[00:00:09]
>> Thank you Lou. Thank you very much Annabel Thanks everyone for coming Can I start out with a show of hands how many people are students undergrads or graduate students a good lot that's great whenever I prepare a talk I would assume there are going to be lots of students in the audience so I hope I've hit the right target there's the title of my talk.

[00:00:31]
I'll try to explain what that means over the course of the talk I work as a Genelia research campuses and I will mention this is a very special place I could spend an hour just telling you about Genelia but I won't let me just say that these are the people in my lab who do the work there's actually you'll see 2 of them had listed here a lab coordinators and 2 are former post op so we have 6 post office in the lab and we have small labs at Genelia but we get a lot done and large part because we have this tremendous resource called the shared resources and every post doc in my lab benefits from men to sleep from help that they get various people in the shared resources so I really want to give a talk not just my lab It's the my lab plus all these other folks to help us get stuff done OK I'm going to talk about brain function today and deconstructing how memory works of the soluble level so get your hopes up I don't have all the answers I'm just going to tell you a few things that we've learned along the way and I'm going to break the talk up into 3 parts you guys are guinea pigs I've never done this talk before in the 1st part I'm just going to tell you you know what the hell are we doing what are we trying to accomplish the 2nd part of the talk about some work that has pertain to the formation of memories and then the 3rd part which is all fluid number 2.

[00:02:02]
About the use of memories. Has Happened To make slides late at night the night before the seminar OK So what are we doing. I'm going to I'm going to show you this video to try to give you an introduction to what we're doing I shot this video with my i Phone yesterday at my in-laws house my in-laws recently moved from Texas to Virginia you'll recognize the strong Texas accents and this world is on a squirrel proof bird feeder and it's squirrel proof because the bar that the birds sit on is spring loaded and if a squirrel stands on that bar closes it's access to the bird team so it's squirrel proof here we go.

[00:02:50]
To go. All. This because got it all figured out. So I think this video is pretty cool just because it illustrates that animals are really intelligent when they figure out what they need to survive they're going to make it happen one way or another and I think as neuroscientists what we want to understand is how the brain and the body working together allow animals to do such incredible things not just physical feats like jumping up on to the feet are but also the cognitive feat of realizing that it has to hang from its back legs it doesn't put too much weight on on the bar.

[00:04:16]
And there are many approaches that people are taking to try to understand intelligence of course is a whole field of artificial intelligence and bioengineering where people are trying to engineer intelligent robots for example there are systems neuro scientists and cellular neuroscientists who are interested in understanding how the brain works and for me personally I really would like I would be find it very satisfying if we could achieve an understanding of the brain at all these different levels so what is the relationship between artificial intelligence and real intelligence how do real brains actually produce intelligence and produce the complex behaviors that emerge from intelligence and what are the properties of the cells to make all of this pop possible as well as the connections between the cells and this is a very traditional way of thinking about understanding complex systems in biology is to break it down into the cellular machinery perform that implements complex functions and try to understand the sit the system in terms of its cellular composition so that is very much my goal as a neuroscientist and I think it's maybe not the same goal that everyone in earth science has but I want to start out by telling you where I come from of how many people in the room don't who this is.

[00:05:39]
Well you got only a couple that's great OK this is Lisa Dole the single is the 18 time world champion at the game of Go which is an ancient Chinese game said to be the most complex game in the world and there are far far more many possible iterations of moves in a game of Go than there are atoms in the universe and therefore it's impossible for a computer to compute all possible moves in any given situation and therefore is much more difficult for artificial intelligence to to actually get good at playing this game so a couple of years ago Google didn't mind develop this artificial intelligence system called Alpha go and they pitted it against the top professionals ultimately.

[00:06:32]
Having having it play against Lisa deal and there's a netflix movie about this I highly recommend that you all watch it it's an amazing. Account of the competition between Alpha goal go and lease the deal Alpha go won the match 4 games to one and it was emotionally shattering to Lisa Dole and really to tens of millions of people who watched this happen in real time to the notion that a computer could be a human at something that required so much deep into a should formed over the course of a lifetime was really shocking to people and I think that the movie conveys that really nicely.

[00:07:16]
Alpha goes into the top did not last long however because about a year later the same group google Deep Mind announced Alpha go 0 which instead of learning from observing humans playing Go games of served are learned by playing against itself and therefore with really little or no human knowledge of humans playing the games this alpha goes era was able to learn the game from scratch and Alpha go 0 was pitted against Alpha go and it won $100.00 to $0.00 so there is now really no chance that a human is likely to be an artificially intelligent system at the game of go anymore so that's done why do I bring all this up I bring it up because.

[00:08:08]
It raises an interesting question which is if we can create intelligence in a computer does that mean that we understand it and is that enough for us to say OK well now we understand intelligence I would say I saw someone shaking her head over here I would agree absolutely not we don't understand it even the people who made Alpha go and Alpha go 0 don't really understand it now they probably have some really valuable intuition into how these kinds of systems these deep artificial neural networks function and therefore they may have some intuition that would be really valuable for us to tap into as neuroscientists in terms of trying to understand how these complex systems actually function or may function in a real brain but there are still things the real brains can do that artificial intelligence systems cannot do and it's a very interesting question why it is that the human brain for example is able to learn things much more quickly even though these systems are very intelligent they require an enormous amount of training that happens as fast computers for very long periods of time and these are things that humans can actually learn things that humans can learn quite quickly like small children can identify the difference between a dog and a cat be easily but artificial intelligence systems have to be trained with millions of pictures of dogs and cats before they can identify them reliably So here's what an artificial neural network looks like it's typically layers of blocks of elements which sometimes people call neurons and these neurons are very simple device units that get input from other neurons they sum them linearly to some threshold and it's a surpassed some threshold of input then they produce output to the next level and by implementing rules about how the weights of the connections between these neurons change in these artificial neural networks this is sufficient actually to produce artificial intelligence.

[00:10:13]
Real neurons are really not like that so this is kind of a cartoon diagram of a neuron in an artificial intelligence system it's just an element that gets the input and an element that produces output probably those if you study neuroscience recognize this is something like a cell body and if in apps and real neurons are no more like this then this is like a real human OK all of you are very familiar probably with what real neurons look like these are 3 neurons that I studied during my Ph D. about 30 more than 30 years.

[00:10:53]
And at the time I became absolutely fascinated with the structure of the dendritic tree of neurons and why it is that they have such an elaborate structure and why cells in different parts of the brain these are all 3 different cell types in different parts of the hippocampus have such differences in their structure.

[00:11:12]
And this is my view of how Durand's work is of what they look like for a long time of course they'll have an X. on that goes somewhere but I really think too much about that and then along came the mouse lite project edge nearly a where the idea was to trace out the X. phone arbors ations of individual neurons with cell cellular resolution throughout the entire mouse brain and I'm going to show you this movie of 3 neurons in the amount of motor cortex of the mouse just because I think this illustrates how incredibly complex individual neurons are even in a mouse brain so these 3 neurons each have about 40 centimeters plus of X. on associated with one cell body so imagine one nucleus has to produce all the machinery to provide the proteins to these axons to support the function of thousands probably tens or hundreds of thousands of synapses these are now the lamb a quarter call neurons that you're looking at and.

[00:12:12]
There's some neurons in the cortex that descend into the pyramidal tract the neurons are incredibly complicated so not only the dendrites dendrites complicated the axons are complicated and I think we all should kind of open our minds to what neurons are capable of in animal nervous systems they're there doing something that's much much more complicated than the simple linear integrated fire operation that's typical of neurons and artificial systems OK So where am I going with all this.

[00:12:46]
In my lab we study the hippocampus because the hippocampus is such an important structure in memory as I told you we want to understand the cellular basis of hippocampal function and its contribution to memory there's a lot of evidence and there's a lot of evidence in both rodents.

[00:13:19]
Rodents and humans that hippocampus is an important contributor to memory. These are some neurons in the big Killam of the hippocampus which is a brain region all talk about more extensively in a moment they also have very diverse a very complicated. Elaborations of their acts on arbors a sions targeting many different parts of the brain and one point I want to make here is that there's a lot of diversity in the cell types in the hippocampus and the hippocampus is sending output to many different parts of the brain so we really shouldn't think of the hippocampus as doing anything on its own it's part of a much much larger system which is the nervous system as a whole.

[00:14:01]
OK so we started studying the diversity of neurons in the hippocampus and we used a method called R.N.A. C. which allows us to determine all the genes that are expressed in any particular cell and we've done several studies using this method and we've created this database called the hippocampus R.N.A. seek atlas or hippest and there are a number of different data sets in this web site that you can go to and explore the data and I just want to show you that this in this one study Mark to browse key found that in the analysis of single cells in gene expression and single cells that in this plot here which is called a tease me plot it looks almost exactly like a map of the globe.

[00:14:48]
India is missing it seems but each of these little clusters of continents of cells are different cell types and we were able to map these clusters of gene expression onto physical locations in this expansive structure that we call the hippocampus and so we've learned that there is tremendous diversity in this cell types in the structure projecting out of the hippocampus to different parts of the brain we also have identified diversity in across the different regions of the hippocampus So for example dentate gyrus ca 3 CA one dorsal and ventral and several other aspects of the diverse collection of cell types have been probed in depth using this approach called R.N.A. seek and I don't I don't want to talk about this in detail today I'll talk about a little little bit later on but I did want to tell you about this resource.

[00:15:40]
OK so I've told you that there's a diverse collection of neurons in the brain including in the hippocampus they have subserve complexity their functions they have complex connections and what I didn't mention is that we should also not forget that there are also asked to sites in the brain that's another important cellular component of the brain and we have recently published a paper showing that neurons and after sites interact in meaningful ways that may mediate some signal transduction and information processing as well OK So that is a sort of fairly long winded interruption to how I view the complexity of the brain now let's talk about real data on forming memories and using them arrays so we're going to talk a lot about the hippocampus and to put you into the right context I want you all to think about where you are right now.

[00:16:28]
And how you know where you are and where you going to be going next at the end of the seminar and how you're going to get there now there are a lot of ways that you can know where you are you just look around the room you're taking if you visual cues and then you make a comparison of those visual cues to some map of some memories of having been in this room before and you say yeah I'm in the seminar room at Georgia Tech you can even close your eyes and you could still know where you are so this is an interesting cognitive function to the brain is able to tell you where you are even in the absence of any ongoing sensory information and you're able to imagine you're able to do mental time travel and think about where you'll be going next and even imagine the past that you'll follow to get there that these are functions that the hippocampus is taught to serve and in the next 2 sections of the talk all provide some data that I think will not solve this problem of how this works obviously but touch on these issues that they just wanted to get you thinking about this issue of how you know where you are how you know where you're going.

[00:17:33]
So obviously memory is a very important part of that we have to remember locations where we've been in the past and compare them to the inputs that we're getting now the hippocampus is known to be involved in this and since we're going to be talking about the hippocampus extensively I want to talk about the circuitry a little bit so you've got.

[00:17:52]
This thing going to work with these lasers when you're over here to dentate gyrus sends its projections to see a 3 C. a 3 senses projections A C A one and C. one senses projections to this a big film there's also enter on a cortex which serves as both an input to the hippocampus and an output of the hip to campus so this is a well known structure of the hippocampus and I want to sort of schema ties it for you in the following way so we can think of this as a big loop the answer Ronald cortex provides input to the dentate gyrus then that goes to see a 3 to see a want is a big deal and back to the end frontal cortex however and around the cortex also provides direct projections to see a 3 to see a one anticipate Killam and see a 3 has recurrent connections extensively within it so there's a recurrent theme here which is there are many many loops in the system there's this one big loop that I talked about we can think about a slightly smaller loop here or a smaller loop here or even it's smaller loop here now to add further complexity to this.

[00:18:59]
There are different cell types and different pathways in this vehicle and I'm going to talk about Mark some routes these work to try to uncover the cellular basis of these different pathways and ultimately their function and he did this with help from a lot of people this work is now published as of last year so I'm going to go through this fairly quickly so one thing that we know is that there are many many outputs of the hippocampus or many different structures that receive input from the hippocampus and we also know that these are divided into different pathways so this is work from Jim can your arms labs as well as many others including our lab showing that there are 2 populations of cells in CA one and 2 populations of cells in the Submit Killam that form these 2 distinct loops here and I'll refer to them will be focusing on to become so I'll talk about distilled the big filament proximal Sobek Ilam and you'll see that process will see a one actually project to distill Sobek human distal CA one project to proximal civic Killam and they're interconnected with medial and lateral and Ronnell cortex in the ways indicated by the colors here and they have different downstream targets as and indicated by these colors which will take advantage of.

[00:20:13]
OK I'm going to show you a bunch of pictures of this big film and they're going to be incremental section here we're going to be zoomed in on the subject a limb here and it's a big film is this region right here this is C A one this is the big deal I'm proximal Is this a big deal I'm closest to see one distal is for the stairway and so what Mark began by doing is showing that in the mouse as the circuitry to be determined in other species like the rat There's indeed this projection specific architecture and he did this by injecting retrograde beads into different structures So here he's injected into sort of these are retrograde virus injections in retrospect you know cortex central hypothalamus or me to lunch Ronald cortex and you see the labeling it's a big deal in here is restricted to the distilled portions of the big Killam and you can see this best because there's this other injection of a different colored tracer in the nucleus accumbens which labels the proximal part of 6 of them so if you superimpose these 2 you can see clear distinction between the distal cells which project to these 3 regions and the proximal cells which are projected to nucleus accumbens as well as other targets that we have been able to hear so we showed in this work that there is a clear distinction to separate sets of cells projecting to different areas outside of the hippocampus and spatially localized to different parts of this victim Here's another example of retrograde labels to prefrontal cortex and lateral in front of cortex and you can see that they call localized with this nucleus accumbens projections and as as always there is one exception there's a small population of cells that project to the foulness and they sit right at the border between proximal and distil civic.

[00:22:08]
OK So we have these We're going to focus on these 2 populations of cells which are projected to these different downstream targets and what we wanted to do we've shown that there is they have different outputs in other work we show that they have different inputs and they have different firing phenotypes as well but what we really wanted to do was get genetic access to these 2 populations of cells so that we could turn them on or off and determine the effects on the behavior of the animal memory kind of behavior.

[00:22:36]
And we were able to do this by injecting viruses into these targets now is sort of now this is not viruses this is beads and then we use these different colored beads labeling to sort out the cells so we disassociate the cells and we sort them out and we do our Any seek and the idea here is to determine gene expression in these 2 populations of cells which we've done and I told you about the website where you can find all the data and we find that indeed in these 2 populations of cells or many genes that are differentially expressed so if we plot the expression of genes from one group of cells against of expression from another group of cells all these colored dots represent genes that are differentially expressed there's hundreds of them in each case and if we look at what kinds of genes are differentially expressed we find that many of them have functions that we associate with the functions of neurons things like act on guidance and he's in molecules peptides and receptor synaptic function voltage gated channels etc and you can see by these colors in these 2 different populations of cells so these genes are expressed very very differently in these 2 populations of cells so there are different cell types are expressing different genes are sitting in different places they're sending their output to different places they're getting input from different places and so on but we did all of this so that and we validated that these differential expressed genes using R.N.A. scope I'm going to skip over this because of time and we've also done this with single cell R.N.A. seek to talk to you a little bit about this before and we find indeed that these distal and Prof know each point on the graph here instead of being a gene this is a cell and is clustered into groups that represent the distal population and the proximal population actually appears to be at least 2 different cell types but we're going to lump them together and treat them as one OK So we have now different gene expression in these 2 populations and what I told you we wanted to do is take advantage of this differential gene Express.

[00:24:36]
When to turn the cells off and. Determine their fact some behavior so here's how we do this we found Korea lines for these differently expressed genes we determined that these. Create the Express differentially in these 2 populations of cells so now we can use this Korea line and we could do an injection of a virus in this case into proximal Sobek Ilam and it's carrying the label M Cherry So the virus is only expressed in Korea expressing cells in the screen line and therefore the population of cells expressing this trancing is restricted to proximal so because I mean you can see this because we just another retrograde chase tracer expressing G.F.P. in the distal subpixel and see these are non-overlapping and they're not.

[00:25:25]
It's not that this is restricted to Proxima subpixel and because that's where we've put the virus for sure the virus can spread over to more disobey Killam it's restricted because of the requirement for the CRE for the expression of the reporter Gene which has its M. Cherry label we can also do the opposite experiment using a different Korea line which is selected for distill so big Killam we inject the virus for M. Cherry Now in the distal subpixel I'm can see that here the G.F.P. label is a retrograde tracer that goes to Proxima so because I'm again non-overlapping projections.

[00:26:00]
OK so now we have fessed behavior so we're going to the what if what you may have noticed if you're looking closely is that we didn't just express and cherry in the cells we also expressed a gym for D H M 4 D is a dread receptor which can be activated by a logon to silence the neurons and we show that this works really really well and so because I'm so we're going to assess behavior while silencing one or the other population of cells and we did several different tests of different behaviors and found no effect but then interesting we found an effect on a test called the Morris water mains probably many of you are familiar with Morris water mains but for those of you who aren't Here's a video complaint the mouse is put into an opaque pool of water they're good swimmers but they don't really like being in water so they're looking for a way out and they can't climb out of this pool so they just swim around desperately hoping for some escape and indeed there is a platform hidden in the water somewhere below the surface of the water and the mouse eventually will find this platform and once he finds it he stands there on the platform and just stays there but initially this is just a search a random search until he finds a platform OK but now if you repeatedly put him back into the same pool of water to learn the location of the platform this is now a different animal but this guy's already been trained and he knows exactly where the platform is and he swims straight to it OK So here's the experimental design every day the animal is exposed to the platform multiple times and every day it's a platform location is changed so it takes a few taught a few trials on one day before the animal learns the location of the platform and then were mean and the next day we switch the location of the platform and they have to learn a new location all over again.

[00:28:05]
OK then we do this 9 days in a row and then one day in day 10 we inject either scieno or vehicle we do the same type of training procedure but then on one extra trial we do what's called the pro trial where the platform is not there and you determine whether the mouse has learned or remember the location of the platform by whether it looks for the platform in the right place and then on day 11 we do this again with either the vehicle or the scieno depending on what we did on day 10 OK so here's here the results for.

[00:28:40]
The control you can see when there's a probe trial the animal swims most of the time in the location where it's expecting to find a platform OK So they've learned that on this day the location of the platform is right here they swim around there most of the time when we silence the proximal Sobek Killam it has no effect but when we silence the distal subpixel I'm the animal shows no sign of remembering where the platform is located so it doesn't learn the location or remember the location of the platform when we've silence that population of cells in the distal subpixel limbs and here's a group data to show you this is true of the many many mice OK So you actually can learn a little bit more than this from these experiments because of an interesting observation that when you move the platform on the next day the animal seems to remember the location from the previous day and the reason we know that is that even though there's no platform here where it was the previous day on the 1st trial of the next day the animal spend most of its time swimming in the area of the platform from the previous day so the animal clearly remembers where it was yesterday and eventually realize it is not there anymore and learns the new location of the platform.

[00:30:01]
So there is an interesting thing here which is you can test whether the animal when we saw it just feels a bit Killam has a retrieval deficit or an encoding deficit so if it's a retrieval impairment what that means is on the scieno day the animal shouldn't remember what happened on the previous day but if it's an encoding impairment then on the following day when the scieno has worn off and the cells are no longer silenced the animal should showed no sign of having learned it on the previous days and what we find is that it's actually very consistent that the retrieval is intact and it's consistent with an encoding impairment OK So even though we've silence distal subpixel in the animal can't learn a new location of the platform it does show silent signs of remembering where the platform was on the previous day now here's of Group Data this result was very surprising to me for the following reason this is a big we think of as the major output of the hippocampus and we think of the hippocampus as the structure that forms the memory.

[00:31:09]
And so my prediction was that well if we silence the output of the hippocampus the memories can still be formed all the circuitry in CA one in CA 3 in the dentate gyrus that's all still intact a memory can be formed but if you can't send that information out to these other areas of the brain the animal will be able to use that information but in fact it's the opposite that the information going out of the submit Killam needs to be there in order for the animal to form a memory this is a very surprising result Now one possibility is that we've disrupted this loop or one of these loops that I talked about before by cutting the output right here so maybe that cutting the output here was less important than cutting the output here and these are the kinds of things we can test in follow up experiments but the major hypothesis I have at this point is that some kind of long range loop has to be intact in order for the memory to be formed that this long range loop for example if you cut it anywhere it's not going to be able to form a memory and again this is somewhat surprising result that changes for me the way I think about how memories are formed they're not just formed in the hippocampus they're informed there are there are formed in the hippocampus and other circuitry that's associated with output from the hippocampus to other parts of the brain question too so so the the other job question was whether the creel on for a specific to dorsal versus ventral Sobek Killam and the answer is that there's some expression in ventral subpixel him as well there's also expression in other parts of the brain but what we've done is by doing virus injections we restrict the region where the drug receptor is are expressed so most of the drug receptor is going to be expressed in dorsal so big healing because that's where we inject the viruses so it's an intersectional approach that allows us to restrict the manipulation OK.

[00:33:19]
Where are we now. Thank you exciting I mean sure why does it matter OK I talked about how we can cut potentially in future experiments cut in different places OK let's talk about what we're trying to do overall formation of memories let's talk about using memories OK So back to this question how how do you know where you are you have to use memories to know where you are I'm going to talk about this work that was done by a post such a new show and this is part of a collaboration with Jeff McGee's lab and use for a long post aka Jeff and he's loved and she contributed also OK So again HIPAA campus is involved in spatial memory and we want to try to understand how the hippocampus knows where we are so we're going to record from neurons in the hippocampus specifically in the CA one region of the hippocampus while an animal navigate the virtual environment while running on a ball the environment looks like this it's an oval track there are 6 discrete locations the animal runs clockwise in this virtual track and there are different cues on the wall at each position to over a period of these as locations $1.00 to $6.00 and cues 8 throughout and will move the cues around a little bit here's a weird sort of hybridized video that gives you a sense of the animal running this video is not working but the animal runs on the ball that controls movement in this virtual space through this artificial environment.

[00:34:50]
So while the animal's doing this we can put an electrode in to a wholesale recording in CA one and CA one sells as a mentioned get input from ca 3 as well as from the in Toronto cortex and you can see a beautiful labeling here as work by another post op care applause this is labeling inputs from the enter on a cortex which are restricted to the distal tough to androids and the rest of the dendrites get their input from ca 3 so we're recording from the CA one sells and because we're doing wholesale recording we can evaluate not only whether they're spiking or not but also what kind of synaptic input they've received so here's how the experiment works the animals in this virtual environment we're recording and in many cases as the animal runs a lap there's nothing there's no firing we don't see any significant puts And we actually take advantage of this and we take advantage of a beautiful we can show that this is fact is a real neuron we can record reconstruct it has to logically later on but it's silent it's a silent cell now we take advantage of a beautiful trick that was discovered by K.T. but neuro post-op can just meet these lab and this is called behavioral time scale synaptic plasticity I think one of the most profoundly important discoveries of the last decade or so in the field of synaptic plasticity and what Katie discovered is that sometimes neurons will be silent and then suddenly they'll form a place to yield and when they do that there's a plateau potential associated in the playfield and she could infect create an artificial plateau potential and artificially create a new playfield so that's what we've done here.

[00:36:30]
On the next lap on lap number 3 we just inject a bunch of current into the cell while the animals out location A We do that a few times and aunt's all subsequent labs to sell fires action potentials in that same location where we artificially injected currents OK and then a follow up paper what Katie showed is that there's some beautiful synaptic plasticity that occurs as a result of this big plateau potential and that's responsible for the formation of this place field in the sea it oneself OK looking at looking at this little close more closely when the cell fires in the playfield there's a few things are spiking of course there's a ramp of the polarisation which has been filtered out here and there are fade off fallacious that increase in amplitude when the animals in the place field.

[00:37:19]
OK so we can induce a playfield and one of the nice things about that is we have control of these cues and we can ask what is it about that location that is causing firing so to do this what we do is we duplicate the cue we move a cue or we duplicate a cue at location one which is an A and we put it over here at location 4 and when we so 1st we induce a playfield in this vocation which is location one and then we move it we duplicate it at location 4 and you can see it produces a little bit of firing this is a little bit wimpy but there's some firing and across the population the effect isn't that small Actually it's reasonably significant You can see significant amount of the polarization at this location where we've duplicated the cue the red is following the duplication you can see there's the polarization that wasn't there in the control there's also an increase in state amplitude and some spiking OK so there's partial induction of a playfield just by moving the cue so it seems like the cue the visual cue is a really important piece of information that the animal uses to produce its map of where it is in the hippocampus now we thought maybe one of the reasons it's only partial is because when the animal's actually at location one it can see forward to location 2 where there's a different Q.B. then when we duplicate cue a location for it sees forward to location 5 where there's a cue so it's seeing different things let's now change the situation it's virtual environment so we can do this however we want and we put a virtual curtain here which blocks the animal from seeing B. when it's a location one it only sees a and a blocks it from seeing when it's a location for it can only see whatever it is that location for and when we duplicate the cue in this way we get a much more striking effect there's pretty much full blown induction of playfield firing at the location of the duplications you can see this in the group data over here so this is evidence that the only thing that's really.

[00:39:19]
Riving place cell firing in this case is a visual input that the animal receives and if we control the visual input so that only Q.A. then we can put Q.A. anywhere in this environment and it will drive firing so that means that the Q. and A is this visual cue is sufficient to drive firing of these cells.

[00:39:39]
Now if we instead of moving a story instead of duplicating Q.A. we move it from location one to location for that move to the location of the place field firing it's gone at the original location can see that in the summary data over here so what that means is that not only is that cue the visual cue sufficient to drive the firing of a cell it's also necessary so without it you get nothing OK So we have perfect control over what's causing this play cell to fire and that's pretty surprising in a way because what it means is that the history of what the animal saw before it encounters a Doesn't really matter the fact that it saw D. than either an absent a is no different than seeing B. thence C. then OK it has no recent history effect on the playfield firing that's a little bit surprised but we knew that the representation of environments of cues in environments is very context dependent the hippocampus So we asked what happens if we take this cue A and presented the animal in a totally different context a completely different environment so the animal learns to run in 2 different environments the oval track and now a triangular track and we have one common cue in the Triangle track that's Q A which we've been talking about is also present in the triangular tract and we can move the animal in virtual reality from one location to the other pretty much instantaneously so what happens call this context wanted context to what happens if we induce a place feels at position one with Q.A. in the 1st context and then move the animal into a new context we introduce the animal into the new context at position why and then from the very 1st time it sees a in this new context it does nothing OK So previously you might have looked at the firing of that cell and say wow that looks like a cell in the visual cortex is just responding to a visual stimulus but this is very different from a cell in the visual cortex because it's completely gated by context and not only is it not firing there's no deep polarization whatsoever.

[00:41:50]
OK So somehow context is an almost perfect gate of the representation of this object in the animal's environment here's another example where when we put it into the 2nd context there's no place you'll firing associate with Q A but we have playfield firing associate with a different key facts and the summary data are shown here here's the control wrap the polarization in the original place field in the original context absolutely nothing when you present it in the new context how are we doing on time here we're going to 1215 or 1210 write something like that OK Good all right so how does this work this simplest way that you can imagine that a cell couldn't code a sensory stimulus and the context is that it received some synoptic input associated with each of them you sum it this goes above threshold for firing and you get action potentials firing this is clearly not how it works and the reason we know this is not how it works is because when we changed the context there was no left over representation of the sensory input the gating is occurring in some other way and the most likely place to think about where that a Kading occurs is upstream in the CA 3 region which provides the majority of excited stories enough to drive to the cells So here's what it ca 3 neuron looks like by the way from this most light project this is so cool I had to show it this is a reconstruction of the X. on Arbor is a sion of a single C. a 3 cell.

[00:43:29]
And this neuron has about 10 centimeters of X. on the so she did it with a single cell and it was reconstructed by the mouth sight team in 8 hours previous reconstructions A C 3 cells were taking months prior to work with this team so that's what the C H 3 cells look like they're providing massive amounts of input to see one so it's a logical place to now go and look for this context dependent gait OK So we do the same kind of experiment now there's a trick there's a difference here important difference which is we did these experiments using extracellular recording the whole cell recording is harder in CA 3 so we started with extracellular recording using silicon probes.

[00:44:10]
That means you can't induce a place field so you have to rely on cells that already have a place field and you don't know quite as well what's driving them but you can infer it so here's an example here's a cell that's responding right around the end of Q A BEGINNING OF Q B type of area OK so now we duplicate Q.A. at a new location so we shift it we don't duplicate it and that moves the playfield far into the new location so just like in CA one it seems like the cell is responding very reliably to the visual input here is another cell where it's similarly pretty good here is the visual input that's driving the cell is probably around and probably a B. here as well and when we move that a to the new location leaving a blank here a position one we clearly induce playfield firing at the new location there's some leftover firing at the original location probably because Q.B. was also driving the spiking of the cell So again very consistent with the notion that visual input is a very important driver of these cells Here's another example where we shift the natural place field to occur as a location for probably largely driven by 2 D. So now we move to different nations for the position one and the firing moves close to that same location there is a little more variability and a little more noise in CA 3 tendency and one here is an example when we moved the location of the playfield to some leftover firing at the original location so it's not quite as clean as what we saw in CA one but nevertheless this is a pretty robust effect that when we move the cues the playfield firing gets shifted from the old location to the new location you can see there's some leftover firing at the original location but not much.

[00:46:06]
OK So what about context context also seems to Gates is a firing so we have we have driving of the firing by the visual input in CA 3 what about this can textual gaining we do the same contextual gaining experiment this is a little bit messy this one the cell has a lot of firing at its original playfield we present the same to you in a new context and there's a little bit left but not much the other cells are cleaner this one has a complete gating of its place feel firing response a context this one also nearly complete getting in this one complete catering so in CA 3 the situation looks very similar to what we saw in CA one now what I told you we've got sensory driven plays fields in CA one there is strongly gated by context and the same thing is true in CA 3 I told you that this model of context will gating of the cells response could not be true in CA one but what about in C A 3 story and instead what we think is happening is in one context you get a population of ca 3 cells are acting as input to CA one producing these the polarizations and CA one which some to drive the spiking of the cell case of the gating the contractual gaining is either occurring in CA 3 or even upstream of ca 3.

[00:47:31]
By the way. I just want to briefly mention this probably don't really have time to go into details this is published last year of beautiful project down by another post chain long issue showing that these CA one place all responses are very voltage dependent so there's an extremely non-linear response to the synoptic input that means that a very small subtle change in synaptic input which is going from the red to the black here produces a big change in the response of the cell in his place field in vivo and he did a bunch of in vitro experiments also determine the cellular mechanisms in the ion channels that are responsible for this non-linear amplification of place all responses and I think this is going to be very relevant to the mechanisms of this can textual gaining OK so I told you that this model could not be true in CA one could be true in CA 3 but based on the spiking data it seems very unlikely that this is true because there's already really strong conceptual gating of spiking in CA 3 so instead I want to present a model this is a cartoon back of the napkin type of model of what I think might be going on and it provides sort of the the basis for what we hope will become a computational model that can produce some testable predictions down the road so the idea is that in CA 3 you have a strong sensory response.

[00:48:57]
And a very strong sensory input and a small contextual input these 2 things serve as input to a CA 3 cell they some non-linearly to produce spiking the same is true in some other cells in CA 3 and because these are all connected to each other by these recurrent collaterals this population of cells reinforces its own firing and you get a population of cells associated with a particular sensory cue in a particular context context or environment now this population of cells is firing in CA 3 serves as input to CA one maybe there's a little bit of contextual input in CA one but this has to be now enough to gate this this firing of in CA one so really what matters in CA One is what population of cells was activated upstream in CA 3 Now to illustrate this for you imagine that we put and by the way I'm describing this is a mechanism that involves cascading on the knee or these in a certain there are nonlinearities that several stages of circuit processing that can result in this very strong can textural gating that we see in C A one and to some degree also in CA 3 OK So now imagine what happens when we put the same sensory stimulus in a different context this small change in the contest will input to a cell that's a maybe it's a c 3 cell can actually cause it spiking to be reduced or even to fall below the threshold for spiking and therefore a different population of cells will be firing in response to this sensory cue in this context this population that responded in context one will not respond in context to because that little bit of input from the context will signal was enough to cause a large number of cells to fall below the threshold for spiking therefore the input to see a 3 cells will be a little bit from this guy because he fired one spike nothing from this guy.

[00:50:57]
And nothing from context and you have almost nothing left in CA I want so this concept of cascading nonlinearities I think is a nice model for how you might get something like an textual gating of sensory inputs Now the reason I asked the question how do we know where we are we have really in these data neural signature that the animal probably knows where it is as soon as you change the context that there is a very strong neural representation that a response to something we know drives the firing in one context doesn't do anything in another context so it's a neural signature of that context of that knowledge if you will of where the animal is in its environment OK I'm going to stop there again so I want to thank many many people I Genelia cling to people in my lab all these other people that I mention and I just would like to tell all of you that are considering for the future of where to go the Giannoulias a great place to think about coming doing a post office or even tasty work if you're at that stage so thanks I'll stop there and maybe there's a few minutes for questions.

[00:52:12]
Lose. Thank you Peter thank you thank you that got The Cafferty File Jack. Really. There are a lot of dives by. Specially for her. Takeaway Q. It's all. It's just. It's work here so. So. So so so the question really is what information is the animal using to determine which of the 2 contexts it's in right.

[00:53:13]
And there are 2 different there are 2 important differences the 1st one is that in the 2nd context there's a novel cue we can label that why on the slides which the animals never sees in the 1st context so in principle that alone is enough information for the animal it's experience both of those environments before and during training so in principle that we're just seeing why is enough to know yeah that's a different time in a different place now there's also different turn angle as you point out which the animal experiences before it encounters the relevant cue so.

[00:53:49]
We don't know how important the turn angle is so in future experiments we want to take that out of the equation either by creating 2 environments both in oval tracks or maybe even trying to create just a linear track environment so you can is have an animal running and repeatedly occurring sequence of Jews that mimic the sense of running around in a circle instead it's an infinitely linear track and then you can take a turn angle out of the equation altogether so we're going to do some experiments like that.

[00:54:21]
If my prediction is that probably that initial Q Why is all the animal needs to know that this is this other environment but it's just not 100 percent clear that that will be true. If you want to. Basically. Yeah I mean we haven't done that experiment but I make a very strong prediction there that if we move it on the other side of the relevant queue to the other side of the virtual curtain they will produce partial firing because they can see past the relevant cue and they can see other things so now their visual input is different here and the Depart of the difference in their visual input is the turning goal coming up.

[00:55:08]
But all this stuff is. Easily addressable future experiments because we have full control over the virtual environment and about. The. The. Next. So. The question is whether history matters more in C. 3 than C. one I just want to point out that history. You know has different time scales right and what we showed is that the recent history doesn't matter that is if Q A is the relevant page will cue driving the firing it doesn't matter if it was preceded immediately preceded by B. or C. or whatever he was going to work on it on the last.

[00:56:10]
The longer history actually does matter and that's the contextual gating because the animal's knowledge of where it is which is also formed by history and by memory is relevant but your question I think is about the very recent history of the Q It's all right before Q. that drives the firing and whether that matters more in C. 3 than in CA one and we don't have any evidence that it does matter more but you're correct in pointing out that it's harder to interpret the extra set of the recordings than the interest of either interest so is beautiful because you can see whether there's any input associated with a cue in this particular context in the exercise you can only tell if the cell is spiking So Jeff McGee's lab is doing wholesale recording in C 3 this is a collaboration between our lab and his SO kill hopefully have an answer to that question within a few months.

[00:57:09]
But for now I would say we don't know but I doubt it that's my guess the reason I said out it is because I think whatever happens whatever the animal uses in CA 3 that information is available to CA one and use could see some signature of that and I think.

[00:57:31]
You know. This. It's. So the model the model is just the drawings that you show your showed that I showed you there that's all that is right now so we haven't assessed how dependent this model would be on the relative strength the sensory vs contextual inputs. I suspect that strong contextual inputs would also work quite well what I was trying to convey by showing the small contextual input is that.

[00:58:23]
The point is that nonlinear amplification can accentuate small differences and it's been shown in many past experiments that animals don't need much information sensory information to distinguish between 2 environments and that's probably a consequence of pattern separation the show them into things that are subtly different and pushes neural circuits into different activity States more definitively patterns I should say based on some small difference in the contextual input so that's what I was trying to convey but I would say it only gets easier as the contextual differences are greater in the contextual signals therefore increase.

[00:59:10]
I would think so yes here Yes QUESTION. 6 months. Or. So is our. So it's because the question is about animals anticipation of the Q and I think what we've shown is that animals although they may be anticipating a cue there's no place feel rip firing a place filled response that is indicative of that anticipation So even though the animals learn that every time I come around past Q.F. I'm going to see Q.A. at this location as soon as we remove that Q.A. the not only is a firing on this and Epic people are zation is completely gone so that would suggest that anticipation there's a signal of anticipation that's visible in the hippocampus at least in the way that we've assessed it yeah.

[01:00:25]
Yeah so there's 2 issues here the question was about non-linear amplification not being static and that as memories are formed you might lose the high gain. There's 2 issues here because we are looking at the hippocampus after memories have already been formed we have a series of. Training trials where the animals exposed to these environments become familiar with and comfortable with the idea of running on this ball to change the virtual world and so we're looking at it after memories have already formed.

[01:01:00]
Wild memories are being formed that is the very 1st time you put an animal into a new environment Presumably there's a whole different set of things that happen to allow memories to be formed one of them is probably this behavioral time scale. Synaptic plasticity that I mentioned before that's also dependent on nonlinearities in the neurons so there are different kinds of nonlinearities that could be invoked there are also neural modulatory transmitters and receptor as I don't mean as the colon is for tone and they can change the state of the system that could in principle turn these non-linearities up or down turn the gain more down because.

[01:01:47]
That's right so they could they could take and regulate how readily the system is prepared to be modulated that is to be subject to the non-linearities they will form new memories versus the modulatory circumstances that may or more readily produce recall of the non the areas of so she would recall of previously forms memories Fritz's So there is it out a whole host of nonlinearities in the system that can be all regulated up and down probably.

[01:02:21]
By.