it's simply looking at how
29:48
the waveforms across time at rest correspond to one another. So, one voxel, its nearest neighbor, its next neighbor, etc., how close do they correspond? Are they doing the same thing across the temporal time that we're, the time frame that we're looking at, or are they actually, do they actually seem to be undertaking different kinds of processes? And temporal autocorrelation is another way of looking at this. This orthogonal way is to say, given one single voxel,
30:19
how much is its processing changing over time? Okay, so is it shifting over time? Is it doing something different from time one to time two? And it turns out, this is some nice work by Eva Brunec in Morris's lab, that you can actually show this distinction within the anterior and posterior hippocampus. So, the anterior hippocampus seems to be much more at a grosser kind of gradient, so the temporal autocorrelation there is much stronger. So, it's the voxels are doing very
30:54
similar things across time, and they're also doing very similar things to their nearest neighbors, where there's a much greater granularity of information processing in the posterior hippocampus. So, that's an interesting difference there, again, in terms of anterior-posterior.
31:10
This is an experiment, a recent study from our lab, which is essentially, again, capitalizing on this, but instead of looking at voxels at rest, we're actually looking to see what happens when you're asked to retrieve a context given an item and how is that representation of that context, how does it differ in the anterior and posterior hippocampus? So, in this case, individuals study
31:35
pairs of objects and scenes, like the seahorse on the background of a beach, and then they're shown the objects again at retrieval, and they're asked to remember, imagine your mind's eye, what that was paired with originally. And then you can see across these various things, you can compare the activation or the activation in different parts of the hippocampus, the type of representation using this multivoxel pattern analysis approach to be able to say, what does
32:04
the hippocampus look like in terms of anterior and posterior? So, in this case, what happens is that in the anterior hippocampus, again, it does not seem to discriminate between seeing, remembering exactly the same scene or a similar scene. The granularity is quite broad, but it certainly differentiates the beaches from the kitchen, whereas in the posterior hippocampus, it actually is able to distinguish between those two different beaches that it was seen. So, again,
32:32
that suggests very much like Brunec's experiment that there's a difference, a distinction from the anterior-posterior axis in terms of that degree of granularity. Okay, for the rest of the talk, then, I'm really going to kind of move. The hippocampus will still be a player because it's always a player, but we'll move away a little bit from it on its own to the other players that are involved in remembering information.
32:56
So, regions that are activated when we retrieve personal memories, there's a long, beautiful long literature on this now in terms of the hippocampus not only acting on its own, but also very
33:09
co-activation of key regions in lateral temporal cortex, lateral parietal cortex, and these, interestingly, these hubs of the default mode network: the precuneus and the medial prefrontal cortex. So, all of these seem to come online when one is remembering this kind of information, this rich autobiographical details about your own personal past.
33:31
But we know from many studies, both imaging and
33:36
lesion studies, for example, that there are encoding and retrieval interactions between hippocampus and various swaths of different parts of the brain that seem to convey different aspects or attributes of our memory phenomenon, right? So, if we think the medial temporal lobes is involved in that binding of information together, well maybe the dorsolateral prefrontal cortex actually adds some degree of organization to the material to be remembered, and maybe the ventral medial
34:03
prefrontal cortex has a different kind of process that it engages in. And similarly, at retrieval, these different regions interact with the medial temporal lobe, so they're not all doing, even if they're all co-activated at the same time, they're not all doing the same thing. They contribute different aspects or different kinds of processes to the overall extent of our being able to remember these kinds of information. So, we can also deconstruct, even without fancy
34:30
neuroimaging, we can talk about these regional contributions to memory deficits in patients. So, I'll just show you two examples of that, because again, many of us aren't going to be working on doing fancy neuroimaging experiments, but we might want to know how we might be able to say, are these different processes involved in our patient population? So, here's an experiment by Melanie Cohn, one of my collaborators, where she was looking at Parkinson's disease and
34:55
a task that she had designed and had shown in previous studies to be very
35:03
sensitive to hippocampal engagement. This is called an associative recognition memory (ARM). And what she does is to get people to study two words, make up a sentence, I use the pillow to sit on in the temple, and then later to either present the item alone, temple, and see, did you see that word or not, or to present the pear, temple, along with pillow, and see, did you see temple or not? Now, you don't have to actually retrieve, work hard to retrieve that association,
35:30
she just wants to see whether or not it was bound together in the first place and whether that binding actually conveys some additional benefit in terms of hippocampal engagement. So, she looks at this what's called a reinstatement score to say, okay, how much more did you get if you saw a pillow, temple, than if you saw a temple on its own? And again, all you had to do was recognize the word temple. And so, in recognizing the word temple,
35:55
controls in PD are exactly at the same level, but in fact, at this reinstatement, she was actually able to show that PD were not benefiting to the same degree, suggesting to her that there might be actually a problem in terms of medial temporal contribution, not so much in terms of the contributions we typically think about in terms of Parkinson's disease, all of those executive components that are important for memory. And again, in the study, what she's able to show,
36:21
I'll draw your attention down here in particular, she, she looked at this associative reinstatement effect but also looked at some parameters from the Hopkins Verbal Learning Test - Total Learning, Delayed Recall, and Recognition- and what she was able to show was that these tests, the Hopkins ones, each of these components seems to be fairly dependent or fairly highly correlated with executive memory components. So that might be part of what's feeding into poor performance
36:49
on the Hopkins, the executive proponent, but the associative recognition was not at all associated with executive components. So, it's truly a memory measure, but it's not picking up so much on those executive components, which again, very important to know about.
37:03
And here's just a nice example, because they're fairly rare, of an individual who has fairly selective perirhinal cortex lesion with an intact hippocampus, studied by Stefan Köhler’s lab, and looking at differences between recollection and familiarity. Remember, I told you earlier that recollection seems to be mostly dependent on hippocampus, whereas familiarity we thought from imaging studies was more dependent on that perirhinal cortex. And this patient NB, who
37:32
you see here in the open circle, shows this really interesting and very profound dissociation. So, her performance is as good or better than a range of healthy controls on the recollection aspect of things. She's actually able to remember items in context, but her performance on the easier, some would say familiarity test, is actually impaired relative to healthy controls. So, that suggests that you can, in fact, find these principles of dissociations between different
38:02
aspects of the memory network in terms of what's contributing to overall memory performance.
38:09
Okay, autobiographical memory. I've studied this for a long time. This is just some very, a few studies showing here an effect size, essentially telling us that, in fact, whenever we look at these internal or episodic details—that is, when you ask people, "What were you wearing? Was there a person, you know, what was the weather like?"—as opposed to, you know, just the gist of the event, patients with medial temporal lobe epilepsy have deficits
38:35
relative to healthy controls. Okay, an effect size of just where they fall off relative to controls.
38:42
When we look at a closer look at the kinds of problems that they're having in general, and here's some left and right TLE patients just from one additional sample, we find that they're not too terrible at the overall gist of the event. They can kind of give you the bare bones of what happened, but it's in fact the very fine details of things like perceptual details—what were you wearing, thoughts or emotions you had at the time,
39:05
temporal information—that's where they start falling apart relative to healthy controls. Again, I'll remind you of this lovely slide from Robin and Moscovitch, showing this. In fact, these data contributed to this lovely slide, showing that the anterior hippocampus in particular might be interacting with this ventral medial prefrontal cortex to come up with the gist of events, but in fact, in order to get these very fine details, you really need communication
39:30
between the posterior hippocampus and that neocortex, where those event details are stored.
39:37
This is just to remind us, because we're moving into the... we're talking about networks. In fact, this is an early study showing that patients with left temporal lobe epilepsy—all we're asking them to do is to retrieve events from their own personal past and comparing it against a bunch of different control tasks—you see here controls showing this nice activation of this autobiographical memory network. In our patients with left TLE, notice that they're not just not
40:03
activating that left hippocampus. In fact, they're showing a downregulation of the rest of the network, and this has been replicated a number of times. So, relative to healthy controls, once even that one bit of tissue is compromised, the rest of that process is degraded, basically. I'll just, just a tiny little point to raise that I find interesting because it's different from other clinical tests. We initially thought, "Gee, this will be a great way of probing our
40:30
patients in terms of, you know, the kind of standard things we want to know about
40:35
them: how bad off is your hippocampus, and will things get worse after your surgery?" But we found over many different studies, there's no difference between left and right patients, and there's no change from pre- to postoperative status. So, in fact, it seems like there's something deranged in the network itself, as opposed to just the contribution of that hippocampus. And we did get a little bit of insight from that somewhat more recently
40:58
by looking at what's happening in our patients with temporal lobe epilepsy. So, in this study, what we're able to show, again, not surprisingly, healthy controls are showing activation, greater activation in that left hippocampus and other parts of the network, but patients are actually showing greater activations just in different regions of the brain. So, they're showing it in this medial prefrontal cortex and in temporal neocortex, etc., regions that we know are involved
41:24
in the network overall, but they aren't the ones that are primarily driven by that medial temporal lobe hippocampus. And when we looked a little bit further, we actually were able to show that the anterior and posterior hippocampus were actually not behaving in the way that healthy controls do. So, for example, in healthy controls, what we find is when we ask them to just construct an event—say, "Okay, who was there, just tell me once you've got that event in mind,"
41:51
versus elaborate on that event—tell us, "Who was there? What were they wearing?” etc. We find that there's a difference, a kind of, between anterior and posterior segments of the hippocampus.
42:04
Their connectivity to one another changes, and their connectivity to the outside world, or the rest of the brain, changes. But, in fact, medial temporal lobe epilepsy patients, who might be activating this or showing strong connectivity, I should say, between these regions, don't show that variation. So, something has gone wrong in that circuitry that doesn't allow them to both
42:26
communicate within the different sectors of the hippocampus, but more importantly, communicate outside to other areas of the network. Okay, one other place that I just want to mention about hippocampal-neocortical interactions that we know are also important, not just for retrieving those kinds of memories, but also for the idea of consolidation of information over time. There's been a long-standing literature, starting with Larry Squire, suggesting that maybe, when
42:52
we first learn information, the hippocampus is really important for doing that initial encoding of information. But over time, information is essentially consolidated in regions of the neocortex such that if you lose your hippocampus, you might still be able to retrieve that. Now, I've already told you, you can't retrieve the very fine details of autobiographical experiences, but how about other kinds of information? So here, we capitalized on something called
43:17
accelerated forgetting in patients with medial temporal lobe epilepsy. Here, they're simply doing an item memory task, and what we see here is the proportion that's forgotten over time, relative to healthy controls, for 90 minutes, all the way up to 72 hours. And you see that even at 90 minutes, when we can fairly well match performance, they start falling off very, very rapidly. So,
43:39
this seems to—and this is well outside the window of opportunity in our standard neuropsychological assessment, right? It seems to be happening over hours and, potentially, over days. And when we look to see, well, what can we understand about their connectivity between the affected hippocampus and the rest of the brain? What we found was that the connectivity that they were showing (this is in resting state, it's not while they're doing the task), connectivity
44:06
between the affected hippocampus and the temporal neocortex on the other side was actually something that was strongly related to the amount of forgetting. So, the stronger the connection, the less the forgetting. So, once again, it's telling us that this hippocampus is not working as an island on its own. In order to facilitate that kind of hippocampal consolidation, neocortical consolidation needs to have that degree of interaction between the rest of the cortex.
44:33
Okay, and for the last five minutes, hopefully, what I'll be talking to you about is this idea of
44:39
networks that we can identify when patients aren't actually—or individuals aren't actually doing anything. They're, in fact, what we might say, at rest. And so, these various resting-state networks, again, can be identified in the brain by simply looking at the correlation of time courses within different parts of the brain, different regions. So, if I drop a seed voxel in here, I can very readily show you the different regions of the brain that we now understand to be the
45:04
default mode network that correspond over time. Their temporal fluctuations at very low frequency are tightly bound together, suggesting that they actually comprise intrinsic networks in the brain.
45:19
Okay, so is it useful to know anything about this? Can we leverage this idea of these default mode networks or any of these resting brain networks? So, we wanted to look at a very simple idea of the connectivity between the hippocampus and this hub of the default mode network, this posterior cingulate. And this is some early work showing, not surprisingly, that these blocks here now are showing where activation—not activation, but connectivity- between these two regions is
45:48
different between controls and patients. So, what we see is, not surprisingly, a decrease in connectivity on the affected side, which we expected—less activation, less connectivity—all makes sense. But we also saw an increase on the contralateral side, which we didn't necessarily expect, but thought, "Well, maybe that's some kind of compensatory response." We also wanted to say, "Well, does it tell us anything about memory?" And in this case, using just a very simple face
46:14
recognition task, in these patients with right temporal lobe epilepsy, we saw that pre-operative performance—pre-operative connectivity between these regions, not activation, but connectivity—is actually nicely related to the performance on the task. So, the stronger the connectivity, the stronger the bond between the two of those regions at rest, the better the memory performance.
46:38
We also were able to show this whole idea of, "What will happen following surgery? And does this at all predict anything about memory performance following surgery?" So, the idea of functional adequacy: again, the stronger the connectivity was before surgery, the more you stood to lose. So, that shows the idea of functional adequacy. The idea of functional reserve: looking at the contralateral side, the stronger the connectivity there, the better maintained
47:05
things were. So, that was very encouraging to us as a kind of first pass at connectivity. But I will say I'm actually just going to skip the slide because it's a little too busy for us at this moment. I'm just going to say that we wanted to look at the idea—or we've now explained—I just want to leave time for questions. The idea of expanding this now to look more broadly at network properties. Because one of the things that we also identified previously was, the more you looked
47:33
like a healthy control, that might be telling us something really important. So, here's—this actual study isn't yet published, although it's just about to be submitted, so I might take this out of the slides. At any rate, we wanted to look at this idea of looking at just this resting-state network of regions that we believe to be involved in memory performance from other a priori reasons. What we do is look at the connectivity amongst all of the nodes in the network. We gather up
48:02
a bunch of healthy controls to look at that connectivity and form a normative template. And then we can look at each individual patient and say, "Okay, how close is your memory network to resemble a healthy control?" So, it's kind of like a psychometric approach. And what we're able to see is that your change from before to after surgery, is strongly related to how healthy your network looked. In fact, in kind of a counterintuitive way, the more healthy it looked,
48:32
the better off you'd be following surgery. But importantly, when we look to see, "Well, what regions are contributing to that?" It's actually regions outside of the hippocampus. It's actually medial and lateral prefrontal regions. So, in fact, it's not so much the damaged tissue that's contributing to that story—not surprisingly, it's the rest of the network. So, what's that telling us? That the rest of the network is actually pretty important to be able
48:58
to consider when we're thinking about what's going to be happening to individuals following surgery. And in fact, here, this study is actually just following up on one that has already been published in terms of language networks, showing very much the same kind of thing.
49:14
Okay, very last slide. Before a conclusion, and this is just to wet your appetite. Okay, so now you've shown us a whole bunch of network stuff. Who cares? Right? Like, I'm telling you—we might care in terms of the cognitive neuroscience thing. There may be reasons to care about it in terms of the clinical neuroscience as well. Can it help us predict what might be happening following surgery or some kind of intervention? And I will say that there are groups now, like Joel Voss's,
49:39
who are using this kind of network neuroscience in terms of targeting areas that might actually be able to change memory performance. So, in this case, using transcranial magnetic stimulation, knowing that you can actually, with certain targets of networks—not so much to the hippocampus, because you can't reach that deep in the brain—but if you can actually reach regions that are strongly interconnected in each individual subject, you may actually be able
50:07
to exert both local effects, which are usually disruptive, but also network-wide effects if you give TMS a number of times before training. So, in this example, what they were able to do was to get people to study and then test their memory for object-location associations. Their TMS targets were here, and here are the regions they were hoping to be able to hit within the hippocampus. And if they looked at performance on this task before and after stimulation, either this type
50:36
of stimulation or sham stimulation, they showed a dramatic improvement in the stimulation. So, people were much better, these healthy individuals, much better at doing the task after stimulation than sham. Finally, they showed in older adults, in another study, that that actually benefited this recollection capacity. So, again, something that's dependent on the hippocampus showing this recollection change, quite dramatic, relative to sham stimulation,
51:03
and actually showing activity in the hippocampus was increased in the hippocampus following that type of stimulation. So, showing you can target areas that might be dysfunctional with TMS. We've done somewhat similar things with deep brain stimulation as well. So, conclusions. Obviously, the hippocampus is the hub, the synoquinone, the master of all masters in terms of episodic memory. We know that it has intrinsic properties of pattern separation
51:31
and pattern completion, and that these enable fine-grain encoding and associative retrieval. Those are the two major things you need your hippocampus to do. We also, I think, are coming to know that a full explanation of memory functioning and memory deficits requires that we appreciate the interactions of the hippocampus and other parts of neocortex to which it's intimately connected. And hopefully, at least I've been able to give you a slight taste of the idea that
51:56
using fMRI, including looking at some of these properties of intrinsic connectivity, allows us an opportunity to explore those interactions and potentially intervene therapeutically, and to predict what might be happening with our patients with respect to those therapeutic interventions. So, just thanks to all lab members, etc., with much gratitude, and much gratitude to all of our funders and our patients.
52:22
I'm happy—only seven minutes, I'm sorry—but I'm happy to take any questions that you might have.
52:36
Thank you. This was such a wonderful presentation. I do research for Carrie McDonald on temporal lobe epilepsy, so I was just delighted to see some of these papers.
52:47
So, with that, giving, you know, I think a lot of our trainees learn about the Wada, right? But we're now realizing that fMRI may be a better tool. And as you were explaining the anterior versus posterior part, and knowing that the Wada procedure doesn't anesthetize that posterior part of the hippocampus, are we potentially missing important information that could be predictive for a post-operative outcome? Yeah, good question. It's always been a little bit
53:14
of a nag to anybody who does a Wada. I suspect, not for the following reason: One of the reasons is some early studies, by Jean Gotman and others at the MNI, had actually shown because if you have depth electrodes in the anterior and posterior compartments at the time that you're doing a Wada, you can actually show that you functionally deactivate the posterior hippocampus. So, even though the circulation isn't approached through our typical Wada injection,
53:43
you still kind of get the effect there. Oh, that's really good to know. And we have another question here. It was mentioned that familiarity is dependent on the perirhinal cortex, so should we see recall relatively dependent on this region?
54:02
So, probably so. So, recall—you mean just—it's much easier to demonstrate it in recognition tasks, where you have designed the test in order to really emphasize fine familiarity discriminations, right? So, for example, it's not only shown in memory tasks, but also just in kind of an oddball task where you show, you know, objects or faces that are, you know, quite close to one another. So, it seems to be important for binding features of objects together. So, if you
54:43
disintegrate those in your recognition test, it's easy to find that dysfunction. But it doesn't seem to be as important in terms of the contextual information, right? Because that's actually coming through parahippocampal cortex. So, you're very clever, like Stefan Köhler, you can design studies where you can get a patient to show absolutely perfect recollection and terrible familiarity. But in our standard memory tasks, that's very rarely the case that
55:14
we're going to be, you know, compartmentalizing those kinds of processes quite so keenly.
55:21
I think that's a tremendously useful point to bring up, though, and I'm glad that you're bringing up these points, because this is a finer level of detail than I think a lot of students learn in their neuropsychology/neuroanatomy courses. And it's very often that I'll see patients have some of those odd performances where they do worse on that recognition.
55:44
And something that you said before, also with the binding of detail and kind of integrating that, really made me think of some recent research that I was reading about the Cookie Theft task and how they're using kind of intelligence, artificial intelligence-based software to predict these performances. And some of the unique things that they were finding is, you know, lack of detail, lack of understanding the associations, kind of putting all the pieces together,
56:14
was very predictive of the earliest signs of a neurodegenerative process, such as Alzheimer's disease. And so, I'm just kind of putting these things together in my head, and I wonder if you could— I know that a lot of this is from the clinical context—but I wonder if you could maybe just comment on the clinical implications of kind of what you're saying in terms of the neuroanatomy. Sure. So, for sure, I mean,
56:39
I think that, as I said, what's always surprised me about things like autobiographical memory—and you can do this very easily too with more complicated kinds of materials. So, we've also done it with showing people little film clips, for example. So, you don't have to, you know, ask people questions about their own personal past if you don't want to. But, being able to recollect these kinds of things in detail, it's always surprised me that we don't see very
57:04
much difference between pre versus post, or left versus right. Or, in fact, it shows up very early in terms of, as you say, in terms of, you know, patients with APOE4 alleles, etc. So, it does, to me then, seem to be very much a principle that the hippocampal component is required, but what's truly required is its interaction with other structures, right? Because it's kind of easy to find a hippocampal deficit other ways. And one that actually scales very well with
57:37
these other principles that I've talked about, right? So, it's kind of easy to see, "Oh my gosh, this person has a massive amount of hippocampal atrophy, I bet they'll be very bad on a, you know, X task or Y task," right? But what I'm telling you is that even the ones with minimal hippocampal atrophy will be very bad at this particular task, right? So, it does, to me, suggest again that it's these network properties. The reason I showed that idea of that interaction between left,
58:03
right, anterior, and posterior in a healthy brain, you're actually kind of communicating internally within that component of the network, but then also externally, you know, to medial prefrontal and posterior parietal. And if you've lost the capacity to be able to do that, then that task may fall apart much, much more readily than other kinds of tests that we use in the clinic. So, I felt bad about, you know, spending a long time of my life looking at it and saying,
58:34
"I can't possibly use this in temporal lobe epilepsy, because my clinical job is not to ask about anterior, posterior, blah, blah, blah. It's about to say whether or not you're at risk for a big memory deficit after your surgery, and this isn't going to tell me that." So, it's interesting from that perspective, but I do think that that other example of kind of looking, you know, we…one of the reasons why I personally use things like the RAVLT
58:59
for temporal lobe patients, rather than the CVLT or stories, is because I know there's a huge contribution of other parts of the brain to those particular tasks, right? Frontal lobes,
59:13
lateral temporal lobes, etc. And if I'm really interested in knowing what's going to happen after a hippocampal resection, that's not helpful to me. So, I really have to be able to use something—and RAVLT is a pretty good pattern completion test—because, you know, you use bits of the information. I've had many people make up stories about turkey farmers wearing hats.
59:34
I had one last question that follows really well into that, and it's about the clinical utility of the pattern separation task. I mean, we're seeing a lot of fascinating research, both in temporal lobe epilepsy but also in MCI and Alzheimer's, and it's showing that it's actually more related to episodic memory than some of the tasks that we use in the clinic. Yeah, no, it's a great question. So, and I'm always reminded by Craig and others to say that
1:00:02
pattern separation is the hippocampal function, the mnemonic similarity test is the test of that hippocampal function, right? So, and that's true. I think, you know, that test is really very interesting to be able to, again, show those deficits very early on. I'm pretty well persuaded by people like Arnold Bakker, who’ve told me, "I’ve tried it with Alzheimer's patients and I find it impossible. They cannot understand what you're asking them to do." And he says that even
1:00:31
once you get kind of more severe MCI, they cannot understand, you know, they're falling apart. So, the kind of standard pattern separation task is very difficult for them to be able to do, that, you know, old vs. new vs. similar kind of thing. But it may be a good harbinger of what's happening before that. So, it may be able to pick up, you know, very earliest signs of deficits before that. And the one piece I'll bring back to imaging or to anatomy is we now know it's a really
1:01:01
good way of looking for what we think is a pathological sign of hippocampal hyperactivation. I showed you that slide from Reisa Sperling showing that you actually, very mild MCI patients actually show more activity there in the hippocampus than healthy controls. And originally, people thought, "Well, maybe that's compensatory." It's now clear that it's probably pathological, and there are studies, a number of small studies and a big trial right now,
1:01:26
giving in a very small dose of an anti-convulsant medication, levetiracetam, which actually reduces that hippocampal hyperactivity and also improves memory performance on the task. So, I love fMRI, but you know, beware of using it because you might be telling yourself a false story. So, it's always really important to be able to look for the same kinds of things that you think you're seeing in terms of those kinds of correlations with lesion kinds of experiments as well.
1:02:01
Well, thank you so much. This was a fascinating talk, and I may be a little bit biased, but thank you so much for coming and talking to our viewers. And with that being said, we have our last lecture of this series next Monday, so make sure that you join us, and the recording of this lecture will be up next week. Thank you!
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat