NIMH Multimodal Brain Stimulation Speaker Series: Bradley Postle, PhD
Great. Well, thank, Tom and Holly, and everybody's who has given me a gracious welcome, and thanks for arranging this idyllic low-humidity weather. It's been fabulous. For those of you who aren't in the room, we're going back to the days when a WebEx complication. I'm showing slides live, and then Tom has my PowerPoint file on his computer, and he's going to be mimicking what I do in terms of changing slides. So, if there's a lag on the order of half a second, or a second or so, it's not technical as much as it has to do with interpersonal communication.
So, with that, I'm going to jump in and just very quickly introduce the setup that we use. So here, you can see a staged photograph of a subject in the lab. Most of the studies that I'm going to be showing involve simultaneous EMS and EEG recording, and we have a system that was developed by the Nexstim Company in Helsinki, and this is technology that was kind of pioneered and developed by Risto Ilmoniemi, who you'll be hearing from I think later in this series, sometime during the autumn, a talk that I'm very much looking forward to as well.
So, we have a Nexstim eXemia TMS compatible EEG system, and effectively what that means in laymen's terms is that there's a sample and hold circuit built into the EEG amplifier, such that 300 microseconds prior to the discharge of the TMS capacitor, the amplifier goes into the sample and hold circuit and stays quiet until a millisecond after the discharge of the CMS amplifier. So, the idea is that you have this silent period, during which the TMS discharges that prevents the amplifier, the EEG amplifier, from saturating.
In a lot of cases, we're using Nexstim, sort of conventional Nexstim TMS equipment that's in wide use around the planet, and the Nexstim -- excuse me, the Magstim system. I might have just misspoken. We use Magstim quite a bit for our TMS. The Magstim system and the Nexstim system play quite nicely together, and that's worked well for us. There's also a Nexstim TMS stimulator that we use in some cases. And, so, in a standard experiment, then, what's happening is that we are, either with a human or with an arm that has multiple degrees of freedom, we position the coil.
You can kind of see on this slide, in the background here, is a 3D reconstruction of the subject's brain. So, we've done at least a T1 structural brain scan, MRI scan, sometimes superimpose the activation hotspots from an fMRI scan, depending on the application. There's a 3D reconstruction of that, and the experimenters feed the 3D reconstruction, which has been co-registered with the subject's head. So, the subject has these infrared fiducials on a nose piece that's being picked up by the Polaris infrared frameless Stereotaxy system.
So, what that means -- and there are also infrared fiducials here on the TMS coil. And, so, what that means is that, in real time, as the experimenter is moving the TMS coil around the skull, what he sees is an animation of the TMS coil moving over the 3D reconstructed brain with a simulated beam of light, as it were, that gives a real-time visual depiction of the E field that's produced, that's estimated to be produced by the TMS. So, we can target, then, structures, and the assumption, of course, is that the gross anatomy of the subject hasn't changed from when we scan them and when we're doing the TMS experiment. We feel pretty confident about that. And, yeah, so that's enough of the background, I think.
Just, quickly, to give you a feel for the kinds of signals that we see, the movie is showing you a series of TMS impulses. Each time the wave form jitters, there's another pulse that's being contributed to this running average. And, so, what this is intended to give you the sense of is the TMS-evoked response looks physiological; right? It's on the range of four to six microamps. The impulse to impulse replicability of the wave form is very high, so it's a signal that we can work with. Now, this is after we've cleaned the signal for any artifacts.
We typically do one, if not two, rounds of independent components analysis cleaning, so we remove anything that's not -- we try to move everything that's not neurogenic from the signal. But the evoked response is one that we feel like looks physiological and is something we can work with in ways that I'll be describing here throughout the talk.
So, when my lab got started, we were using equipment that belonged to the Laboratory of Giulio Tononi, who is up here, a colleague of mine at the University of Wisconsin, and with Giulio, and in particular Marcello Massimini, who is now in Milano with his own lab, and Fabio Ferrarelli, we started in my group inspired by, among other thing, by a study that Massimini and colleagues had done to look at the physiological underpinnings of the behavioral state as subjects transitioned from quiet wakefulness to sleep. And this came out "Science" in 2005, and this was around the time that my group got involved in TMS, and I won't go into a lot of detail here.
But, in effect, what they saw was they had subjects lying in what amounted to a dentist chair who had been set up with a EEG, and they just waited until the subjects fell asleep, and they periodically administered pulses in TMS. And what have shown here, just to give you a gist, this is a butterfly plot of multiple stimulation events, but point is that over the course of the propagation of the TMS pulse from 10 seconds post-stimulation, all the way out to 280 seconds, which is what you see here, the reconstructed activity in the brain evolve markedly, such that the region of activation gets larger, it moves more anterior. It moves along the midline. It bifurcates, such that you have a hot spot in the contralateral hemisphere and the frontal cortex, as well as back hear in the parietal cortex, all the way in the contralateral hemisphere, back to under locus of the original -- the point of stimulation, which is the area that's highlighted here in at ten milliseconds.
So, the point is that you have in complex propagation of a signal that bounces back and forth quite a bit over the course of these 300 milliseconds. Whereas when the subject is asleep, there are two things that are noteworthy; one is that the magnitude of the initial TMS-evoked response is markedly higher than is the magnitude of the TMS-evoked response in the wakeful brain. So, this is early stages of non-rem sleep shortly after the subject fell asleep.
The other thing, three things I guess, one is that the initial response is higher in magnitude. The second is that the reconstruction of the activity remains quite localized. So, you still see the spread in anatomical space of the signal, so much less anatomical propagation, and finally, at around 120 milliseconds, 130 milliseconds, the response dies out. So, it's also not extended in time in the sleeping brain in the way it in the wakeful brain. And the Tononi group has a nice account of ways in which the state changes between the waking state and the sleeping state entail, among other things, a dramatic reduction in large-scale conductivity, and this is an empirical illustration of the fact that long-range connections between many neural systems believed to be functionally shut down.
So, really, the idea here was just to give you a sense of the context into which we stepped when we started doing this work. And so, initially I'm going to show you some studies from this group. Jeff Johnson and Steve Emrich were post-docs at the time, Bornali Kundu, an MD/PhD student in the lab.
And in my group, we're interested in visual short-term memory and visual attention, visual awareness, and so the initial bit of data that I'm going to show you from my group is going to look at this phenomenon of the systematically differential propagation of the TMS call as a function of state, but now what we're varying is the cognitive state of the individual.
So, what I'm going to show you is how the TMS pulse propagates when the subject is passively fixating versus when the subject is fixating in the middle of a working memory task. So, in the task context, the subject has seen four stimuli, and at the end of the trial will, depending on the construction, either have to make a recognition judgment as to whether this location on the screen matches one of the four locations that had been highlighted, or if it were an object trial, the subject would see a complicated shape and would have to make a yes/no discrimination as to whether the shape that's presented as a memory probe matches one of the [inaudible].
But the point is, we're contrasting the effects of delivering a pulse of TMS when the visual conditions are identical, but the difference is you're either in the middle of a memory trial or you're in a fixation block when we aren't controlling what you're thinking about. And pulses of TMS were delivered in a pseudorandomized basis during the play.
And what I'm show you here are three different synthetic measures that were extracted from the EEG signal. So, these were measures that were developed by Adenauer Casali, who is a physicist. He's in Brazil now. He was working with the Massimini group in Milano when he developed these methods. And for the purposes of this talk, I'll just say that these are rectified signals that you can think of the significant current density, for example, you can think of as a more sophisticated version of a global mean field [inaudible]. So, essentially, it's one number that gives you a sense of how much energy was produced by this pulse. And, so, during task, there there's double the magnitude of energy, as indexed by this significant current density measure than there was during rest. And you can see how that played out over time so that the initial response of this to the TMS is markedly higher during task than during rest. Remember, this is a behaviorally identical situation, and so the cognitive context is different.
The other measures that we have are indices of the current scatter, which is also higher during tasks than during rest, which is to say that areas distal to the TMS stimulation are more impacted by the impulse, and we can look at that more specifically. The green arrows are pointing to where the TMS coil was located. And what we're going to do is just look at this measure of significant current density, which, again, you can think of as a rectified index of the energy of the TMS-evoked response, both at regions of the brain that are local to where the stimulation was delivered, as well as in -- so the stimulation was delivered to area seven. It's relatively medial portion of the pera parietal lobule, and we'll compare that to area six in the premotor cortex, where any signal that we pick up has to have been synaptically transmitted, physically far enough away that none of the magnetic energy from the TMS will affect this area.
So, that's what I'm showing you here. The red wave forms are from broad [inaudible] area seven, so in the parietal area proximal to the TMS, and the blue wave forms are from area six, the frontal cortex. And, so, initially, I guess if we look at the solid lines, which are the lay period of the short-term memory tasks, versus the dotted lines, which are the fixation on a cross control condition, you can see that the amplitude of the TMS-evoked response is markedly higher immediately after the impulse in the task condition. So, in a way, this is the opposite of what I described between quiet wakefulness and sleep; right?
In effect, this same [inaudible] cortex is more excitable when the subjects engage in a task than when the subject is overtly engaged in the same behavior but in the context of passive fixation. Subsequently, over the long term you can see that, in general, there's just more energy locally in the TMS-evoked response in parietal cortex, the response of asymptote around 300 milliseconds.
Now, in area six, in contrast, there's an initial sort of ripple associated with the discharge of the TMS apparatus. But it's really tens of milliseconds later that you see the biggest excursion in the amplitude of the response. And, remember, this is at least one, if not multiple, synapses away. It includes thalamic relays as well. But there's some complex filtering that's happening. But you can see that the response in frontal cortex -- the biggest response in frontal cortex is markedly delayed relative to what we saw in parietal cortex.
Another thing that's interesting is that between task and rest you don't really start to see these differences until several tens of milliseconds after the impulse. And, so, presumably it's in the recurrent interaction between these two areas that I'm showing you, as well as other regions, including, importantly, subcortical region, where we see real differences in the energy of the TMS-evoked response. And, so, this is one way of quantifying, and we'll look at other ways we've tried to quantify, in effect, the changes in effective connectivity that are revealed by delivering an exogenous source of energy as a function of different states, whether it be cognitive states, behavioral states, other physiological states that play out over shorter periods of time.
So, here, what I'm going to look at is differences in the TMS-evoked response as a function of where in the cycle of neuronal oscillation you deliver the impulse. So now, really, we're talking about what's the difference if we stimulate ten milliseconds earlier or ten milliseconds later. And in order to do this, this is where Bornali Kundu, she did a simple -- she looked at the global mean field amplitude of the EEG response to TMS and did a median split between the high amplitude response trials and the low amplitude response trials, in red and in blue, and then looked at the analyzed the data to predict what factors in the state of the brain prior to the delivery of TMS predicted whether a trial would be in this high GMFA category or in this low GMFA category.
So, here, what I'm showing you is the intertrial phase coherence. This isn't power. This is the trial-to-trial coherence of oscillatory activity at varying frequencies, going from low frequencies at the bottom the vertical axis here, up to 50 hertz at the top. And outlined in red in the -- excuse me, outlined in these black squares are portions of the gamma band that fire to the delivery of the TMS pulse predict the TMS-evoked response. The reason that they're in dotted lines is that these were areas that were significant before Bornali applied direction for multiple comparison.
So, after more rigorous correction, what she found was that this narrower range in the beta band, right around 20 hertz, at about 150 milliseconds prior to the TMS pulse, that the phase coherent of the gamma -- excuse me of the beta oscillation predicted whether the response would be of higher or of lower amplitude. And to look at that a little more precisely, she looked at -- and it turns out that when we're stimulating parietal cortex, not only is it the pre-stimulation phase angle in the beta band that seems to be most important, but the most important determinant of the amplitude in the global mean field amplitude, after the pulse, is also carried in the beta bin.
So, this is actually, when we're targeting parietal cortex, the amplitude of the ringing in beta accounts for most of the difference in energy that you see in overall TMS-evoked response. And, so, what I'm showing you here is -- these are the different phase bins into which Bornali divided the beta cycle. But the summary slide down here shows you that, in particular, there are phase angles that are close -- that are just prior to and just after the trough of the beta oscillation, at which the TMS-evoked response will be maxed. So, at these phase angles of beta are when you would time your TMS to get the biggest evoked when you're stimulating parietal.
And one question so that we -- so just as an aside, so this is a comparable analysis, looking at power. So, what this indicates is, in this particular experiment, there was no evidence that pre-stimulation variations in power at the location that Bornali was stimulating had any influence on the amplitude of the TMS-evoked response. So, the predictive power is knowing where you are in the instantaneous physiological state in terms of the oscillation, the phase angle of the oscillation, rather than the ongoing power.
One question that is going to be sort of underlying a lot of our thinking, in that I'll just call it to your attention now, is captured in this figure. This is a paper from Rosanova, el al. This is also Marcello Massimini's group. And here, in effect, what they did was they delivered single pulses of TMS to different areas of the scalp, and what they found was, when they targeted area 19, the dominant spectral evoked response was at around 11 hertz, and when they targeted area seven, the dominant spectral response was at 20 hertz, and when they targeted area six, the dominant evoked spectral response was at 31 hertz. And this remained stable when they varied the amount of energy that they were delivering, and the stimulation.
And, so, the argument that they made is that these different corticothalamic circuits may, in effect, have different natural resonating frequencies, so that when you ping occipital cortex, it will naturally resonate in the alpha range. When you ping the parietal cortex, it may naturally oscillate primarily in the beta range, and so on. And, so, this may have important implications; for example, if we think about different disease states like schizophrenia, where gamma abnormalities in the gamma band have been observed, to what extent is that a byproduct of the fact that there is dysfunction in the circuit that involves the frontal thalamic system; right, as opposed to something about gamma, per se; right? In effect, the oscillation may be a byproduct that just arises from the architecture and the physiology of that system, as opposed to the oscillation having some function that's independent of the brain system [inaudible].
So, one question that we wanted to ask was what might be the functional significance of these effects that we've been seeing in the beta band when we stimulate the parietal cortex. So, this is now a study that Jason Samaha published recently, and he's taken advantage of prior work that's shown that one can produce a visual phosphine evoked by TMS so that when you stimulate, for example, in the upper -- in a superior portion of occipital cortex in the right hemisphere, you'll produce the percept of a flash of light that's perceived in the lower quadrant of the left visual field. And it's been shown previously that not only with occipital stimulation but also with parietal stimulation one can produce this phenomenology of visually evoked -- of CMS evoked phosphine.
So, here's Jason, and what he did was determined the 50% detection threshold for each subject before the experiment started. And the subjects then experience a series of pulses at this 50% threshold, both in extrastiate cortex, as well as at this area of the intraparietal sulcus, highlighted here. And after each pulse he asked the subjects to indicate on this sliding bar how confident are you that you saw a phosphine. So, we have a continuous measure that we discretize between zero and 100, that quantitatively, for each of our subjects, receding here are each of the ten subjects, and you can see the distribution for each subject in terms of the number of trials in which they reported seeing nothing, the number of trials in which they reported that they were sure they saw phosphine, and in between.
And in his analysis, then, instead of just doing a median split or some other kind of courser average, what Jason did was a trial-by-trial regression, where he looked at what elements of the EEG signal predicted on this 1-to-100 scale the reported visual phenomenology from the subjects.
And, so, what you're looking at here is the beta coefficients, then, that came from this regression, and on the top is a time by frequency map of beta coefficients that are associated with the occipital TMS, and on the bottom is the parietal TMS. And, so, what you can see is, if you look to the right of zero, times zero when we delivered TMS, the response in the EEG that is associated with receiving a phosphine is identical in the two stimulation conditions. There's this increase, this positive association with power at very low frequencies in the theta, and maybe the low alpha range, and then this large decrease in power that you see in alpha reaching up to around 20 hertz that you see both with occipital and parietal here. So, it's as though -- at a simplistic level, it's as though the percept of a phosphine doesn't care where the source of the energy was that produced it; right; that the neural correlates of perceiving a phosphine are the same regardless of where you stimulate it.
However, if you look prior to stimulation at what is the oscillatory state that's fixed, what the subjects report will be, what you see is that in occipital cortex it's really concentrated in the alpha band; that's negative relation between alpha band power and subsequent phosphine report. Whereas in parietal cortex, it's centered, really, here in the low beta band and sort of between the low 20s and the high teens, where the oscillatory state about a half a second before the delivery of TMS predicts what the functional effect of the TMS is going to be.
So, this is nice, I think, because it links function to what we've seen previously with the physiology in that there's something functionally important about beta band oscillations in the parietal cortex, versus lower frequency oscillation.
So, some questions that we need to ask, I made this point about how the pose-stimulation time frequency analysis looks very similar. But, of course, we stimulated different areas, and so one question that we want to ask is about the direction in which the TMS impulse profligate after parietal cortex; right? In a very simplistic way, you could imagine that the parietal cortex has to feed back and be gated through earlier visual cortex, and then the signal runs through the same route to achieve a conscious percept; right?
Another possibility is that there are local and/or read forward phenomena that are primarily responsible. And, so, we are in the process now of applying an analysis that we did a few years ago to a different kind of dataset, and I'm just flagging for you this approach that's been developed by Barry Van Veen and a student of his. So, Barry developed the methods for source localization back in the '90s, the Veen former method, for example, came out of Barry Van Veen's lab. But that's a solution of the inverse problem, and he doesn't actually sort of believe the results of that anymore. And, so, he prefers to use this model testing approach, where he developed a multivariant regressive model that allows him to estimate directed conditional granger causality, such that you can test ideas -- this is during a working memory task, what I'm showing you here -- where a change in the load was associated with an increase in alpha band power, the directed causal flow of energy was from frontal to parietal cortex.
When we added interference, what we saw was an increase in beta band power that seemed to flow directionally from frontal and parietal areas to extrastiated areas. And Van Veen, and his student, Chang, have extended this method, such that they can apply this multi-variant auto-regressive model to a known exogenous source of stimulation. So, this is what we're going to apply now to these Samaha visually TMS-evoked phosphine data to try to get a handle on what might be the temporal and putative causal chain of events that leads to phosphine production.
So, what I've been talking about up until now is variability in the physiological state of an individual that will predict different physiological and behavioral outcomes. We've also seen evidence for considerable variability across subjects. And, so, these are 11 individuals, and I'm just showing you the TMS evoked response during task and during rest, not unlike what I showed you earlier when I was showing you the significant current density facts and parietal and premotor cortex.
And, at least to the naked eye, what this suggests to me is that there's much more similarity within a subject in terms of what her TMS-evoked response might look like, even across cognitive states than when we look across individuals who are within the same cognitive state. And this is also evidence when you look at the time frequency representation. So, these are four different subjects who were brought back between three days apart and five months apart, and I'm just showing you the so-called event-related spectral perturbation from one electrode from each of these subjects. And you can see again, at least informally, and we've done this quantitatively with permutation testing, that the test/pretest reliability of the spectral fingerprint, if you will, of an individual is quite stable. And, so, then a question becomes, well, are there functional consequence s to the fact that different people seem to have different oscillatory profiles. And the answer is, yes, at least when we assay that with TMS.
And, so, what I'm showing you here are the effects of the distal TMS-evoked response. We're still stimulated this parietal area in superior area seven. And what you can see is that -- and, again, the difference between the black and the gray is whether the subject is at rest or whether the subject is engaged in a task. So, it's the same data, actually, that I showed you earlier, where subjects are fixating in both cases, but in one case it's the delay period of a working memory of a working task. In the other case, it's fixation.
And what you can see is that the raw power in the theta band in parietal cortex predicts what will be the TMS-evoked power in the theta band in the superior frontal cortex; that when we look at area nine, a little more laterally and prefrontal cortex, individual differences in the power in the time frequency representation that we see in the alpha band or in the theta band, and this is in parietal cortex, predicts what TMS evoked response will be in the theta band, in both cases, in this more lateral prefrontal area. And we see differences in the distal TMS-evoked response in occipital cortex as well. So, marked differences, both in the beta band and in the theta band. These are the beta and theta in parietal cortex, which is where we're delivering the stimulation, that will predict, then, the individual differences in the TMS-evoked response back in occipital cortex.
So, this is interesting to us, and we feel like there's a rich number of possibilities that could be pursued, that if one were smart enough in knowing which dimensions of individual variability of anatomical physiological individual variability in your subjects were important before they came into the experimental room or into the clinic, that one would want to tailor the stimulation protocol differently, as a function of, for example, different resting oscillatory traits in the area that you're going to be targeting.
So, moving thematically then, from looking at individual differences in physiological traits to how we can use this to probe pathological states, I'm going to harken back again to work by my colleagues in the Tononi lab. This is Fabio Ferrarelli who, initially, in a sleep study described reduced sleep spindle activity in schizophrenic patients relative to controls and relative to a comparison group with major depression, and so here in the power spectrum during sleep, I'll just high like that in the spindle frequency band here, between 1 and 15 hertz, is markedly lower power in the schizophrenic patients versus the healthy controls and the depressed individuals. And where, on the scalp, they found these differences was this central area that's highlighted down here in the bottom panel.
And, so, with the TMS/EEG approach, what Ferrarelli and colleagues did was delivered single pulses of TMS to this area that had been identified in the sleep study. And now these are awake resting individuals, patients and control subjects, and the findings, here's a comparison between the patients who are in red and these are the neurologically psychiatrically healthy controls in blue, is that there are differences in just the raw evoked response that you can see here, and that translated in a spectral transform to evidence for a markedly decreased TMS invoked response in the gamma band in the patients relative to the healthy group and relative to the comparison depressed group as well. You see that both in the power in the gamma band, as well as in the intertrial coherence in the gamma band.
And, so, with that background, we followed up with a more targeted hypothesis about what might be the source of this difference between the patients and the healthy individuals. And we used the logic that the reticular thalamic nucleus, which is this thin sheathe of GABAergic neurons that imposes an inhibitory clamp on the thalamus, the totality of the output of the thalamic reticular nucleus is focused on this element. But, most importantly, the TRN has been known for a long time to be the pacemaker that generates sleep spindles. So, a number of very elegant studies have shown that if you completely deafferent or deefferent the structure, spindles are preserved.
And, so, the hypothesis, then, was that there may be some dysfunction in the thalamic reticular nucleus in schizophrenia that leads to this sleep spindle abnormality. And, so, we tested that with TMS and fMRI, and we had to go to fMRI because we can't get at the physiology of the thalamus in any meaningful way with EEG.
So TMS fMRI, here is our setup. It's a conventional figure-eight TMS coil in which all of the ferromagnetic elements have been stripped out, and it's a shielded cable, this is shielded. The TMS stimulator sits outside in the control room, and we run this cable through the penetrator panel into the Faraday cage. It's run through the back of the bore of the scanner, and we had, in our machine shop, developed this PVC arm that has different blocks of heavy material where we need it and has many degrees of freedom so that we can move the TMS coil in as many different dimensions we need to, within reason, to be able to fit it inside the bird cage head coil. This is mounted on the so-called doghouse of the scanner with a plate of PVC that's underneath the doghouse that's screwed into the housing of the scanner. So, essentially, our TMS coil is bolted onto the housing, and the scanner is how we hold it stable.
Eva Feredoes was the post-doc who first did these studies, and so this is kind of what it would look like to the subject before the subject slides in. And the point of this slide is just to highlight that the we use a button press behavior as a behavioral comparison for TMS. And so just to give you a feel for what the TMS-evoked response looks like, this is a statistically thresholded, but unmasked statistical mass button press related activity, and what I'm going to do is mask it so now you just see the button pressed evoked response. This is the subject who is pressing a button with their right thumb once every 20 seconds. So, this is a first scan.
In the second scan, they do the task again. And some of these boxes have been painted blue. It's kind of hard to see. In the third scan, over the left hemisphere, the hand area of primary motor cortex, we delivered single pulses of TMS, and the single pulses of TMS produced elevated activity in the primary motor cortex, same way that the button presses did. And what I will just show you now is the hemodynamic-invoked response trial average to the time of the button press for the first scan and the second scan of the button press, and then here is a comparable number of TMS pulses delivered to scan one. And, so, the TMS-evoked response within the scanner looks comparable to what we see from a simple behavior, so we're going to take advantage of this now to look at the TMS-evoked response in the thalamus.
So, Ellen Guller to the lead on this study, with a lot of help from Alex Shackman, as well as Fabio Ferrarelli. So, now I'm showing you a comparison between schizophrenic patients and healthy control subjects in which, initially, we're looking at the response evoked by a right thumb button press. So, what you can see is that the response is quantitatively higher and somewhat lagged out in time in the patients relative to the control subjects. These are not reliable differences. These are the effects in cortex. Here is the effect in cortex of the pulse in TMS.
So, the one thing that this illustrates is using an exogenous source of energy, an exogenous source of energy, an exogenous perturbation, over which you have control, can, especially for patient studies, give you more experimental control, in the sense that the brains of the patients of the control subjects received identical stimulation on each trial. So, there's a lot less variability in the TMS-evoked response than what we saw in the button press invoked response.
Now when we look in the thalamus we see this marked reduction in the amplitude of the fMRI response to a button press in schizophrenia relative to control subjects. But this would be hard to interpret, for the reasons that I just showed you, that there's already difference, in fact, going in the other way; right, at cortex. So, how do we interpret these differences in thalamus? The argument that I would make is it's easier to determine the differences when you look at the TMS-evoked response for which the proximal cortical-invoked response is very stable across the two diagnostic groups.
But, now, in the thalamus, we have this marked difference, both in the amplitude, as well as in the timing of the TMS-evoked response. And so, remember, the hypothesis is that there's dysfunction in the thalamic reticular nucleus of in schizophrenia. We can't hope with our resolution, with fMRI, to isolate the TRN. But because we know the totality of the TRN's output is focused on the thalamus, the logic was that the thalamic-evoked response would be a reporter to us of the integrity of that system. So, this is consistent with the idea that the thalamic complex is implicated in the pathophysiology of Parkinson's.
Another way in which for potential diagnostic applications, this approach is promising to us is that when we look at the cross the sample of control subjects and patients in term of the amplitude of the evoked responded, at the group level the patients are lower than the controls, but you can see that the specificity isn't what you would need for a diagnostic tool. In our hands, with 12 subjects in each group, the lowest TMS-evoked response of a thalamus of a psychiatrically healthy individual was higher than was the highest TMS-evoked response in the thalamus of a schizophrenic patient. So, the implications for diagnosis, I think, are pretty clear.
Another thing that we saw was that areas in cortex that were distal to the site of simulation also looked different. So, in this midline superior frontal area, as well as in the insula bilaterally, we also saw lower TMS-evoked responses in schizophrenia relative to the control subject. And when we looked at how we could understand this effect, which would have to be a polysynaptic effect, because you're stimulating M1, we reasoned that if we did a connectivity analysis in which we correlate the time series between two areas, that could give us some insight into the differences in the responses in these two groups. And one of the nice things inferentially is that we can call this an effective connectivity analysis rather than a functional connectivity analysis, because we are introducing a known sort of perturbation whose effects on these different regions of the brain that we are going to evaluate. So, there's a causal element in this analysis that you don't get from a traditional functional connectivity.
So, a couple of things I'll point out. One is that if we look at the connectivity of the TMS-evoked response, we can't explain it in terms of a difference in the controls in the patients when we look at precentral gyrus to superior frontal gyrus connectivity or when we look at precentral gyrus to insula. When we include, however, the thalamic-evoked response, we see that by running through the thalamus we can now explain a significant amount of the variance of the difference in the insula-evoked response and in the superior frontal gyrus-evoked respond by taking into account the differential response in the thalamus.
So, the suggestion, then, is that this may have implications or ideas about whether schizophrenia has been described as a disconnection syndrome, but whether that disconnection is corticocortical or subcortical is an important question. This is one way that we can start to get at those kinds of questions.
Empirically, just as a follow up to make this point about effective versus functional connectivity, when Ellen did the same analysis, looking at portions of the data in which there was no TMS, she did, in effect, a resting state connectivity analysis, she was unable to find evidence for disorder connectivity between any regions. So, the highlights the fact that it has been at least implied through everything I've showed you so far, that one of the advantages of using TMS and these kinds of studies is simply that you get a better SNR, because you're introducing more energy into the system, and in many cases, that just increases your sensitivity to pick up differences that may be too subtle to pick up with more traditional physiological.
Okay, in the last bit of time that I have now, I'm going to move on to looking at the effects of repetitive TMS. So, we're going to get into this question of virtual lesioning and training or otherwise influencing physiological processes [inaudible]. And the first thing I'll show you is the results of a paper that just recently in press from Regina Lapate, who did this work in my lab, and she was interested -- she comes from Richie Davidson's group where she has been studying -- had been studying control of emotional responses with stimuli.
And earlier work that they had done with this affective coloring task showed the following phenomenon: That if you subliminally show either happy or a fearful face, that's typically backward math that otherwise made not consciously accessible to the subject, and then showed an overtly perceivable neutral face and asked the subject to make a judgment, how much do you like this person, it's a well replicated finding that the subject's evaluation of the neutral individual is biased by what the affective state had been of the prime -- the subliminal prime that had been presented just prior.
And in fMRI work that Hegina had done prior to this, what she had found was that lateral frontal cortex connectivity with the amygdala was inversely related to the susceptibility to make this emotional misattribution. In other words, the less top-down control one saw on the amygdala the greater the susceptibility to this emotional dis-attribution bias was observed. And, so, to test that causally, what she did, with help from Jason Samaha, was deliver continuous data burst stimulation to the lateral prefrontal cortex, as well as to a control area in primary somatosensory cortex.
Just as an aside, this is something I have discussed with a couple of folks during my visit. We feel like in our lab that there is no good alternative to an active control, because sham -- among other things, sham stimulation doesn't control for what I think is the most important factor, which is the introduction of electromagnetic conduction in the brain. And, so, if you, with your sham, control, produce these peripheral sensations that are similar you're not actually doing anything to the brain of your subject.
And to really make a -- if you think about, this is a little bit oversimplified. But if you think about a neuropsychological study, for example, you get a double dissociation so that you're convinced that area X is responsible for behavior A. It's not enough to have a control group where you tap them on the head in a different area, from where their lesion is. It shows that tapping them on the head doesn't have an effect on behavior; right? So, to really have confidence that your effects are specific, you want to show that the same stimulation procedure in a different area that's not implicated in the behaviors that you're interested in does not have the same effect. So that was the logic here.
In the aggregate, what was found was consistent with the prediction, which was that delivering TMS to the prefrontal cortex had the effect of exaggerating this emotion attributional effect, the biasing of the judgment of a neutral face, in a way that delivering TMS to the S1 control area did not have.
But to more specifically get at the question of whether it was an alteration of function in the prefrontal cortex that was responsible for this effect, what Hegina did was regress the variability in the TMS-related change in alpha band power in prefrontal cortex against this misattribution effect. So, to the extent that subjects are falling below or above zero means that they are differentially susceptible to being biased in the negative or the positive direction. And what this shows is that it's the subject whose alpha band power increased as a result of TMS, and, therefore, the subject whose prefrontal cortices are believed to have been most affected by the theta burst stimulation, which intended to produce a hypometabolism target area, those with the subjects who showed the most pronounced.
Another interesting finding from the study was that the subjects were then contacted three days later, shown the same faces, and asked to make the same judgments. Importantly, though, this was done from their home, where they controlled the case of presentation to stimuli, and there was no affective time. So, the question is, is there an enduring influence of what you experienced in the laboratory three days later? And it turns out that there is; right?
Now, the conclusion is not that the effects of theta burst stimulation were somehow still influencing their performance three days later, it's that the theta burst TMS intervention in influenced their initial processing of these stimuli when they saw them in the laboratory, and that this, then, is what they encoded, was that this was a person who I don't like or a person who I do like. And regardless of what the factors were that caused the subject to make that initial evaluation, that's what stuck with them. So, in effect, this was a memory of how I had evaluated this person three days before.
Okay. What about real-time delivery of TMS while the subjects are engaged in a task, so online as opposed to offline TMS? So, this is a study I've described to you previously, the idea, subjects see four of these abstract shapes, and, depending on the instructions, they're either remembering what the shape looked like or where it appears on the screen, and they have to make a yes/no discrimination about location or about object identity. On half of the trials we delivered 10 hertz TMS, either to superior parietal lobule or to this control area in somatosensory cortex.
The initial findings for us were disappointing and were, in fact, reinforced earlier what I said about the importance of an active control. So, the influence of parietal cortex TMS on behavior was only seen in the location memory condition, not the object condition. But it was one of improving performance, and that performance improvement was comparable, whether the area targeted was S1 or was the superior parietal lobule. So, we thought we had an odd result, and that this was a result that was going in the proverbial style drawer of failed TMS experiments.
What Massy -- this is Massy Hemidi-- was look at individual variability in these data, and so what I'm showing you here is a group map that the top [inaudible] or band pass filters the alpha band. Here on the time frequency representation, you can see that in the parietal cortex there's markedly greater alpha band power during the object pass than during the location pass, during the TMS passes trial.
These are the TMS present trials, and you can see that TMS has produced visibly more energy in these lower frequencies. Now, remember, that the TMS is being delivered at 10 hertz, and, arguably, what's happening is this high alpha low beta activity has been modulated. But the important step, though, was then to aggregate across the delay period and look at the affect that TMS had on activity in this high alpha band. And what we saw was that the extent to which TMS had the effect of decreasing power in this high alpha band improved your performance, versus the subject who experienced an increase in alpha band power got worse. And this effect was specific, in that so here's a correlational map showing you the correlation of the TMS-related change in power to the TMS related change in behavior for a function of frequencies in time. This is what it looks like in a source localization.
But the specificity of this influence on individual variability is seen here in that it wasn't sufficient to deliver 10 hertz CMS to the superior parietal lobule task, because we didn't see the effect on the object task. We only saw it on the location task. And similarly, it wasn't sufficient to have TMS being delivered while you're doing a location task, because the effect was only seen when the location task was being performed, and we were targeting the experimental area and not the control area.
So, even though on the aggregate level we didn't see differences between our experimental area and our control area, we did see differences in terms of how the TMS influenced individual physiological profiles, which, in turn then, influenced the individual behavior or our intervention of our interventions.
I'm going to skip through the details of this last study. This is just a comparison of visual working memory to visual attention, and the point to be made here is, again, that the primary functional effects of TMS, we're at higher frequency -- in higher frequencies than the 10-hertz stimulating frequency at which we delivered our stimulation.
And, so, this leads to the question of whether the functional consequences of TMS may be happening in frequency bands that are different than the frequency band at which we are delivering the stimulation. I'm going to skip over this as well, in the interest of time, and answer the question that I just had.
Okay. So, this is work from Jeff Johnson that we haven't yet published but that starts to get at this question. So, looking in the delay period, when we're delivering a train of 10 hertz TMS to this parietal area, if we look at the time frequency representation at the group level, you can see that there's this band of elevated alpha band power in both the color memory and the color memory location tab. When we introduce high-frequency TMS, most of the energy is at 20 hertz and above, even though we're stimulating at 10 hertz, and, initially, this was puzzling to us. Here's the data from a single subject. What I had shown you before was group data.
But then Jeff generated this ERP of the same data, so at times zero is when the trial starts, and you can see, even in the back of the conference room, with your naked eye, what the TMS-evoked responses look like. So, I'm super imposing here these bars I show you when each pulse was delivered, and you can see that at this parietal electrode there are two full cycles of the evoked response that are generated by each pulse of TMS. So, in fact, even though the exogenous energy is fluctuating at 10 hertz, the proximal effect on the brain is to introduce this 20-hertz variable -- time varying signal that wasn't present.
Now, remember that I told you that occipital -- excuse me, that parietal cortex seems to resonate in the beta band, whereas occipital areas resonate in the alpha band. So, another question that we wanted to look at is what -- if we're delivering this same 10 hertz TMS procedure to a different patch of cortex. So, this is data from Drew Sheldon, among others, and what I'm showing you here is the ERP from the delay period of a motion, short-term memory task. And I'm showing you the electrode at area MT, which is an extrastriate cortex rather than a superior parietal cortex.
Before and after we delivered TMS, the bracket is just to indicate one millivolt -- one microvolt, excuse me, and so here's one microvolt in the time series with TMS introduced, and so right away you can see that the amount of energy in the ERB is markedly greater in TMS. This is where each of the TMS pulses was delivered. And an informal inspection suggests that what we see is a fairly robust initial one-cycle of an evoked response that looks fairly clean, but then that dies out for the remainder of this period until the subsequent pulse of TMS. And so, just at this gross level, the details of the evoked response are going to look quite different depending on which cortical area you're targeting.
This is looking at the same MT electrode when we're targeting this meta-sensory cortex, and, essentially, we see the same thing. It's just in a phase because we're stimulating a distal area. We're looking at the response that's several synapses away. But you get the same kind of thing, a one-cycle fit into a hundred-millisecond period of time, as opposed to two clear cycles that we got -- that we saw when we targeted [inaudible].
So that's just the last slides that I'll show you are looking at the spectral transformation of those MT data that I just showed you. So, this is the trial without TMS, and here we see the power associated with a 10-hertz train of TMS is delivered during the second before the trial started. And what I'll call your attention to is the structure of the oscillatory activity during the delay period of this working memory task. It's almost unchanged; right? So, the TMS effect itself is dramatic.
But if this 10-hertz train of TMS was having a training effect, you would want to see it persist through the delay period, and we don't see that when we look at the power. We also don't see that when we look at the intertrial coherence. So, this is the trial-to-trial oscillatory coherence, which is dramatically elevated from trial to trial when the TMS is delivered, but there's really no influence, no evidence of a residual influence of this 10-hertz train during subsequent portions of the trial after the TMS pulses have been [inaudible].
So, when you think about this, in conjunction with the three days later emotional misattribution effect that we showed you earlier, we're talking about different time scales, but one suggestion might be that if you want to influence a habit or a stereotype behavior, you might really need to show that you're having that effect at the time that you're delivering the TMS, and if there's some way in which you can fundamentally change the way people respond, to certain stimuli for example, that can be learned in the presence of the stimulation, that may well affect behavior subsequently in time. But it's probably not the case that the activity of the brain later in time is still reflecting the effects of the stimulation; right? So, any learning or any modification would have to happen while the stimulation.
I've run over my time, so I'll just very quickly tell you that one area that we're looking at to try to get more sophisticated in terms of identifying the influence of TMS on neural activity is this signal decomposition approach that Roemer van der Meij has developed with parallel factor analysis, and, in essence, what it does is it allows us to separate sources despite the fact that they might be overlapping in space and somewhat overlapping in spectral patterns. So, it gives us a principled way of deconstructing -- Saskia Haegens has been involved in this work -- in some initial analyses she's done, what I'm showing you are just the data from three different subjects, showing that there is this component with this scalp distribution in these three different subjects, with this spectral distribution that's centered over 20 hertz in response to the 10-hertz stimulation that we see elevated with TMS delivered versus TMS not delivered.
So, this is a way that we're going to try to, in future work, try to tease apart what are the proximal effects of TMS that are not related to the endogenous physiology that was present in the brains of subjects before we stimulated to help get a handle -- to get a handle on whether the behavioral effects of TMS that we're seeing are related to de novo energy that we're introducing in the system or rather, whether what we're doing is biasing endogenous structure in those systems as it would perform the task in the absence.
So, what I've shown you is that when we paired TMS with concurrent physiological measures, it markedly increases our experimental sensitivity and we can start to make sense out of these trait, state, and disease-related patterns of variation, especially in connectivity, and test hypotheses about the functional relevance of these variations in physiology that we may not otherwise be able to test without a causal intervention.
When we deliver repetitive TMS, the effects are highly variable and so without an independent measure of the physiological consequences of your stimulation, you're often left with a null aggregate effect on the behavior. There may be some richness there if you have an independent physiological regressor that you can use to make sense out of the variant behavioral data. But without that, you're left putting your study in the pile drawer, which I think has happened a lot over the course of history with that.
Concurrent EEG can predict the magnitude and the direction of behavioral effects, but one of the things that we need to get smarter about is doing this a priori. So, all of the correlations that I've shown you, the relations between TMS and individual variability or the effects of TMS on behavior are things that we calculated post hoc. And going forward, of course, to really exploit this approach, what we need to do is no a priori, what are the factors, what are the dimensions in somebody's physiological makeup that will predict whether this set of interventions will have a positive or a negative or a null effect.
And the working hypothesis in my lab is that you feel like although these evoked effects are unquestionably the largest in magnitude that we see in the signal, the evoked effects seem to have little influence on function, and that the most traction that we're getting on influencing behavior is through modulatory influences of TMS, which are more subtle, but which, in effect, can be summarized as modulating or biasing the gain in oscillatory structure that was present in the subjects data before you delivered this innovation. So, your biasing was already there, as opposed to introducing a de novo regime that's influencing behavior the way that you want to.
So that's it. Thank you very much for your attention. I'm happy to take questions [inaudible].
If anyone online has a question, they could enter it in the chat window.
Yeah. Right. So, to what extent are these effects due to gross anatomical differences versus something more interesting? I think that the bit of data that I showed you where we saw differences in the primitive rendition of the physiological fingerprint of individuals where there's considerable variability. Well, so it's sort of a cross -- a broad range of frequencies and the finding was that when we targeted one area, the distal evoked response in prefrontal cortex, midline frontal cortex, occipital cortex could be quite variable across subjects and was predicted by the oscillatory fingerprint of this readout from a parietal electrically. I think that just simple changes in anatomy -- differences in anatomy would be able to account for that.
They could, but I don't see how that would influence -- how that would account for the differences in the readout of the TMS-evoked response in these anatomically distances. However, what would the frontal cortex response be different in a predictable and frequency ban-limited way due to the vagaries of holding parietal cortex.
Yeah. Yeah. But, I mean; right. So, one thing that I think, just to pursue that a little bit, if we're smarter about using data reduction approaches that are able to sort of abstract across some of these variability, that's probably not interesting, to highlight the factors that are. So, I'm thinking of Kristof Michelle and Micah Murray have these approaches that give you, like, a 12-dimensional brain score [inaudible]. And, in principle, something like that that should be less susceptible to, for example, the vagaries of the total depth might be a good way to go forward.
Right. Yes. How do I square it? Well, there's broad spectrums oscillatory activity in every area of the brain. If you remember, from that figure, the peak spectral invoked response, the way they defined it was the frequency at which the evoked response persisted the longest. So, it wasn't the case that you had this very narrow band at, you know, 30 hertz that kind of stuck out and that there were no influences at higher or lower frequencies. It's also the case, from thinking of electrocorticography work this Brad Voytek has done, for example, that every cortical territory that's studied can be construed as having low frequency carrier waves or may be more trending into the theta band or the alpha band, depending on where you are, on which are nested, then, higher frequency bursty oscillations that may correspond to the organization; right? So, and that's seen in frontal cortex as well.
So, there is always -- and you'll see one over F structure in the EEG everywhere. So, it's not as though you don't see alpha in the frontal cortex; right? So, in this case, it's using alpha band power as the gross proxy for overall state of excitability of a patch of cortex [inaudible]. So, the prediction would be -- and I would have to look for this -- but I imagine if you delivered theta burst stimulation across the whole of the scale, if you were effective in inducing a hypometabolism in the targeted area, you would see elevation of resting alpha in all of those.
My question [inaudible] the TMS study showing differences between patients [inaudible]. You made the comment TMS allows to apply the same energy to the representative range. How do we know the response [inaudible] generated electric field [inaudible] due to the [inaudible]?
Yeah. So, what I assumed when I was making that statement was the fact that the proximal -- the spatially proximal tissue in which the -- the tissue that actually experience the electromagnetic induction, was that precentral gyrus, the hand area of the motor cortex. And the physiological evoked response, as measured by fMRI, was no different between the patients. So, I took that as evidence that the local effect of excitation was comparable to [inaudible]. That allowed me, then, to interpret the differences that I saw in the downstream area, whose activity had to have been a trans-synaptic [inaudible].
So, I don't know, is it the integrity of the axons between the motor cortex and the thalamus? It is a pathway that I wasn't taking into consideration that went trans [inaudible] and through the post-central -- there are all sorts of possible explanations, but the most parsimonious one that seems to be from the data is that reproduce comparable subcutaneous excitation in the overlying motor cortex, and that it was the influence of the results in volley of actual potential that hit the thalamic reticular nucleus and thalamus that accounted for that. Another thing that I can point to is that the functional connectivity measures cortico-cortical connectivity measures, whether looking without TMS or with TMS, the comparable between the groups. So, in terms of my measure of the patterns of covariation between the area that we targeted and, let's say, the frontal midline area, those didn't differ when I only included those two nodes in the network.
When I included the thalamus, and allowed variation in the integrity of the somatic response to account for some of the variability, then we were able to predict those distal effects of the cortex.
[Inaudible]. The first one, might we use QEG as effective [inaudible] test the individual states?
Right. That's kind of what I was getting at earlier. I think something more sophisticating than what I've been showing, which are essentially these ERSTs that are taken from an arbitrarily selected electrode or group of electrodes is certainly going to be a more sophisticated way to go forward. I think what we have so far, what I've shown is promising, that with these more simple methods [inaudible]. I think that this source decomposition approach that I hinted at right at the end of my talk either more promising ways for us to go forward.
[Inaudible]. I've had three clients who received TMS. [Inaudible]. Perform patterns of EEG [inaudible].
I would have to know what the TMS -- specifically what the stimulation protocol was before I would venture a -- this person wants to go offline.
I didn't hear the first of.
Do you have a [inaudible] over a long period of time [inaudible]?
We've looked at whether there are enduring residual effects, and all we've seen are comparable to what I showed with this affective misattribution pass, is that if we can produce some systematic variation behavior in the lab while we're stimulating, that that will be what the subjects obtain when we bring them back and evaluate them later. We have not seen -- we haven't tried to, and we also have not any evidence of prolonged physiological alteration working. Thanks.