7th HLF – Lindau Lecture: Space and time: Internal dynamics of the brain’s entorhinal cortex

Video in TIB AV-Portal: 7th HLF – Lindau Lecture: Space and time: Internal dynamics of the brain’s entorhinal cortex

Formal Metadata

7th HLF – Lindau Lecture: Space and time: Internal dynamics of the brain’s entorhinal cortex
Title of Series
No Open Access License:
German copyright law applies. This film may be used for your own use but it may not be distributed via the internet or passed on to external parties.
Release Date

Content Metadata

Subject Area
In mammals, space is mapped by specialized position-coding cell types in entorhinal cortex and hippocampus, including entorhinal grid cells, which are active only when animals are at locations that tile environments in a periodic hexagonal pattern. I will first show how space-coding neurons in the medial entorhinal cortex (MEC) collectively form a low-dimensional representation that persists across behavioral tasks and activity states. This low-dimensionality points to network architecture as a determinant of firing patterns and continuous attractor models have been proposed to account for the dynamics. Current models do not easily account for the representation of the unidirectional flow of experience, however. In the second part of my talk, I will thus ask how entorhinal networks are organized in time. Which trajectories in high-dimensional state space do cell ensembles take during experience? To determine how activity is self-organized in the MEC network, we tested mice in a spontaneous locomotion task under sensory-deprived conditions, when activity is determined primarily by the intrinsic structure of the network. Mice were head-fixed and ran on a spherical cylinder in darkness. Using 2-photon calcium imaging, we monitored the activity of several hundreds of MEC layer-2 neurons. Both linear and non-linear dimensionality reduction techniques were applied to the spike matrix of each individual session. When the cells were sorted according to their contribution to one of the first principal components of a principal components analysis, stereotyped motifs appeared, involving the sequential activation of neurons over epochs of tens of seconds to minutes. Transitions between cells that were close in principal-component space were favored, while transitions between clusters farther apart happened with a lower frequency than chance. Such stereotyped sequence elements may be recruited during encoding of space, and more widely experience, in the entorhinal-hippocampal network. Deficiencies in these mechanisms may be at the core of neurological diseases characterized by early entorhinal cell death, spatial disorientation and memory dysfunction, such as Alzheimer’s disease. The opinions expressed in this video do not necessarily reflect the views of the Heidelberg Laureate Forum Foundation or any other person or associated institution involved in the making and distribution of the video.
Area Positional notation Internet forum Spacetime Musical ensemble Lattice (order) Quantum chromodynamics
Functional (mathematics) Information Code Texture mapping Electronic program guide Temporal logic Function (mathematics) Code Formal language Connected space Function (mathematics) Internet forum System programming Self-organization Energy level output Spacetime Lie group Physical system Quantum chromodynamics Spacetime Physical system Exception handling
Area Gateway (telecommunications) Complex analysis Focus (optics) Functional (mathematics) Observational study Host Identity Protocol Observational study Surface Multiplication sign Surgery Mereology Connected space Neuroinformatik Similarity (geometry) Integrated development environment Semiconductor memory System programming Video game Physical system Position operator Sinc function Physical system
Point (geometry) Metre State observer Cognition Code Multiplication sign Combinational logic Mereology Field (computer science) Neuroinformatik Mathematics Different (Kate Ryan album) Cuboid Category of being Position operator Area Dot product Touchscreen Mapping Information Basis <Mathematik> Position operator Uniform resource locator Integrated development environment Spacetime
Area MIDI Link (knot theory) Theory of relativity Code Direction (geometry) Multiplication sign Source code Combinational logic Graph coloring Frame problem Disk read-and-write head Degree (graph theory) Different (Kate Ryan album) Order (biology) Cuboid Directed set Ranking output Curvature Position operator
Area Multiplication Multiplication sign Covering space Field (computer science) Uniform resource locator Integrated development environment Phase transition Computer cluster Cuboid Self-organization Spacetime Pattern language Position operator Spacetime
Metre Group action Greatest element Divisor Distribution (mathematics) View (database) Direction (geometry) Multiplication sign Branch (computer science) Distance Disk read-and-write head Field (computer science) Number Frequency Mathematics Computer cluster Geometric series Cuboid Form (programming) Physical system Self-organization Module (mathematics) Area Scale (map) Scaling (geometry) Gradient Maxima and minima Flow separation Type theory Uniform resource locator Network topology Time evolution Order (biology) Self-organization Natural language Modul <Datentyp> Spacetime
Point (geometry) Trail State observer Functional (mathematics) Digital electronics Standard Model Multiplication sign Direction (geometry) Similarity (geometry) Disk read-and-write head Attractor Neuroinformatik Number Element (mathematics) Mechanism design Different (Kate Ryan album) Velocity Computer network Energy level Circle Analytic continuation Address space Spontaneous symmetry breaking Tunis Predictability Artificial neural network Standard Model Basis <Mathematik> Bit Multilateration Continuous function Disk read-and-write head Connected space Degree (graph theory) Type theory Attractor Word Uniform resource locator Velocity Ring (mathematics) Order (biology) output Pattern language Right angle Quicksort Local ring Spacetime
Direction (geometry) Combinational logic Set (mathematics) Field (computer science) Dimensional analysis Theory Revision control Hexagon Different (Kate Ryan album) Phase transition Computer network Energy level Integrated development environment Hardware-in-the-loop simulation Self-organization Predictability Mapping Standard Model Computer network Distance Connected space Similarity (geometry) Attractor Integrated development environment Ring (mathematics) output Pattern language Quicksort Data structure Row (database)
Predictability Point (geometry) Scale (map) Observational study Standard Model Standard Model Computer network Distance Connected space Proof theory Hexagon Cross-correlation Phase transition Phase transition Connectivity (graph theory) Orientation (vector space) Pattern language output Self-organization Pattern language Right angle Data structure Condition number Self-organization
Predictability Slide rule Mapping Multiplication sign Computer network Quantum state Connected space Number 2 (number) Quantum state Pointer (computer programming) Cross-correlation output Cuboid Pattern language Diagram Right angle Codec output Computer architecture Data structure
Point (geometry) Trajectory Group action Standard Model Multiplication sign Mereology Theory Sequence Cross-correlation Computer network Ranking Spacetime Diagram Software testing output Position operator Trajectory Computer-generated imagery Computer network Integrated development environment Order (biology) output Pattern language Local ring Row (database) Spacetime
Computer virus Quantum chromodynamics Computer-generated imagery Surface Multiplication sign Moment (mathematics) Expression Bit Approximation Medical imaging Integrated development environment Computer network Self-organization Energy level Condition number Spontaneous symmetry breaking Quantum chromodynamics
Area State observer Random number Connectivity (graph theory) Multiplication sign Principal ideal Laplace-Operator Instance (computer science) Variable (mathematics) Trans-European Networks Sequence Plot (narrative) Dimensional analysis Frame problem 2 (number) Number Component-based software engineering Wave Principal component analysis Principal component analysis Pattern language Matrix (mathematics) Reduction of order Row (database)
Personal identification number Trajectory Diagonal Multiplication sign Structural load Binary file Binary file Sequence Frame problem Sequence Similarity (geometry) Single-precision floating-point format Raster graphics Principal component analysis Computer cluster Matrix (mathematics)
Frame problem Mapping Direction (geometry) Multiplication sign Graph (mathematics) Line (geometry) Binary file Binary file Sequence Trans-European Networks Frame problem 2 (number) Similarity (geometry) Sequence Frequency Personal digital assistant Different (Kate Ryan album) Graph (mathematics) Computer cluster Energy level Quicksort Condition number
Lattice (group) Multiplication sign Point (geometry) Computer network Dimensional analysis Control flow Sequence Digital electronics Data structure Capability Maturity Model 2 (number) Sequence
Trajectory Constraint (mathematics) Direction (geometry) Range (statistics) Standard Model Computer network Disk read-and-write head Connected space Sequence Attractor Mechanism design Medical imaging Integrated development environment Computer network output Representation (politics)
Metre Pulse (signal processing) Trajectory Image resolution Multiplication sign Direction (geometry) Open set Surface of revolution Graph coloring Medical imaging Medical imaging Integrated development environment Universe (mathematics) Energy level Cuboid Position operator Distortion (mathematics)
Noise (electronics) Trajectory Personal digital assistant Multiplication sign Computer network Open set
Point (geometry) Digital electronics Standard Model Multiplication sign Direction (geometry) Attractor Theory Number Inference Machine learning Computer network Energy level Extension (kinesiology) Physical system Area Noise (electronics) Multiplication Mapping Software developer Interactive television Physicalism Computer simulation Limit (category theory) Internet forum Phase transition Order (biology) Musical ensemble Row (database) Spacetime
[Music] after the during lecture we will have
another special lecture the Lindo lecture it is electric given by a local area to will normally attend the Linda of Nobel laureate meetings this years Linda lecture will be given by Edward Moser he's a psychologist and neuroscientist he has a professorship for psychology at the Norwegian University of Science and Technology in Trondheim and he shared the 2014 Nobel Prize in Physiology or medicine with my blood Moser and John Okeefe for their work identifying the place cells that make up the brains positioning system my positioning system tells me to leave the staging of the floor to edvard moser good morning
thanks a lot for the invitation I will as you understand talked about the brain and I will focus particularly on space but also somewhat on temporal organization or brain activity my reason
for focusing on the brains space system or the brain system for helping us find out where we are and how we get from one place to another is not only that I think this is an interesting problem by itself but more the fact that among the high level brain functions this is perhaps the one at least one or very few where we can describe activity in some mechanistic detail and where there is some theoretical background that guides our investigations so as you heard in dr. Banjos also the brain is very hard to access and particularly the cortex because neural codes first of all occur across 100 billion neurons and these neurons are very small particularly the connections between the neurons synapses were information lies and the code is distributed intermixed and combinatorial and definitely we have a curse of dimensionality so it has been very very hard to pick up activity from the brain that can be related to any function in the cortex but there is some exception for the brain brains navigation system which you will understand very soon that makes it possible to access even if it is kind of in the middle of the brain far away from sensory inputs and motor outputs so
where in brain is this well for such a complex function of course almost the entire cortex is involved but still there are two areas that are particularly important and historically focus has been on the hippocampus so hippocampus is in your medial temporal lobe it's not on the surface as indicated here but actually more on the inner side where the cortex goes curves around on the inside but nonetheless this is cortex and all part of cortex and it has been investigated intensively for the last 60 70 years since patient got surgery in this brain area and it was found to play a major role in daily life memories so this patient then lost his memories but it was also found later on that the same brain area and the same cells are actually important also for for representing or position in the environment so positional memory are very strongly linked and a lot has been learned about both in studies of hip more recently another area has come into the picture and that's the entorhinal cortex which in many way is a gateway between hippocampus and the rest of the cortex so hit enter in our cortex connects hippocampus with the rest of the cortex all areas of cortex almost but it also has functions on its own as we will see very soon but this is hard to study in humans for obvious reasons so most of the work has been done in rats and mice for many reasons but one of them being that especially rats have hundred years of experimental histories we know quite a lot about a rat behavior and rat brains and mice are not all that different so now we had transgenic mice so that has also caused a lot of advance in our understanding of these systems but their brains and their cortices even are organized very very similarly to the human cortex of course they are much smaller but the cells they consist of the connections the functions of the cells are very similar and they are also it turns out very important for helping rats and mice to navigate which they are very good at and actually do in many ways the same way as we do so a major
advance happened in 1971 when John Okeefe used micro electrodes these are very thin electrodes at that time probably 30 40 50 microns wide a much thinner now that he put into the hippocampus of rats so you see a rat brain here and the red is the hippocampus the blue is the internal cortex and not really knowing what he was going to find he started recording from rats with hippocampal and electrodes connected to an oscilloscope and a computer and started listening to their activity and found that this in the hippocampus indicated the position of the rat as it was running around freely in the environment so I want to show you such an experiment what
you see here is a rat in a box the box is won by meat one by one meter lodge you see it from above the rat is I'm going to start a movie very soon but you will see a rat running around chasing small chocolate pieces so the rats like chocolate so that can keep them running for for long long long times and of course we do that to have the rat visit every possible place in the box in every possible direction and at the same time we record activity from electrodes that are inserted into the brain so we can pick up the electrical and discharges of neurons in the hippocampus and through a cable that is connected for the occasion and then convert the electrical changes the voltage changes to a sound so we can listen to the activity of a single cell in the hippocampus so we are going to do that now and at the same time as we hear the sounds of that cell you'll also see red dots appearing on the screen when whenever the cell is active so this is one single cell in the campus typical cell and you see that this cell is while the red dots doesn't quite match the sound but anyway the point is that you see the see the cell being active only when the rat is in the upper left part of of the box and since the rat since this cell is active only in one particular place it was then called a place cell by o'keefe and uhm cannot be
illustrated by a color code here so right this high activity blue is low activity and this is the box seen from above so different cells in the hippocampus have different place fields that means even different parts of the box and there are many hundreds of thousands of these cells in a rat brain so you can imagine that at every possible place in the box there's a different combination of active cells and that combination is unique across the environment so that any one computer or brain area that listens to the combination of active cells can figure out exactly where the rat is at any given time and because this implicitly in has in or contains a code for location in the environment then a few years later O'Keefe together with his colleague Lin Adele suggested that this was the basis of a neural map for space so such a map had been proposed 20-30 years earlier by Edward Tolman based on purely psychological observations or rats but this was the first time that anyone actually found something that might correspond to a neural map of space a map of space that contains information about your coordinates but also about the experiences you make at different positions in the environment so this was in the 1970s and more was learnt about place cells but then another major
discovery was made in the mid or late 1980s when Jeff Todd and Jim rank found that in a different brain region called the priests victims which is somewhat outside the hippocampus but still connected to hippocampus cells in that region fired overactive in relation to the direction that the rat was facing in the box so imagine again rat is running around in the box and each time the rats head faces a certain direction could be in the east or west or south or but only of course in relation to the room frame then the cell is active so one example here this cell is active only when the rats head direction is about 300 degrees in the room and otherwise totally silent other cells could be active at other directions and again in combination this provides a very precise code order that's instantaneous direction so there's a code for position as a code for direction but how are they linked there was a big question at that time and also how are they generated people also started thinking about that but it was
motivation enough for us to start recording in a third brain area called the entorhinal cortex area that I showed on the initial slide in blue and this is in a rat brain seen from behind and the area and color here is the entorhinal cortex and we recorded up here because this is the area that is most strongly connected to hippocampus so the reason for starting to record there was first that this is the connect connects the place else with the rest of the cortex and we found already then that places probably cannot be the source of the place code so it made sense to once go one step out and also the fact that this brain era connects the direction cells in the priests vikrum with the place else in the dorsal hippocampus it might be an interesting place to look if you want to find out how they are integrated and what we observe there are cells that
are somewhat like the play cells in the hippocampus in the sense that they are active at particular locations so here you see again the box from above in gray is the trace of the rat as it is running around in the box for 30 minutes and each black dot is the location where one particular cell in this brain area is active so you can see there are active at distinct positions but multiple positions and those positions form a very very regular pattern that stretches across the entire environment so the pattern is indicated here when I put these red lines on top you can see it's really repeating hexagonal pattern that causes the entire space like a grid and for that reason we call these cells grid cells and there are many many grid cells
in the rain area so this shows when you record several of them at the same time so these are the electrodes each of them approximately 17 micrometers wide so if you are above the cells you record then you can imagine we call the blue and the red and the green one here and this is then the grid pattern that you see of each of them and you will see that that each of them have grid patterns but they are shifted in X Y location or they have different face as we call it and this is quite typical that there is not very much at least very very little organization with face that means that at any position in the in the anatomical environment any position in the brain in the entorhinal cortex you have all faces represented so often we call this salt-and-pepper organization however there also there is some other
organization in the sense that the frequency or the scale of the grid changes from the top of the brain to the bottom so the colored area here is again the entorhinal cortex this is the brain is side view of the brain this is hippocampus this is the internal cortex you start from the top and then you have high frequency or grids and then the frequency goes down and you get wider wider spacings as you get lower so but this is not the continuous gradient actually consists of quite discrete modules at least five probably it's in the order of 10 where you have first one group or very small scale grid cells and then another one with higher scale and then a third one with a larger scale and so on where the distance if this is a box where the rat runs the distances here may be 34 centimeters between the clusters and down here it's several meters and these these modules as we call them they are organized they're actually overlapping quite a lot in in in the brain but nonetheless very very clearly separated and funny thing is that there is actually quite rigid scaled relationship between them so it's organized like a geometric series where you multiply by a constant factor to get from the smallest one to the scale or the next one multiply again with the same factor to get the next one and so so on and this is probably a way the brain organizes its great modules in order to to represent location combinatorially with a minimum number of neurons so just to be sure we don't
believe this is special for rats and mice it's not so this shows the field of genetic tree for mammals starting with the common ancestor down here humans are up there rodents are here and they all have these grid cells and head Direction cells and also several other types of cells but they are also present in bats which are a completely different branch of the tree for mammals they are found in somewhat similar form at least in monkeys and in humans so these cells are present everywhere through mammals and it seems like this system was developed a long time ago at least in mammals and and has been preserved so then I wish to
say a few words more to getting more speculative but an obvious question is how is this localized activity generated I mean that this is really what we understand because if we understand how this example of very very clear patterns how does this arise among neurons in the brain that would put us on the track of understanding neural computation or computation in neural circuits because most likely what the brain does for one function like space is also used in different ways for other functions so that's why we put a lot of effort on this but it is also a different difficult question because since this is circuit I get activity involving many hundreds probably thousands of neurons that interact you can't really easily infer it from activity of small numbers of neurons at the level so you had until now so until now technology has only allowed us to record a few dozens or cells at the very best but this is totally changing now and and we are now up in how many hundreds or thousands of cells at the same time and that creates a new time where you actually can address mechanisms of population activity so I
wish to just illustrate the more mechanistic approach by giving some examples and showing one type of models for localized activity that have been around for some decades now and these are the continuous attractor models that have been introduced to explain both directional tuning in the head Direction cells and spatial tuning in place or later grid cells I will begin with the head Direction cells because this is one-dimensional and the easiest to understand so the question is how can you get cells that are active only at a particular location or only on a particular direction that you're facing your head and how can this activity then change as you turn you around how do different cells take over and one you get around 360 degrees the same cells take over again so what is the neural basis of this and the models that were introduced based on some more general models earlier on are as follows so here you have cells each cell is one circle here and they are connect that in orange so they're organized according to their direction and firing preference so all those that are on the top here fire when the are active when the send when the head is pointing in the north direction and then those that are actually down here are those that are active when the rat is facing to the west and so on so the point is that there are strong connections in this network between cells that have a similar function or similar tuning so all those that have the north preference are strongly connected and the closer there are in the directional tuning the stronger the connection this is enough then to create excitations among similar cells so that creates a bump of activity in this network but of course in order to control it you also need inhibition so the idea is that there is inhibition to those cells a little bit further away like in Mexican hats often called a Mexican hat pattern and in that way you can keep a bump of activity active without this spreading uncontrolled to the whole ring then the second element of those models is that there is some input from the outside about the actual movement of the head so the direction or the velocity the angular velocity of the head comes in to this network and when it for example the rats head is turning to the right or or clockwise here then this bump or activity also is shifted in in the clockwise direction when that input is counterclockwise than the bump of activity because of these inputs is shifted the other way around and then it can go around and that then explains how you have a localized activity and how it can move around in space and this remained very theoretical for a long time it made predictions that were consistent with observations but only very very recently how such a network actually been observed in flies whether in such a direction as such a ring structure has been seen directly activity moving around ring on connections between cells as predicted by this model so that sort of puts some actuality on them but these models were then is soon also extended
to two dimensions so the two-dimensional version is again that here you have cells now imagine these are place cells organized again according to where they are active in the environment and the strong connections to those cells that are active have placed fields in the same place and then avoid the ring of inhibition around which is not shown here and then you also have inputs about speed Direction coming in and then moving this bump of activity around so this was proposed for place cells in the mid 1990s but since place cells there's one problem with them anomaly that they are different they have different Maps or different combinations of activity in almost any different setting so it would be quite difficult or you would run out of cells if you would have to create such maps for every possible place in the environment so when the grid cells
came that became more more likely explanation because grid cells are similar everywhere the only modification you have to make is to explain why do you get bumps of activity in many places in the network and that's fairly easy because you have the bump of excitation locally and then you have the ring of inhibition around and that ring of inhibition then sort of prevents activity in each of the these bumps that may arise in different places to sort of go into each other and of course the most stable pattern that you can get is a hexagonal pattern where all the bumps are as far away from each other as possible so these are and then again you have inputs about speed direction that could move this activity around in the network so these are theoretical models and the predictions I mean to really test them one should observe this at the population level recording from many many cells in this net at the same time but until then that
hasn't been done here this on the way but there are still some predictions are consistent with this idea and one is that if if these models really depends
so strongly on the connections between the cells much more than the sensory inputs from the outside then they should perhaps be quite rigid in the sense that they persist across different environments and so they do so this
shows just this is just a schematic illustrating the point so here you have a rut running in one room and then in a second room and then back in the first room and if you record three grid cells for example the phase relationship between them so red is to the right of the blue and the green is below the blue and the red dot pattern you see here is still preserved also in the other room and of course when you go back and this you can show with cross correlation studies that that is essentially just one pattern of grid cell organization that is maintained in every possible room or every possible condition where this rat is is behaving and so this is what you would expect if it is really the intrinsic connectivity that matters so much of course this is not proof of the models far from that but it is enough to consider further experiments
and one other prediction is also that that if it is so much dependent on the connectivity then perhaps these grid patterns are even present in sleep when the the rats aren't really behaving and sensory inputs of kept away and this is some work that was just published by Richard Gardner in our lab and this
shows essentially shows the essence in this slide so here you have two grid cells one shown in red one in black and they have identical faces just on top of other each other in this box and to the right here use cross-correlation diagrams that shows the probability of activity in one of the cells given that the other cell is active to the left you see that cross correlogram for when the rat is running in the box and of course this is what you would expect so if cell number one for example the black one is active then you would at this time zero see a higher probability of activity in the second cell as well and that's what you see you see a peak around zero seconds and at other times it is no but the more interesting thing is that if you now take the same cell and ask what is the co activity of those two cells in sleep I don't slow asleep orals in REM sleep and you see that there is still a peak that means that those cells that are active at the same time in the running state or when the rat is running in the box those cells are also Co active in sleep and opposite those cells that are not active together like here in the running state they're also not active together when the rat is sleeping that suggests is essentially just one map that is preserved are also
shown here for a group of cells recorded so what you see here is again time on the x-axis each row here is one cross correlation diagram but now the cross correlation is color coded using a GLM here and the red is high correlation so peak like this and black is negative that means they dip like you saw in the other image so this is what you get when you rank order the cells according to correlation in the running state but if you then record the same cells in sleep you can see is essentially the same which that means that for the whole population the pairs of Co active cells are just the same when the rat is sleeping as when it's awake which means these patterns are likely not strongly depend on any sensory inputs that come in the actual position of activity in the network of course is when the rat is running so the the bumps of activity move around depending on on where the rat is or what kind of speed inputs come up to the entorhinal cortex but but in general the the fact that there are such locally localised activity points that is probably created by the brain itself ok so now in the tour in this in the
last part of the lecture I want to get to where we are now questions you are working on now because these models are quite idealized and one of the reasons is that there hasn't really been opportunities for testing them out there are among the most dominant theoretical models in neuroscience and but the testing has been lacking due to lack of tools but one of the question that they don't handle very well is how his brain activity organized temporally and what trajectories does activity take in multi-dimensional space as that's running around in the environment and this is work that we are doing now with especially Flavio Donato and sold our con own host urban house who have been recording activity from the medial entorhinal cortex in mice in a very
simplified environment so here this mice they run on a ball in an environment that is completely dark they were run on a ball and they don't get forward and that allows them then to be head fixed under a microscope that then allows us to image activity in the brain and specifically in the medial and renal cortex activity or money many cells while the rat is running stationary in complete darkness will no particular goals but mice like running so they just run a run around take a little break and run and then because this is so stereotypical behavior it it has some advantages if you want to study the organization of the activity so just how
is this done well a virus is injected when soon after they are born into the medial internal cortex this virus causes the expression of fluorescent cultural monocle so gcamp and this molecule when expressed in the cells about two weeks later then when we look at them under the microscope through a prism onto the entorhinal cortex we can look at a surface here then we will see yourself blinking depending on the calcium level in the cells at at the moment so you see different blinks here there are all different cells in the network that is a bit hard to see but there are actually hundreds of cells in this image so you can in and this calcium activity is a good approximation of the firing that you heard on the movie the pop op op op op it's just that it is has a slightly slower dynamics but you can still infer the timing of the so-called action potentials when the cells discharge so while doing that then here's another
recording for 200-plus cells over a time of somewhat more than 1,600 seconds and each black dot is when the cell is active and here you see this is typically what we see only record from many cells in in in the brain area and the cortex it doesn't really look like that is much pattern here but actually the nice thing about the entorhinal cortex is that activity is organized according along a low number of dimensions so doing a principal component analysis make some sense so what solar dot in Allah cogno did was that she organized the cells according in sequentially here and divided the time into small frames approximately 150 milliseconds for example and then she did a PCA with the cells as variables and the time frames as observations and then she for each sense you know then she organized or ranked the neurons reorganized them according to how much they loaded on each of the components for example principal component one the center had the highest loading on that component was then put on the top and so on depending on on the loading and if you did that then suddenly much clearer a pattern appears and you see that there is a repeating sequences or activity that go occur all the time during the recording and if you magnify a little here you see that is actually sequences these aren't instances of simultaneous activity on and off but it actually their activity runs through a long long sequence or many many tens of seconds and which go over and over and over again and it's also if you're the same
that you see if you use if you use nonparametric methods for sorting or reducing the dimensions you get the same sequences repeating all the time so how
to quantify it while it is possible and to give to to create a matrix here where you divide not only time into bins but you also been the loadings onto the principal component which might for example be principal component one so let's say you divide it in ten then in so for each one tenth of the data you can then define for each time frame is the activity higher than you would expect just from the general background and then you can ask for each time frame which been on this principal component axis here is the most active one and you put a frame around the most active one for each time and you already here see the example of this is not random in time it actually it's sequences that are on through time and this is what you then
get if you plot so called transition matrices so what is the probability of a transition from one pc beam to another PC beam and you see a clustering along the diagonal which then means that that most often activity transitions from one principal component pin to one that is very near it and see the same also if
you plot it like graphs here so these are again the pc bins going from blue to yellow and you see that most of the lines of transitions are actually between neighboring bins rather than randomly across the data as as if you keep the original sorting and if you
then have these graphs like this and then you go back to the data analyzing frame by frame so what we did here was that we looked for all possible transitions or possible sequences among those that were significant here looked for them in the data and then marked in this case with blue over red depending on their and whether they went up and down all those that matched a possible sequence in in this map here and then sorted indeed asked for different time bins anything from one second bins to 40 second bins and then took the one that gave the biggest difference from the chance level and what we then found in in the data in almost all cases is that there are these long long long sequences that can last for in some cases up to two three minutes that are significantly way significantly above chance that that happened when the mouse is running on this wheel in this very stereotyped condition you see an extreme case here here the period is approximately a few tens of seconds so activity sort of repeats itself all the time at a very very regular frequency this frequency depends also on the age of the animal in
the beginning there is nothing really so but then when it gets about twenty years twenty days old and up to thirty and it starts to organize itself and ends up with relatively short periods of about some tenths of seconds which then coincides exactly with the timing of maturation of that network so network is full only fully connected really when you get these sequences here and we're
still looking working on whether this really are sequences most likely it seems now from data that they are actually circular should I just go wrong around around in a loop when the lattice Mouse is doing this very stereotypical behavior and quite independently apparently or its actual behavior what else tops or runs so I think what this
tells us then is that this is a quite hard wired network that generates similar a lot of its activity through intrinsic mechanisms intrinsic connections of the network that define when cells fire and in what sequences and that well sensory inputs may shape what directions it takes in the real world but it is strongly dependent on connectivity that is all already there that involves the whole range of ourselves in the network and finally
just to find out what is happens when that actually is not only in such a stilted environment but runs in a real environment then we are now setting up experiments with with mice that have that have this very very small microscope weighing two grams that can be put on the head of a mouse when the mouse is running around in the environment and this
Microsoft connects so this is the work
all host open house and raging song based on a microscope that song and colleagues developed in in at Peking University two years ago which connects is meaner to a microscope to a laser that is outside through a cable that manages to keep the precision of pulses at the femtosecond level over two meters without distortion so this is quite some
revolution because it then allows two photon imaging of activity at very high resolution so again here you see many cells active at the same time and now you also see the mouse running around in the box color indicates direction and this is whether the mouse is in the box and at the same time you see 200 something cells and the activity of these cells Y in the mouse is at different positions in the environment and this just shows again pretty good
signal to noise ratio so you can see you can easily capture the activity of cells or in this case of a 259 cells that passed signal-to-noise ratio or 1 to 5 and those are then shown in the network here and this is just the beginning so I mean expect these methods to be in the many thousands very soon so this is the
people who involved in it so all work together with my Britt Mouse Richard Gardner for the sleep in particular Flavio Donato soldered Congo and host open house for the MSE trajectories and also for the for the new work on on the mini scopes and host open house and wage on song particularly and of course lots of people pay for this so with that I want to thank you a lot for your attention well thank you very much for this fascinating talk and again we have time for probably two questions from the audience yes lint it's been surf thank you so much for a really interesting talk it
seems to me that all the advances that we make in physics and medicine chemistry and biology are preceded by new engineering systems for measuring things more accurately and this wonderful interaction between scientific speculation and engineering development is what contributes to all these successes so I'm wondering what's the next thing you're going to try to measure well I think we are at a point now where we actually are tipping over in a completely new phase of neuroscience because of I actually put it very nicely because it is it is because of the new technologies that allows us now to record many hundreds thousands of cells at the same time which you really need to when you want to study the networks but it also this can't go alone it has to be go together with theory and models computational models of how the brain might work which tries to reduce it and looks at the possible mechanism and ignore all the noise and those two when they go together that's what creates advance and luckily enough this brain area on this brain system is one what this exists so I I think I will stay in that area and our idea is one of the quite I mean we want really to see this attractor in the activity of these many cells at the same time I mean sometimes I say it's like seeing the black holes finally right so so I think this is because it has been very much inferences so far but of course I also hope that that this now will stimulate more theoretical development because that actually was lagging behind I think the here the methods come all the time but theory develops much more slowly and there's a high need for more work on that short question complementing the previous one how far we can go in this approach mapping high-level concepts into activities of cells how much you think we can go in in which direction mapping concepts like time space or into activity of cells you mean technically how far we can go yes oh yeah much further I think so I mean just within a few years there has now changed from dozens to a few hundreds and already the recordings from a few thousands so I think indefinitely what I think perhaps is more important and only increasing the numbers is to do multiple sub circuits at the same time to see how they interact so but but these devices all get smaller and smaller so that you can I mean this will continue for quite a while still but of course then you also need to develop the analytical tools as well at the same time which I've totally new now so the question is to what extent by studying only brain activity we can understand what happens higher level which yeah yeah that's the question yeah yeah no I I think we can but what the limitation there that's another limitation because if you want to understand high level brain activity we also need better methods for studying behavior so the psychology is also like lagging behind because quite often we don't really have good ways to measure for example how a mouse is planning although definitely a mouse is planning where is going what's going to do but all these behaviors have to be quantified better I think this will come and machine learning is definitely important actually in order to do this nowadays but when this comes I think that we can also understand much more advanced behaviors then then space space is very simple after all [Music]