[00:00:05]
>> OK thanks for that introduction it's great to be here. So the subject of my talk will be. How basic sensory stimuli are processed in kind of their probability context and the primary method I use is to put on calcium imaging but has to do a lot of local field potentials and it's on Awake mice.

[00:00:29]
So. So basic cortical processing of sensory stimuli is significantly influenced by the context in which that stimulus appears so if you were to suddenly appear in a context with a bunch of other fruits you might expect one level of response and primary visual cortex whereas if that same stimulus were to appear in a context where there's many many apples it's expected it's a done deal but we actually observe suppression of responses to that stimulus and then convert that stimulus would appear in a context where the observer expects oranges you get a sort of amplified response that stimulus so this basic sort of contextual modulation of basic sensory processing has been studied traditionally using oddball paradigms so this paradigm involves is repetitive presentation of the same stimulus over and over and over to the point where that stimulus becomes redundant and then every now and then throwing in in unexpected or a deviant stimulus this is then recorded humans non-human primates rodents and so forth.

[00:01:37]
And the idea of using for instance E.G. is that the deviant or the unexpected stimulus elicits a stronger response than the standard or the redundant stimulus and one way this has been quantified is where would miss national get to Vittie So what you do is while recording your response from the scalp you average the responses to the repetitive stimulus and you compare that to the waveform of the average response to the deviant or unexpected stimulus.

[00:02:05]
What you get is a sort of negative going potential here at about $100.00 milliseconds or so after the onset of the stimulus so why this is particularly interesting is that we know that individuals with schizophrenia show a smaller mismatch negativity meaning that they have a smaller difference in their evoked response to the unexpected stimulus versus the expected or the deviant versus the redundant this is relatively selective to persons with versus bipolar disorder or other mice major psychotic disease.

[00:02:36]
So. So what this in particular buys us is that if our overall goal is to understand the neurobiology of a complex disease like schizophrenia. What we essentially need is to understand how variation at the genetic level gives rise to sort of changes or path ology at the molecular cellular circuit systems and eventually behavioral level we need to sort of understand how these these different levels interact and we need to be able to sort of pinpoint where exactly things kind of converge in this disorder.

[00:03:10]
But one major complication is that in patients we can't directly access these levels of function non-invasively So we need animal models so far. In a bigger problem is that how do you study a complex disease like schizophrenia in a mouse the idea that you could directly observe the presence of delusive delusions or hallucinations or disorganized speech in behavior is somewhat somewhat insulting to our sensibilities in the sense of how do you know if the mouse is psychotic How do you know if it has schizophrenia it one way to get around this is by studying context processing so instead of trying to study the holistic disease of schizophrenia you study sort of the mismatch negativity or a biomarker of the disease something which we know is relatively selectively affected in schizophrenia but it's perhaps more proximal to the cell and circuit level dysfunction which might underlie these so this is the general strategy of my work.

[00:04:10]
And this is the way my my my talk will be set up 1st I'll kind of discuss the paradigm and how I study sort of sensory context processing and mice and then after I have that kind of model hammer down I'll show how this context processing is carried out at the single neuron level at the level of intern or ons at the level of local populations.

[00:04:34]
Enter enter laminar circuits. Granular supergroup you learn for granular and then also how this is carried out across just the brain regions and then looking forward. Bring up some sort of ideas and hypotheses as to how these different levels of function interact to give rise to sort of basic.

[00:04:54]
Contextual modulation of sensory processing so to begin with can we study kind expressing in mice or does or does the mismatch negativity biomarker translate well so this is this will be the majority of the data that I showed today will be in this very basic paradigm is awake mind.

[00:05:12]
Had fixed on a treadmill and was showing them full field visual squarewave grading stimulus so in this case you just present the mouse with the same oriented stimulus over and over and over to the point where it becomes redundant and then every now and then change your name Taishan and sort of quantify the responses to the deviant stimulus verses that are done in stimulus in order to sort of measure cortical responses we use to photon calcium imaging.

[00:05:37]
So this strategy what you virally express G. can't. Have been all the data today are from using G. camp 6 S. and 30 Hertz imaging they express these calcium indicators in primary visual cortex and then this is just data from an awake mouse. Without the presence of sensory stimuli you can sort of see hear individual neuron fires an ash potential or a burst of action potentials the neuron gets brighter in the field of view so you can.

[00:06:08]
Get activity information from individual neurons and also get an idea of how they they relate to one another in a local area. So it's a potentially powerful technique and here's some real time data on the left here is the visual stimulus that the mouse is seen in on the right is calcium imaging from primary visual cortex during this paradigm you can see that the majority context is relatively silent even though you had this very big large full field oriented grating but then when the deviant stimulus appears you get a very large response from a large number of neurons so if you're having a hard time seeing you're going to sort of focus on a few of the neurons in the field of view and then they're mostly silent and you get this deviant stimulus and they'll come on line so you can what they go through another time or 2 but the idea is that this responses has a robust enough and it has a high enough signal to noise ratio that we can observe it in single neurons and single trials.

[00:07:06]
So this is averaging over trials each individual row is a single neuron. And also a lot of plots like this time is the X. axis and you can see that the average response. To the stimulus when it's contractually deviant is significantly lower than the average responses to the stimulus one.

[00:07:28]
Deviant didn't versus redundant and the color represents how large the calcium transience were as this is sort of an index of bursting. So if we come back to this you can imagine at least 2 things are going on here 1st you have large suppression of activity repetitive stimulus so the stimulus is occurring over and over and over to the point where it's redundant and most of the neurons are not firing at all even given what you'd expect.

[00:07:58]
Basic sort of. Orientation selective responses in primary visual cortex you still see essential nobody firing but then I'm a conversely So you have suppression of responses to repetitive stimulus but you also have this a large augmentation of responses to a deviant stimulus this is kind of by directional you get sort of suppression you also get a deviance detection and the way to tease those out is by using a number of different paradigms.

[00:08:23]
So 1st is run your typical oddball paradigm where this particular 135 degree grading is your conventional deviant and then you can run it again by flipping the D.V.D. in or Donda and so you can get responses in the same cells to the same stimulus when it's in a different context when it's expected and then you can run a 3rd paradigm which I'll call the many standards frequency control in this case the stimulus is neither a Dundon or deviant it's in the context where there's many different orientations occurring so the stimulus is still rare but there's nothing unexpected about seeing this stimulus so you have these 3 different contexts here another way to visualize this is this is just the same sort of schematic but plotted in a different way here in the context neutral control there's many different orientations whereas in this particular paradigm that stimulus is the conventional deviant and then this one the same one is the redundancy so we run each one of these each one takes about 15 minutes with about 5 minutes in between.

[00:09:21]
And so here are responses of the yes. Yeah yeah OK so this is a full 1st half it'll just be layer $23.00 rounds so yes it's a good clarification. And a lot of these properties are of our most dramatic in Layer 23 it's kind of a tease there but but yeah so this is maybe about 150 to 250 microns from the surface.

[00:09:55]
So. So here you have responses of the same neurons so each row is a neuron in low to 3 a visual cortex. To the same stimulus when it's when it's kind of context neutral it's neither redundant nor deviant when it's conditionally redundant and when it's. So if you average over over these neurons and each one of these contacts what you see is the spider actual photos talking about the black line a sort of you're neutral context and this is the Y. axis is sort of scored calcium transients you can see this sort of one level of response to is expected in the sort of baseline condition when the stimulus is contextually redundant or repetitive you see a suppression of responses and when the stimulus is going to actually deviant you see an amplification of responses so this bar graph is the same data just kind of collapse over time.

[00:10:52]
This suppression of activity to repent of stimulus has been referred to as stimulus specific adaptation whereas the amplification of activity is what I call deviance detection. Right so this is this is what's going on in individual neurons when you average over neurons when you average over trials and also typically try to record local field potentials during this paradigm as well.

[00:11:17]
So this is from a 16 channel change from their nexus and if we just take one of the traces and average over these 3 contexts you can see that the sort of average L.S.P. waveform are also modulator by deviant versus redundant contexts and if we look at what the human E.G. response looks like in this case this is just taken from a scalpel of trode in Italy sort of a sensibly.

[00:11:44]
Sort of wave forms like relatively similar only pick from deeper layers and you also have this sort of. Negativity that we observe in mouse visual cortex which is very relatively similar to what we observe and human patients so we can only say to some degree we're recording something that looks like this sensory biomarker of schizophrenia in a mouse model.

[00:12:07]
But what's not typically done in the human literature is to split it up into adaptation versus deviance detection and what you can see from P. which has a little bit better time resolution this sort of adaptation response is occurring a little bit earlier so the blue line is the average response when the stimulus is redundant.

[00:12:26]
And sort of that sort of stimulus Pacific at attention occurs about $100.00 milliseconds and whereas your deviance detection or your major difference that's a listed by unexpected Simulink occurs a little bit later so we also had this time differentiation between the 2 effects as well which is nice.

[00:12:44]
Another another thing you can do with this technique is you can look at current source density profiles and I haven't dug super deep into this yet but dismiss service you kind of you take this 2nd spatial derivative you can get an estimate of local current sink and sources just from.

[00:13:03]
Kind of. $35000.00 foot level observation you can tell that sort of there there's definitely a different deviant response than there is to the control and then also contractual redundance also have a significant decrease in their current source density profiles as well so there's a lot to kind of dig into there which is yet to be explored and then also sort in the human literature and human E.G. literature one thing that's been identified in this paradigm is that by breaking things into the time frequency domain you get a much more reliable estimate of context processing so what this is is just taking a single trace.

[00:13:44]
Using credit moving window Morely wavelets decompose the single trial. Response into a time frequency. So here you have time on the X. axis this is milliseconds and you have frequency on the Y. axis and this is the typical response to one of these visual stimuli. Time frequency domain so you had this nice kind of low frequency increase in power and you also had this sort of gamma band increase in power 2 to a visual stimulus Now what's interesting is one with stimulus is redundant you don't see a whole lot of modulation in the low frequencies but you see kind of a suppression of this gamma band response whereas when the stimulus is contextually deviant most of the modulation is down here in the low frequency so not only do we have a time.

[00:14:32]
In the F.P. not only sort of time differentiation but we also have a frequency domain differentiation devious detection appears to occur mostly kind of in your data alpha beta. Oscillatory range or as adaptation tends to occur a little bit more in your Gamma band range as well. So to summarize that sort of yes this is just the same data but.

[00:14:57]
Sort of averaged over time after the stimulus you have frequency on the X. axis and power on the Y. axis so you can kind of see how the conditional deviant. Amplifies power kind of in your low frequencies where is your redundant stimulus results in a modulation of gamma band activity so if we kind of just summarize this here we know when the stimulus is conditionally redundant you get a suppression of responses and neural spiking and you also get a suppression sort of gamma band activity or as I want to stimulus is going to actually deviant you get an amplification of responses in single neuron spiking and also a increase in low frequency activity so.

[00:15:41]
To summarize the 1st part can we study context processing in mice and local sensory cortex I would say yes to some strong degree in even to some degree looks a lot like the human mismatch response. So now that we have this model now we can start to then look at individual levels and manipulate with dreads and not the genetics in different optical techniques so the 1st part I'll discuss how sort of contacts processing is carried out by local Internet rounds in the visual cortex.

[00:16:12]
So. The majority of neurons in neocortex are pyramidal cells so these are about 80 to 85 percent these are excited to are neurons and they have relatively. Stereotyped kind of morphology in firing patterns and they both locally and distantly whereas a subset of your neurons in neocortex are we call intern runs in these tend to be gather their inhibitory and there's a there's a large variety of subtypes in the to all mainly be focusing on the day or some modest and containing intern runs in poverty and can data into neurons so.

[00:16:52]
Make up about 40 percent of your of your gabber Dick cells. And they kind of broadly target all cell types in the local area they they target both criminal cells and understand are neurons and even themselves other P.V. intern runs as well as they have author apps as well so they're kind of broadly.

[00:17:15]
Applying blanket inhibition and they have a much faster fire and rain they tend to support gamma band oscillations. So mad a stand containing insurance especially within visual cortex in mice. Are tend to be dissin have a Tory whereas instead of targeting everybody relatively equally there synapses on the part of human interference are much stronger.

[00:17:36]
So they tend to relatively selectively. Subserve disinhibition So when they're active they suppress Peavey's which doesn't have its local pyramidal cells and they have a much more moderate firing rate and they're thought the sort of support Alpha and Beta band oscillations as well. So it's particularly interesting about these 2 cell types in this paradigm is that both are known to be affected in schizophrenia so looking at post-mortem brain samples we see this.

[00:18:07]
Especially in visual and auditory cortices you see a significant reduction in P.V. M R N A And so modest an M.R.I. which brings up the idea that maybe sort of our contacts processing deficits might be due to. Deficits in the function of these subtypes of interference so. That was kind of a hypothesis and so the goal was to 1st kind of suppress some at a stand intern runs and see what how that affects kind of your mismatch negativity so in order to do this they use dreads which stands for design or research there's a schools of lead activated by designer drugs so and this involves is sort of virally targeting a sort of inner inhibitory receptor to some out of Senator neurons which typically doesn't do anything to the function of these cells but then when you add a inert Weigand close opinion oxide the combination of these 2 elements leads to the suppression of synaptic output from these cells so we can sort of measure a sort of context crossing deviance detection and so forth 1st and then apply the drug in which the process matters stand Enron's and see how that affects the.

[00:19:20]
Detection. So here this is before suppressing some other stoner Owens again you have amplified responses that they can totally deviant stimulus in red and you have suppressed responses to the can actually were done in stimulus and blue. And then after we suppress them out of stone or neurons on the whole.

[00:19:38]
Local pyramidal cells are still sort of. Responding to the stimulus and you still have stimulus Pacific adaptation you still have suppressed responses to redundant events but you no longer have a devious that action so taking some understand and runs out of the picture relatively selectively affects just this part of the sensory context process and that's what goes on at the single neuron low when we then look at the L A P Again looking at time frequency responses again as I mentioned before your deviance to this is before suppressing some modest an intern runs in again you have sort of frequency on the Y. axis and time on the X. There's your responses to the stimulus in the neutral redundant versus deviant case and as I mentioned before you have amplified responses and the low frequency domain kind of track deviance detection or as modulation the gamma domain track adaptation so we have this frequency differentiation this is the same data again just average over time this low frequency augmentation is devious detection whereas.

[00:20:44]
Similar specific get attention is this gamma band thing when we suppress some at a stand in neurons. You still have stimulus specific adaptation just like you had in your in your single neuron responses but the low frequency modulation is gone. This is kind of this is just comparing the time average data back to back here you still have this gamma bend augmented kind of suppression to return and stimulate both pre and post but you no longer have this.

[00:21:18]
Low frequency deviance detection response. So to some degree of showing that aspects of deviance detection are dependent on the function of some medicine or ounce but what about P.V. intern or ons the other kind of self I chose to focus on so use the same technique used drugs to suppress T.V. Internet and this is just focusing on basic sensory processing of moving gratings and what you see.

[00:21:46]
Just to show that the suppression of T.V.'s is working well it doesn't have its the earlier response so again if you have time on the X. and frequency on the Y. to distant habits the early visually evoked response but you no longer have this kind of sustained gamma band oscillation when you suppress Peavey's So this is expected in this tells us that our manipulation kind of was consistent with the literature says it should do also if we look at spontaneous activity after we suppress Peavey's here the left and right show the same data but the right as individual neurons so these are non P.V. cells.

[00:22:24]
And this is the spontaneous activity before and then after suppressing Peavey's So the identity alone would mean they had the same activity before and after but that's not what you see when you suppress Peaveys as expected you disinhibited spontaneous activity so I show this to say of course manipulation is doing something but it's not doing anything to our context processing so before and after suppressing Peavy internet runs you still have augmented deviance detection augmented responses to deviant stimulating and you still have suppressed responses to redundancy mainly So so this deviance detection appears to be relatively selective at least when we're comparing the psalms in visual cortex to the question is why we know that manipulating schizophrenia relevant affects the schizophrenia relevant biomarker But this is sort of ongoing work that I hope to hammer out we need to sort of record directly from each of these cell types during this paradigm and sort out one hypothesis for why this might happen is so we know that if you repetitively stimulate P.V. cells what you actually get is in adapting response so this is just sort of.

[00:23:38]
Door according from a regular spoken cell to. P.V. cell. And you see that sort of Peavey's adapt to repetitive stimulation whereas some understand and runs through the opposite so as you repetitively stimulate them they get larger and larger posts not the potentials so putting this these 2 facts together with the fact that some out of center on suppressed T.V.'s you can imagine during an oddball paradigm as you are perfectly stimulate the local circuit Peavey's are decreasing their activity whereas some out of step in and neurons are increasing their activity and this would lead to kind of a general disinhibition so that when the deviant stimulus comes in everybody's disinhibited and you get a sort of super threshold response so that's something that test.

[00:24:25]
And that's a work in progress so how is it carried out in the local circus where we know some out of center and they're involved. But they sort of zoom L. and kind of instead of focusing on the single neuron case kind of get an understanding of how local intern runs interact during this paradigm so.

[00:24:45]
Again you can see that we're able to simultaneous or core a large number of neurons during this paradigm and this is this is what it looks like if you just average over each one of these. But one thing we know is that there's a lot of variability from trial the trial and how individual neurons respond to the same stimulus even if 2 neurons had the same orientation selectivity they're not necessarily going to fire exactly the same to the same stimulus across time so.

[00:25:12]
Also if you look at this you can sort of or at least this was the thought process that we had. It's not exactly clear whether the same neurons are doing both of these modulations whereas on average the whole population is kind of suppressing the redundant stimulant amplifying to deviant stimulate it may be the case that subsets of neurons do different aspects of this context processing and so the North to explore that kind of looked at this variability and ask the question is it kind of.

[00:25:45]
Is it. Are there sub populations of cells or sub clusters that kind of act differently across contexts and across time. So I get into the exact methods I used to tease this out someone's interested but essentially you do K. means closer analysis and ask the question how well is the within cluster distance is reduced if you add 2 clusters 3 clusters 4 clusters 5 to the solution and you compare that to a shuffle dataset and what this told us is that a solution with 3 clusters are sort of 3 cell subsets which do different things across these contexts was kind of the most optimal but we don't really need this analysis to tell us that if you just look at the 3 D. scatter plot you can really see that there are there are 3 clusters there each one of these dots here is an individual neuron and the X. axis is the average response kind of redundant stimulus.

[00:26:43]
The Y. axis is the average response to the control stimulus in the Z. axis is the average response the deviant stimulus so you can see the different subcultures in neuron seem to be for in different contexts here. And so if we take that same data and we sort them out into their subgroups that's exactly what you see so this 1st cluster here we can call deviance detectors and what the idea there is that they're mostly silent when a stimulus is redundant or kind of contextually neutral but they can really come on line in fire only one that's an unexpected event so we can call these deviance detectors and they'll be in red there's another subset.

[00:27:21]
That seem to be suppressed during the oddball paradigm in general regardless of whether the stimulus is redundant or deviant but they're typically active during kind of a more neutral context so we can think of those as generalized adapting cells or perhaps they're deviant suppressed cells not exactly clear and then there's another subset which doesn't seem to have any clear contextual modulation at all if we just take the same data here and we just average within groups to simplify things here are the 3 clusters you have your general adapting cluster which are relatively silent.

[00:27:53]
During an oddball paradigm but they tend to be active during your typical multi orientation paradigm you have your deviance detectors which are I think the most interesting subclass are and then you had this kind of non modulator mix modulation so close to it as well and this wasn't just sort of an artifact of shifting fields of view or anything so if you just take the same cells and you record again an hour later.

[00:28:18]
The same cells preferred the same context so you start your generalize adapting cells are still selectively active during the kind of multi orientation paradigm whereas your deviance detectors are still devious detecting and so forth so this is also true if you just take the even trials in the new and you make the clusters on that and compared to the trials so this was pretty robust.

[00:28:41]
And what's important is that sort of these cells are spatially intermixed So if you just kind of plot where they are in the field of view it's not like there are certain kind of sub regions of primary visual cortex which have devious detectors versus adapters and so forth.

[00:28:56]
So. And then again it doesn't look like there's sort of a shift in a field of view you're generalized adapting cells really are active during your control run but they're pretty silent during the oddball run and then vice versa your deviance detectors they're still there still in the field of view the signal still still there it just kind of comes online during when there's a deviant So another interesting aspect of this is that you're subtypes your kind of context preferring cells also show more correlation within type than between so if you ask the question kind of you have all these deviants detectors in the field of view and you look at spontaneous activity when there's no visual stimulus.

[00:29:36]
You see much stronger correlations between cells which prefer the same context than between cells which prefer different contexts so it doesn't necessarily purely show that these cells are synoptically connected and this is something I think that look at but at least suggests that these do form. Some sort of sub networks are context before and so no it works they could be sharing similar input or they could be sort of parsing different internet runs may target one type versus another.

[00:30:07]
But this was this is relatively surprising because a lot of the ways we've previously analyzed in the same way everybody else had previously analyzed that sort of assumed that the cortex was doing by marginally. Better actually modulator instead of there being sort of subsets of cells that do different parts of it so we also looked in audits where cortex.

[00:30:27]
Here again using awake mice on a rotating treadmill again using success but the difference is sort of the head plate has to be pretty far over on the side of the head. And. Sort of moving some muscle but after we sort of work those kinks out it'll work very nicely another thing that you need to do when you're doing primary auditory cortex vs visual visual cortex is very large an easy to find sickly but with auditory cortex you definitely need to do a mapping experiment 1st to determine whether you're looking and primary versus secondary or tertiary areas so you just do sort of.

[00:31:07]
Kind of a pure tone you solely modulating gratings and map out your areas 1st with the one photon approach and then you can focus as we did using to put on imaging just on primary auditory cortex the question being does primary auditory cortex and primary visual cortex both show the same sort of context preferring ensembles.

[00:31:29]
The stimuli we used for the eyeball paradigm were not pure tones but in stead they were sort of these. Time frequency gratings and just kind of on the service they look a lot like the visual stimuli that we used now is intentional because so this is these stimuli go from.

[00:31:46]
3 kilohertz all the way up to 90 killer it's and they're sort of son you saw in modulating and time I don't have an example or at the bat but they do sell some kind of funky but the idea is that these these stimulate the whole tone a topic auditory map and so again we can sort of use a.

[00:32:06]
Flip flop auditory auditory eyeball paradigm and also in many centers frequency control. So here is after averaging all over all of your auditory neurons again you had this by a directional effect as you saw in visual cortex these are calcium transients you have your conduct surely redundant stimulus leads to suppress responses relative to the context neutral control whereas your conditionally deviant stimulus is an augmented response and then again we do a K. means closer analysis what you see very clearly is a subset of auditory cortical neurons prefer different contexts so actually the largest chunk from this is just from 3 mice This is an ongoing work but the largest chunk of cells actually were devious detectors and primary auditory cortex I'm not sure yet if that's going to hold but if we just kind of compare back to back here that this is that these were the sub clusters in visual cortex and these are the closers in auditory cortex and you can see very nicely you have these devious detectors.

[00:33:13]
In both cortices you also have these. Cells and both cord sees what's different is you don't have generalized adaptation an auditory cortex and I'm not quite sure why you have much more specific thing so this subset of cells isn't doing deviance detection they don't amplify their responses to deviant stimuli but they do suppress their responses to redundant Simulink So this tells us both stimulus specific adaptation and even sedation can be studied auditory cortex but again just like visual sort of carried out by different subsets of Layer 23 neurons.

[00:33:47]
So. So how is sort of context processing carried out in local ensembles will or local populations was done by a subset samples actually rather than all equally and all neurons. Again as as was brought up this is mainly been done in Layer 23 So the next question is What about if we sort of extend to Layer 4 and layer 5 Do we still see.

[00:34:14]
Devious detection Do we still see these contacts prefer ensembles. So you just kind of. Summarize your canonical Neo cortical circuit this is highly simplified normally leaving out layer one and layer 6 here but the majority of your bottom up them foot comes from specially and primary visual cortex comes from the.

[00:34:37]
Basic really relate the elements to Layer 4 cells where it's fed forward to layer 23 neurons. And then sort of you have a lot of sort of local or lateral connections you can think of this is this sort of local computation and then these sort of synapse on layer 5 cells which send driving input the higher and also sort of modulatory feed back to your you're really down it's sort of you can split this into sort of input local computation an output if you want to use such a heuristic.

[00:35:10]
So. All the work I've done so show you so far is from Layer 23 but working with wage and Yang who is a post-doc in. In the used a lab and now he's assistant professor at U.C. Davis he sort of developed a technique where we could measure 10 different points at the same time with 10 hertz imaging so the rest of what I showed you is from 30 Hertz But but 10100 still gives us a pretty good picture of how the neurons are responding to these stimuli.

[00:35:41]
So this involves injecting kind of a red calcium indicator in these deep layers in a green calcium indicator in the superficial layers. And then using 2 different lasers one targeting deep to hit your Jr get go and then one targeting shallow and then. Special light modulators to sort of split this beam into a different Z..

[00:36:09]
And also a lesser tunable lenses to very rapidly switch the Z. focus over time so with this gave us is exactly that. All very large number of neurons from from a nice sampling of cortical that. And then sort of the goal was to record the same paradigm across these layers so 1st off this is from Layer 23 again and this in a totally different set up in somebody else's hands replicated.

[00:36:39]
The effect of sort of deviance detection is mostly carried out in a subset of cells and they make up about 40 percent so that was nice to see and also you had your generalized adapters are also present layer 23 so this was right down to the proportions of cell types was replicated what I saw in the previous experiments.

[00:37:02]
But we looked at deeper layers you saw much less deviance detection. So on an average the number of cells which are classified as deviance detectors and Layer 4 in layer 5 was about half to a 3rd and then if you just sort of average over all the cells there's really no deviance detection going on in these deeper layers which was kind of interesting it seems like it's mostly a layer $23.00 function.

[00:37:29]
So you're the majority of cells are kind of active in your control paradigm and largely suppressed during the oddball paradigm. So exactly why. Isn't clear but this is just kind of another depiction of of all is going on here the X. axis is yours its depth you have layer 234 and 5 and these lines indicate the absolute number of neurons that are present from each cluster in each one of these layers do you see the generalized adaptors in gray and they're relatively the same overall number across areas but you have much fewer deviance detectors the dotted line just tells you.

[00:38:09]
Relative to the missing in other Z. axis here but tells you the relative portion of cells in each layer which were driven by the visual stimulus so that wasn't really different what was mainly different was just the absolute or so the relative number of deviance detectors so devious to take action on average and also the proportion of deviance detectors is much more of a layer 23 phenomena.

[00:38:32]
So. So that's something to kind of a little deeper into as well. OK So we have kind of what's going on. Sort of contribution in turn around to play what's going on when you consider populations and what's going on when you consider different layers of cortex and sort of the next scale to look at is could only consider different brain regions so visual cortex is not an island but it receives a lot of sort of bottom up in from stillness and also top down input from prefrontal areas so this was the next level of function to kind of depict context crossing at.

[00:39:11]
So what I mean by bottom up is I mean to input strictly from our midst and this is a real life so they receive their primary input from the retina in the eye so this is kind of the initial relay of sensory information to visual cortex the question is What are these inputs doing during this paradigm are they carrying deviance detection or not and then also asking sort of are are down circuits so we know mouse prefrontal area C.G. or a CA which stands for anterior cingulate area in the mouse this sort of higher level cortex projects directly to the layer one of the one so we have to question how are these 2 subcircuits contributing in a context processing Do they kind of contain these bills or not or or they causally related.

[00:40:00]
So how do we probe these inputs or one way to do it would be to sort of express your calcium indicator G. camp and then carve out cortex and in focus your microscope directly on L.D.N. but you damage a lot of tissue in the process but another way to do it is just sort of express your calcium indicator and then focus the microscope still on V one but making sure that you're very kind of careful about leakage and spread.

[00:40:30]
You can actually just kind of directly observe these boot ons and directly observe the sort of axons from. The one and so this is kind of some real time data. This is in this is in visual cortex but these are the lamb a quarter call axons and you can see you can resolve individual acts on all branches and and do tons so this is just spontaneous activity and it's nice and and clear the signal to noise ratio by using this approach is comparable to that using a.

[00:41:03]
By by using custom indicator and focusing on the soma So here is sort of your D.F. over a trace of sort of burst firing in your V one neurons and you have the same thing in these sort of X. all inputs. So using this technique we can actually get information about what signals are being sent directly from your relay to visual cortex and this is what we did so again using the same paradigm here is.

[00:41:32]
Responses of the same sort of mic axons to the same stimulus when it's contractually neutral versus redundant versus contractually deviant. And so that you have. The bad you can see that kind of modulation isn't quite as dramatic as what we saw in the one. When you average over the lamb according to the inputs you definitely see stimulus specific adaptation so the inputs from balanced of the one are still suppressed when the stimulus is repetitive but you no longer have deviance detection in these inputs so the red line is the average response of the devious stimulus right on top of the kind of neutral control.

[00:42:12]
In the slammer cortical inputs if you just kind of compare this back to back this is what was going on in the V one neurons you had amplified input amplified responses to deviant stimulus and suppress the redundant you only had this redundant suppression point McCorkle inputs. So no devious attention at the population level when you average over inputs to the one.

[00:42:34]
And then if you do the same sort of K. means clustering analysis this was the result from your view one neurons your your. Dilemma cortical inputs you only kind of see 2 clusters there using the same approach in those 2 clusters neither one of them is doing deviance detection as well so you have this kind of smaller subset which doesn't.

[00:42:58]
Guard the context which were seen everywhere but the rest of them about 2 thirds were just your stimulus specific out of adapting so neither one of these subs as was doing the direction this tells us some aspect of this is a devious direction isn't kind of very simply and directly inherited from the film this is the least computed after the film is somewhere in visual cortex or beyond.

[00:43:27]
Yes Yeah yeah. Yeah yeah right so well well actually there there there is some more interesting tuning in the cells. It's not a strong yet so so right. At the very least we don't see devious detection there and but I mean it's. Just at least tells us is not simply inherited from there but you that's right in the mouse it's it's a little more than higher mammals for sure but but you're.

[00:44:04]
So there may be some other stimulate that would be nice to test and see if they're also showing it there also showing that's right but I did see some orientation tuning among these inputs was that yeah different wasn't as robust. So. So you know just to summarize that we see not only devious detection and adaptation in Layer 23 but these nice kind of segregated sub ensembles.

[00:44:32]
Prefer in different contexts Whereas in your bottom up and from this we have neither devious detection or deviant preferred ensembles. So that's the bottom up story what about your top down inputs from from P.F.C.. Again we use the same approach so you express G camp only in these cells and then you focus your recordings on Layer one of visual cortex and these are again sort of accidental segments.

[00:45:05]
From P. of SR ons in visual cortex. So doing the same analysis here each row here is an individual or a single it. Was just sent back to the one using oriented stimulator again and you see the you see definitely clear visually evoked responses in these top down imports.

[00:45:29]
And you see what appears to be at least in a subset of them simulate specific adaptation so these top down inputs are. Responses to are done stimuli but again they're not on average doing deviance detection it doesn't appear so so that was somewhat surprising as we suppose that perhaps this deviant signals computed later on and sent back to the one especially given how late it occurs but it doesn't appear that at least when you average over all these points of view one that they're doing devious detection.

[00:46:03]
And there's just again directly comparing the P.F.C. responses to the one you don't have the Amplified response to deviant stimulate Now when we did the K. means analysis the story gets maybe a little bit more complex so again you have these non modulator inputs from P.F.C. to be one but you also have this subset which is doing kind of a by your actual modulation So it's almost as though these are these cells are kind of relaying back exactly what the average of you want responses back to the one they're doing both deviance detection a little bit and some real specific adaptation so there is this cluster and P.F.C. which leads us to believe it could be at least to some degree contributing to what's going on in the one during this paradigm.

[00:46:54]
So that's where the stories right now. We know that sort of matter stand intern runs play a role we know that this is not this is sort of different contexts are preferred by different subsets of cells. Deviance detectors are more common in Layer 23. And this isn't directly inherited from.

[00:47:17]
Us. But. The kind of the remaining question that's. One which is mainly unanswered by just observing the activity is sort of the top down P.F.C. inputs player calls will roll either in. Your your average responses to different contexts and also your your contacts prefer in ensembles. So in order to test this using our genetic approach again this is recording neurons in V one with calcium imaging but arch T. was expressed in P.F.C. neurons and then recorded.

[00:47:58]
Paradigm in V one neurons and then had another run where we suppressed the inputs from P.F.C. to V one using up the genetics and every other trial so. This is this is what looks like a baseline of showing a lot of plots like this this is again you have your increased responses to deviant stimuli these are V. one neurons and suppress responses to redundant stimulus as expected when we suppress P. of C. inputs to V one.

[00:48:30]
We lose deviance detection so there's no longer amplified responses to a deviant stimulus you still have stimulus Pacific adaptation if anything it appears a little stronger but your sponsors in the neutral case are essentially equal to your responses and deviant kids almost as though now all stimulate if they're rare they're unexpected.

[00:48:51]
Or treated as deviant and this also makes your view one neurons look a lot like the average input from Dallas which is interesting. In this could be due to 2 different effects you have suppressed responses now when you take C.E.O.. Response deviant stimuli your smaller but also responses that control or neutral stimulate your larger and this is the same data just with the bar plots here.

[00:49:18]
You have devious detection and adaptation whereas. When you suppress P.F.C. you only have your attention but you can no longer have deviance detection. So that's when you average over all neurons. But suppressing P.S.C. didn't really affect kind of ensemble number ship which was interesting to us so you still had the same relative proportions of deviance detectors in the same relative proportion of adapters before and after taking P.F.C. inputs and out of the picture.

[00:49:51]
This is just kind of the same data but you can just kind of directly see this is before suppressing P.F.C. intern or P.F.C. inputs to view one you have your deviance detectors in your doctors and your non modulator This is after suppressing P.F.C. you still have the cells still retain their same i DEN Are these your kind of contacts preferring sub ensembles and the one still prefer their preferred context the main thing that's changed is the relative response magnitudes So this is before and after suppressing see Davey's detectors seem to decrease their responses whereas your generalized adapting cells tend to increase their responses so kind of flipping this emagine P.F.C. seems to be suppressing these cells whereas it it's sort of.

[00:50:38]
Exciting these cells either directly or somehow indirectly via disinhibition And so that needs to be worked out as well. You know. How can you go from kind of. Sub cluster in a non modular cluster to these nice kind of. Relatively complex effects on the subclass Toure's here in the sort of a cluster wasn't really affected much it was slightly decreased after you took P.B. out of the picture.

[00:51:08]
So this is you worked out kind of and I think a nice way to do this would be. First by model modeling and getting an idea of what could be going on we know a lot of Bella Hussam out of stone or on target also subsets of some medicine neurons who they target and where and when.

[00:51:26]
I haven't looked at or explored V.A.P. into. Neurons but they could be an interesting candidate as well as T.V.'s The 1st sort of workout how how this circuit could underlie this complex effect and then sort of. Stimulate P.F.C. input swallow. Using drugs to suppress different subtypes and see how seeing how that effects a lot of experiments the kind of explore this model.

[00:51:52]
And get a deeper understanding it also would be important to kind of measure. Time frequency responses during this paradigm as well to link not only what's going on in this kind of single cell and multi-cell level but also. Match it up to something we can measure in patients.

[00:52:09]
Sort of oscillatory response to these stimuli the kind of get a nice full picture of how specific. Things affecting individual intern runs might give rise to what you can observe with a noninvasive E.G. So that's kind of the bigger picture here but instead of focusing on A and a mouse model of a very complex disease we can sort of focus on a particular domain of dysfunction which we can translate them eyes and start to kind of link up our individual cell types and circuits and perhaps sort of neural modulators give rise in module A This response to get a bigger picture of that function rather than a complex disease so that's that's everything so.

[00:52:53]
I think my sort of my postdoc advisor Rafael used to. Give was a Ph D. candidate who worked with me on a lot of this especially the auditory stuff in the form of cortical inputs and also sort of way gin and shooting Han they did. The technical and actual experiments in the multi-plane imaging paradigm so so thanks everybody.

[00:53:47]
Those years those for some other reason. As a good catch I should update that. There was one way I originally had conceptualize this that some medicine arounds might disinherit the circuit both by suppressing Peavey's and V.I.P.'s we know Sam's and V.I.P.'s have this kind of mutually into going to stick relationship in so so that was when we started think about it just to simplify the picture.

[00:54:12]
But but you know in this case actually and no line I'm aware of are they targeted the Peaveys And these are targeted not Sam's it's. A behavioral changes due to the smell of stone suppression. It's a question I had. They didn't change sort of locomotion. And it was more or less the only thing we quantified there wasn't any sort of very obvious changes in their behavior and again it's.

[00:55:02]
Very selectively suppressing some out of stone or neurons in visual cortex. And so it's you know it is possible that they might behave really respond to visual stimuli different but their overall behavior was unchanged and is also true that some some of Scieno is converted to close of pain which is an anti-psychotic the thought being that it's not enough to to affect their behavior and if you do the right controls you can really kind of what are the effects of Scieno cause of being on some inner ons versus the generalized Thanks.

[00:55:42]
Nick. Yeah. Yeah that's that's that's a good question I mean. I don't know I would it would be nice to see sort of on a downstream target whether devious detectors and your more general sells synods on the same cells or do they actually continue to perform different subnet works.

[00:56:31]
I'm not sure so the. So the so this particular paradigm has in the past then been used to say you know patients showed destructive mismatch negativity and this is perhaps due to P.F.C. I don't know. When you're looking at the aggregate Yes I mean if you just sort of average the overall view on output it does look like B. of C. is doing that but there's still some computation that's that's ingrained in the circuit the circuit is still able is still the technique deviant versus not.

[00:57:04]
So so I think there could be many ways to lead to abnormal contacts processing at the behavioral level. But yeah I would it would it would be good to know that I was suspected probably do target the same neurons and there is kind of a waiting that happens but I don't.

[00:57:27]
See it in. Using. It which may using transgenic animals. Other animals rats and and primates. Well so the so. Has definitely been studied using kind of E.G. and rats a lot. But it hasn't been done. You know I haven't looked at specific cell types or circuit manipulations at all usually it's done in a very clinical setting sort of not clinical but very much will just record one trace and we'll just compare compare There are done into the D.V. and see if there's a difference so I think there's a lot of really good work that needs to be done in other animals as tools are developed even as basic as using kind of more thorough experimental paradigms or a or as measuring different layer layers of cortex.

[00:58:30]
And in one human primates has this kind of been done in the right death I think so. Yeah. Yeah. Yes So the that's a good question so I did your non modulator cluster is they tend to not be orientation to neutral so they're they're more or less responding to or so that they're neither context tuned NORRINGTON tuned but both of these subsets do show orientation tuning.

[00:59:16]
The one issues it nor to assess orientation tuning in this paradigm you need to do many many paradigms because you need to have a paradigm where or the stimulus is the D.V. and then where this stimulus is the D.V. and where this one is but as far as I can tell there are still showing about the same orientation tuning as your generalize adapters so not only are there deviant devious detectors but there are devious detectors for a specific stimulus.

[00:59:43]
Which is even more interesting is that sort of nested sudden networks there of different preferences.