PRESENTERS

• WAYNE FENTON, M.D.
• STEPHEN MARDER, M.D.
• CAMERON CARTER, M.D.
• ALAN BREIER, M.D.
• JOHN JONIDES, Ph.D.
• JAMES HAGAN, Ph.D.
• TREVOR ROBBINS, Ph.D.

CONTENTS

PROCEEDINGS

	DR. GEYER:  People can grab their coffee and take seats.  We're going to try to get started and try to stay on time.
	So the program for today is an attempt to integrate the discussions that we had yesterday.  The process that's involved to try to build a consensus for a research agenda, particular the cognitive effects of schizophrenia.
	First will be, in the morning, a series of reports from the breakout group co-leaders.  Each of them being 15 minutes of presentation, with 15 minutes of discussion.  I want to remind everybody, when they're asking a question or making a comment from the floor, to use a microphone and identify yourself.  The transcripts will all be--as has been the case for the Matrix Meetings, in general, transcribed and put on the WEB.
	So, please remember to identify yourself and use a microphone.
	As far as how the groups will go--a little out of order because of travel schedules, instead of the normal one through eight we're going to go--let's see if I can get this right--one, two, six, three--[laughs]--
	[Laughter.]
	--and then four, five, seven, eight--in the second part of the morning.
	After lunch, then--and at lunch, if there are particular areas of synergy or discord that are identified in the overlap from the morning breakout groups, I'm going to invite you all--those most passionate about those areas of synergy or discord, to sit together at lunch and try to resolve them and build upon them--preferably--so that we can integrate our thoughts and develop in the afternoon session, as explicit and concrete a research agenda as possible.
	The afternoon session, then will essentially be this integrated discussion, the first part of which will invite the speakers from the morning session yesterday--who I conceptualize as essentially the consumers of this research agenda-- representatives of NIMH, industry and academia--to comment upon what they heard from the breakout groups; what gaps have not been identified that they can identify that still need discussion; and then the co-leaders of the breakout groups will also join in that discussion, as will the audience.  And that's certainly the most extemporaneous part of the whole meeting.
	So--rather than take any more time in administration, I'm going to introduce, first, Cam Carter, who you know served multiple roles as a member of the New Approaches Committee, and led one of the breakout groups, as well as gave one of the presentations yesterday morning.
	Cam?

Biomarkers for Clinical Trials I-Neuroimaging

  DR. CARTER: [Comments off mike.]

  DR. GEYER: It will be better on the tape. Can you actually--for purposes of 
  the transcription, would you be able to use the lectern, by any chance, or do 
  you need to be near your--

  DR. CARTER: [Off mike.]

  [Discussion off mike.]

  DR. CARTER: So, we had quite a spirited discussion in our breakout group, and 
  we were able to converge on several key problems that need to be addressed. 
  And I'm going to try to follow the format everybody got yesterday and identify 
  three top problems that need to be addressed.

  There were a couple of general things that, I think, were quite helpful and 
  that were clarified in our discussion.

  One was that--and can probably be elaborated in this group--one was that--from 
  industry, I think, it was clarified for us that we didn't need--we had doubts 
  as to whether we had what could be considered to be true [INAUDIBLE] biomarkers, 
  or surrogate markers for treatment. And it was clarified for us from the people 
  from industry that that isn't really what they particular need. It was unlikely 
  that anyone would want functional MRI to be--of cognition in schizophrenia to 
  be approved as a biomarker. What was needed was some information that would 
  help govern decisions early in drug development, and that should we have reliable 
  measures of brain activity associated with cognition that would be sensitive 
  to change with medication, that that would be quite sufficient.

  I think another point that was clarified was that despite the general confidence 
  of those of us that practice FMRI, both in basic healthy populations and in 
  clinical studies, there's not necessary the same perception of this method in 
  the field; and that is, in part, driven by realistic concerns about the method 
  and, in part driven by the fact that there aren't a lot of--there's a relatively 
  large and growing literature and a number of discrepant findings. And, again, 
  those of us in the field often feel pretty comfortable with the discrepant findings. 
  We think we can reconcile them, in terms of differences in task design, and 
  patient populations, and the way the performance is addressed in the studies. 
  But we were certainly sobered, I think, by the perception from other scientists, 
  as well as from industry, that it's a confusing literature, and that raises 
  considerable doubts about the potential usefulness of this approach.

  [Slide.]

  So the three general issues that we identified were: the first, obviously, which 
  aspect of cognition would one try to use to engage relevant brain circuitry 
  and to use as a target for drug development. This is something that breakout 
  four--I think it is--is specifically focused on later in the afternoon. And 
  our research agenda associated with those questions will be discussed quite 
  soon. But this was obviously the very first step--identifying the domains, identifying 
  the relative tasks that will be used; experimental designs that would be appropriate; 
  and the psychometric norming of those tasks. Obviously garbage-in-garbage-out 
  in any neurometric studies. So if one uses tasks that either aren't' valid or 
  aren't reliable, the FMRI signal that's associated with them is certainly not 
  going to be interpretable or valid.

  A second general issue, then, relating to the method itself is that it needs 
  to be validated. There are only a few studies that have addressed the issue 
  of test-re-test reliability. And it's unclear whether those studies are really 
  enough to satisfy people who might be considering investing large amounts of 
  money in using this technology to guide vital decisions. So I think there was 
  quite unanimous agreement that we need to do more in order to clarify what the 
  test-re-test reliability in the method is, and also how reliable it is across 
  centers-- so--within centers and across centers.

  There was also, I think, fairly clearly, a need for some study--study or studies--that 
  would address some of the most obvious conflicting results that are in the literature. 
  One of the examples that came up a couple of times was the results of FMRI studies 
  using the in-deck task. And as sort of informal meta-analysis that one of the 
  non FMRI neuroimages had made, which suggested that it was 50 percent one way 
  and 50 percent the other, in terms of the findings. So, again, those of us that 
  do this could probably propose some experiments that would resolve these discrepancies. 
  And it did appear that that would be something that would be helpful for the 
  field; it would increase confidence in the method.

  [Slide.]

  And then the third set of problems--so, just to go over this format here, I've 
  identified priorities, and talking about obstacles--so, for the first priority, 
  the obstacle is we need the data about--we need a consensus about domains. We 
  need to have some consensus about the tasks, and we need data about the psychometric 
  properties of those tasks.

  For the second, we need data about--additional data about reliability, and perhaps 
  need some data to address some to the discrepant results that are out there.

  And then the third key point was, again, a largely technical one related to 
  potential drug effects on the very signal that we're studying; so, the coupling 
  of neural activity to the hemodynamic response, the potential for drug effects 
  to alter the blood flow signal itself in various ways. This is something that 
  would, in practice, need to be addressed on a molecule-by-molecule basis, but 
  I think what is needed is an approach to do this--the development of an approach 
  to do this.

  And so that takes me to the notion of the research agenda. Essentially, with 
  regard to the first need--the need to clarify development domains, develop tasks--I 
  would suggest you see breakout group four. But also, to note that the psychometric 
  data will be key, and at the end of that norming process that might be part 
  of a research agenda for breakout group four, we still might need some additional 
  tweaking. The assumption is that most of the tasks that would be development 
  from an experimental cognitive perspective would be easily implemented in FMRI 
  studies. But there still might need to be some modification of those tasks to 
  make them totally appropriate for use in the MRI environment. And perhaps a 
  quick re-evaluation of the psychometric properties of the slightly modified 
  tasks, if that were to happen.

  A second component of the research agenda would be to validate FMRI itself, 
  and to resolve some discrepant findings. Test-re-test reliability, again, is 
  something that almost no one does. There are couple of published studies of 
  FMRI that we all refer to when we're asked about it. Those studies have very 
  small numbers of subjects. They tend to look at reliability at the region level, 
  rather than that the voxel level. And they focus on one aspect of the signal 
  or the other, and I think there are multiple aspects of the signal that we get 
  from FMRI that are information, and so understanding what the kind of neurometric 
  properties of that signal would be would be extremely important and extremely 
  helpful in terms of building confidence in our ability to use this as a so-called 
  biomarker; information that would be useful for decisions during drug development.

  And there was also the sense that perhaps, as part of that agenda, it might 
  be worth trying to resolve some of the discrepant findings; so, something more 
  of a scientific idea, with one, at least, of the common tasks that are used 
  in neuroimaging studies, where we might perhaps propose using identical versions 
  of different groups' tasks, and within subject and test predictions about what 
  might be the aspect of those studies that will lead to the apparently discrepant 
  results: is it task design, is it how performance is handled? Is it some other 
  factor?

  Then the third component of the research agenda--again, trying to deal with 
  potential obstacles to using this method as a target--include developing and 
  testing methods for evaluating the effect of drugs on the coupling of neural 
  activity and the hemodynamic response. This might involve animal models. It 
  might involve combining non-hemodynamic methods, such as cortical EEG, or MEG 
  with hemodynamic methods. And, again, in practice, this would involve a module-by-module 
  testing, but developing a creditable methodology for testing this would be an 
  important step towards making this a more realistic approach.

  The same can be said for developing methods that would address the potential 
  drug effects on the cerebral vasculature, so we will know that many of the drugs 
  that affect receptors in the vasculature--and although it's often assumed that 
  this is not a significant effect, global effects in blood flow can affect the 
  ability of cognitive tasks to induce relative local change, and there all methods 
  currently available that can measure blood volume--cerebral blood flow--directly. 
  Thee are not quite ready for prime time, but in principle, since slightly more 
  invasive methods, such as the use of contrast agents with FMRI, or the use of 
  methods such as PET--methods such as PET, could help to address these problems.

  So we believe that some work to develop and validate these methods would be 
  very helpful to the field. They could then be important into studies where each 
  individual molecule could be examined for these kinds of effects prior to going 
  to FMRI.

  And this particular component would probably also need to be accompanied by 
  some proof-of-concept work, with one or two agents that might be known to either 
  affect the coupling of neural activity to the HRF, or to directly affect the 
  cerebral vasculature itself, so they could be validated in a design that involved 
  pharmacological challenge.

  And that's our three suggestions for a future research agenda.

  Do we have our discussion now, Mark?

  DR. GEYER: Yes. We're open to discussion from the floor.

  There are microphones around. And--this is the opportunity.

  [Pause.]

  DR. HAGAN: [Off mike.] Jim Hagan. On the vasculature issue--

  Just to make a comment on the pharmacological effects on the vasculature, I 
  think that's a fundamental issue. And this is probably an area where small-animal 
  work can really help to try and understand what the potential influence is going 
  to be. And I think the literature on this is already beginning to grow. We've 
  been able to do some studies in our unit to look at various mechanisms, and 
  to track and to try to understand how vascular effects track with the pharmacological 
  FMRI effects. And I think that's a very valuable approach, but you need to understand 
  that before moving into man.

  I think it's very dependent on mechanism. It's something which you need to address 
  on a mechanism-by-mechanism basis. But there are ways in which you can do it. 
  I think the greater issue is--or the more difficult issue is determining effects 
  on micro-vasculature; so, determining or understanding gross hemodynamic or 
  vascular changes is do-able, but I think the issue that we face is what happens 
  when you get down to the level of the micro-vasculature, and that's difficult 
  to pick up those signals using conventional hemodynamic recordings. But I think 
  that's an area where the field needs to focus, and the potential is there to 
  do it.

  DR. NUECHTERLEIN: I was wondering, Cam, was there any discussion about whether 
  there are any methodological obstacles to doing test-re-test reliability studies 
  with the--oh, not close enough? I'm hearing an echo--

  COURT REPORTER: Your name?

  DR. NUECHTERLEIN: Oh, Keith Nuechterlein--to doing test-re-test reliability 
  studies with FMRI--are there sort of any technical problems that would get in 
  the way of doing such studies? Or is it more that investigators simply haven't 
  pursued them, as it hasn't seemed a high enough priority? Are there any problems, 
  for example, with the kind of relative nature of the signal, instead of the 
  absolute sort of quantification?

  DR. CARTER: Greg might want to jump in here, too.

  I think there definitely are issues. In addition to the issues of test-re-test 
  reliability of the tasks that you're using to engage the brain activity, I think 
  the technical issues around head registration and those sorts of things are, 
  I think, you know, completely manageable. In terms of which are the appropriate 
  measures, and how those things are affected by, you know, different scanners, 
  different tasks, and different procedures, are pretty complex. I know that Greg 
  has been involved in the BIRN initiative, which I one of--I think the only kind 
  of group effort here in the U.S. to try to address issues like this--methodological 
  issues like this.

  Maybe he wants to comment a little bit more.

  DR. McCARTHY: Yes--Greg McCarthy.

  We're been involved with the Biomedical Informatics Research Network, which 
  is an NCRR-funded infrastructure for imaging research. And, currently, there's 
  11 centers doing test-re-test reliability in two or three functional MRI challenge 
  tasks that are then being applied to a population of schizophrenic individuals 
  in which there will also test-re-test reliability. So, it's an extremely important 
  issue, and it's something that the community, I think, is just beginning to 
  address. But I think it's being addressed finally.

  DR. CARTER: Were their specific things, Keith, that you had it mind? I mean, 
  one can think of all sorts of technical issues, from how you position the head 
  in the scanner, to what your task is like, and how much movement it elicits.

  I think, again, all of those things can be more or less controlled to the level 
  where one could do studies that, in least of the context of the particular paradigm, 
  could then make claims. And I think it's probably valuable to use multiple procedures 
  to get a sense of whether reliability really does vary according to the design 
  of the experiment. I think that's really important.

  And then there's the question of what is it that you want to be reliable? So 
  is it--and this--in the field itself, I think one would need a little bit of 
  discussion to see which things are relevant? So is it percent signal change? 
  Is it number of activated foxiles? Are you just interested in suprathreshold 
  statistical maps? You know, what kind of bases would you introduce when you 
  did that? So it is fairly complicated.

  Yes?

  DR. JONIDES: This is John Jonides. Let me just continue on the reliability issue.

  Presumably the reliability of the signal test-re-test--for example, on the same 
  subjects over long periods of time, or over a single session--is going to depend 
  a lot on which areas of the brain you're measuring. Certain signals--at least, 
  in my hands--turn out to be much more reliable than other signals--especially 
  the most posterior regions of the brain we yield, I think, much more reliable 
  signals. Greg might want to comment on this also--compared to the more anterior 
  parts of the brain.

  So I'm wondering where there is a kind of systematic effort to map out, as it 
  were, the reliability of the signal as a function of spatial position in the 
  brain?

  DR. McCARTHY: Is there a systematic effort to do that in the BIRN? I'm not aware 
  of a systematic effort, but it obviously will be valuable to not just use motor 
  strip or visual cortex to test that.

  The published studies by Dadenal and Darin Anewack look at prefrontal cortex.

  DR. McCARTHY: [Off mike.] This is Greg McCarthy, and I speak to the--what the 
  BIRN is doing on that. There's actually several calibrations of style s, like 
  sensory, motor, visual motor-squeeze, and then breath-holding tasks, and then 
  there's a variety of auditory-sensory tasks, and then executive-function tasks. 
  So, yes, we're trying to look at all aspects of function, and I actually agree 
  with your point about posterior being much more stable than prefrontal.

  DR. NIMERICK: [Off mike.] Jeffrey Nimerick. Although I agree that FMRI probably 
  holds the most potential--was there discussion of other applications, as well, 
  such as spectroscopic imaging with structural [INAUDIBLE].

  DR. CARTER: There was a fairly extensive discussion about PET. And so--yes--I'm 
  sorry--the sense was that the limitation with PET is the available good, well-behaved 
  ligands. But in cases where there all well-behaved ligands, it's a completely 
  invaluable and quantitative tool. People, in fact, from industry, I think, felt 
  a lot more comfortable with PET, when the ligands were available, and with the 
  notion of using functional MRI.

  DR. NIMERICK: If I could ask one other question. I was interested--was there 
  discussion about the difference--the distinction between bio- and surrogate 
  markers?

  DR. CARTER: There was.

  DR. NIMERICK: Maybe you could just quickly summarize that for us.

  DR. CARTER: Well, I'm not sure if I can. The key people who contributed to that 
  were Shitij and Dan VanKammen. Are either of those guys here? Maybe Shitij can 
  summarize it.

  There was a bottom-line conclusion that came from Dan, which was that he really 
  didn't care whether this approach met criteria for a formal disease biomarker 
  or not. He felt that what he would like would be some information about the 
  effects of drugs on brain circuitry related to cognition, regardless of whether 
  that was a core biomarker of a disease process or something else.

  Shitij?

  DR. KAPUR: Shitij Kapur. My recollection of that discussion was the following: 
  that when it came to biomarkers, there was a wide variation in definition of 
  what people were calling biomarkers. But when it came to surrogate markers, 
  the understanding was that the surrogate marker is some marker that will be 
  accepted--perhaps by the FDA--in place of the primary endpoint. And the very 
  rapid conclusion was that there was nothing currently available that could possibly 
  be accepted as a surrogate marker. So it could not be the endpoint of any registration 
  trial or anything like that.

  So the main value of when you're imaging would be to be a biomarker. But there 
  was a very wide variety of definitions as to what a biomarker could be.

  The only commonality among those definitions was that it should be reliable; 
  it should help us predict something more efficiently than measuring that endpoint 
  itself. But precisely what that was was totally open and different across the 
  participants.

  DR. FOX: Gerry Fox. I just had a couple of points on FMRI.

  If there's an interest in comparing pre-clinical with clinical FMRI data, you 
  might want to take into consideration whether you image awake versus anesthetized 
  animals. And there's a few groups starting to do awake animals.

  Second point is: there might be some degree of overlap with the issues that 
  you identified with some similar issues that NIBIB might be grappling with. 
  So it might be worth talking to them, as well.

  DR. GEYER: Mark Geyer. Given what Greg told us about the very exciting BIRN 
  project, it sounds to me--and correct me if I'm totally wrong--that at least 
  some of the test-re-test reliability issues have already achieved prominence 
  in a funded mechanism. And so I want to bring the group back to the specific 
  need for the specific research agenda that would build on what will be available 
  from BIRN, and what are the missing gaps vis-a-vis pharmacological treatments 
  of cognition and schizophrenia that your group sees as--given BIRN--what else 
  is needed now?

  DR. CARTER: Well, again, I'm not sure that there is a sense that BIRN will provide 
  all the data that we need, given that, I think BIRN is largely addressing issues 
  related to kind of Problem Two--the problem of the reliability of the method 
  itself. The emphasis would probably be on part two of point two, which is this 
  issue of resolving discrepancies. There's definitely, I think, a scientific 
  need to resolve discrepancies in the literature, and point one, which has to 
  do with, you know, what are the domains, what are the tasks, and what are their 
  properties in the MRI environment, and point three, this very important issue 
  of what drugs actually do to the measurement method itself, and how can that 
  be addressed on a molecule-by-molecule basis.

  DR. GEYER: So it strikes me that--Mark Geyer, again--it strikes me that that 
  further need will need to be integrated with what carving cognition at its joints, 
  and the distinguishing the different domains, because what strikes me is that 
  FMRI will provide--given test-re-test reliability can be ascertained--good tools 
  for addressing the separability of the domains. So there a couple of other breakout 
  groups that, it seems to me, we need to integrate with to achieve a specific 
  research agenda.

  John, you were waiting patiently.

  DR. KRYSTAL: Sure. A couple of comments: one is that it seemed one of the issues 
  that came up is the issue of the relative nature of the bold signals as a measure, 
  as opposed to absolute quantitation. And one area that we discussed somewhat 
  as a potential value to this kind of work are the development--technological 
  development--of the quantitative techniques that might yield absolute quantitation 
  for perfusion and functional activation; one example being achuros bin labeling, 
  but other approaches that are being developed, as well, that may yield absolute 
  measures because of the possibility that drugs may affect, not only the extent 
  of activation received, but also baseline activation, which we may not be able 
  to measure quantitatively.

  Am I not speaking close enough? Okay. I don't think I need to repeat that, though, 
  right?

  COURT REPORTER: But your name.

  DR. KRYSTAL: Oh, my name. My name--John Krystal.

  The second point, and the second link that I think we haven't talked about in 
  this is the way in which pharmaco-FMRI can serve as a link between animal models 
  and the disease state, so that we can model things in humans with drug administration, 
  and evaluate putative pharmacotherapies in those human models. And I think we 
  know relatively little about that, and that's an area that needs to be a research 
  agenda, because I think that knowledge gap is an important one, in terms of 
  taking full advantage of our very well developed animal literature, in terms 
  of thinking about how to develop new drugs for people.

  And I just wanted to make a brief comment about Jeff Lieberman's prior comment 
  about spectroscopy, which we didn't talk about at all in our meeting, and just 
  to point out that there are sort of two issues there that he raises in a way: 
  one is that there is an emerging application of spectroscopy as a tool for human 
  cortical-neural chemistry. So if you want to know whether a given drug reduces 
  cortical-glutamate release in humans, we now have tools that we can begin to 
  directly and quantitatively answer those questions. They're very expensive and 
  cumbersome, but I think that there will be progress in those areas, and the 
  development and the further refinement of those technologies--I think it will 
  be a useful and important research area.

  The other issue--which we didn't talk about--is--but is, I think, relevant to 
  cognition--is the extent to which drugs are affecting the integrity of cortical 
  architecture and thereby indirectly affecting cognition so that the drugs that 
  have neurotropic or neuroprotective effects, we may be able to use structural 
  and spectroscopic imaging to help to document those effects, as well as the 
  acute effects on cognition, and thereby provide unique end points which we couldn't 
  get from the clinical outcome measures about ways in which these drugs are affecting 
  function.

  DR. GEYER: We have time for two more. Martin?

  DR. PAULUS: Martin Paulus.

  I was wondering: was there any discussion about key pharmacological probes that 
  would need to be tested in healthy volunteers to get a better understanding 
  of what actually--some of the known compounds do to specific tests; say tests 
  of working memory or of other executive functioning?

  DR. CARTER: There was actually, you know, a couple of fairly extended discussions 
  about sort of human pharmacological models of disease, and the potential usefulness 
  of those. Yes.

  DR. HOYER: Daniel Hoyer.

  We have indeed talked about compounds, but the problem with using compounds 
  in a paradigm which is not very well established--and Cam has summarized those 
  studies--we have quite some reservation about the use of FMRI, because--not 
  that the technique is not valuable, but the technique is not validated--right? 
  So we would like to know how the brain of a schizophrenic patient, doing different 
  tasks--which are going to be defined by Group number Four--is looking like. 
  It is nice to take any compound into a human volunteer. It's not going to tell 
  you much about what this compound is going to be doing in someone suffering 
  from schizophrenia and undertaking a cognitive task.

  We've been all playing with these sort of things. Pharmaco-EEG has been in the 
  industry 10 or 15 years ago, and it hasn't led anywhere. And the problem of 
  any compound taken into any volunteer is that you're going to get a signature 
  which may be largely dependent on the compound, but not tell you much about 
  what this compound may be doing in that structure: in the schizophrenic patient.

  So the idea was that we would, indeed, favor the use of FMRI in schizophrenia, 
  in cognition, but first validate the method and define which brain regions are 
  being affected during the task which we have assigned the different groups to 
  come up with. And then, when this is done, then start with treatment, assuming 
  that the treatment is doing something in the first place. I mean, what's the 
  use of taking compounds right now when you don't know that these compounds have 
  an effect on schizophrenia and cognition? It's useless. It's interesting. It's 
  going to produce nice images. You can publish it. But so what?

  So that's why we--this is why we haven't addressed compounds in the first place.

  DR. CARTER: Right. I would kind of not that that gives us a little bit of time 
  t o validate the method. Hopefully, we will indeed have an effective molecule, 
  and will be able to test the method with that molecule. But we have some time 
  to do that.

  DR. GEYER: So, I'm going to introduce the co-leaders of the next group, which 
  was "Distinguishing Separable Domains of Cognition in Human and Animal 
  Studies: What Separations are Optimal for Targeting Interventions"--Keith 
  Nuechterlein and Trevor Robins were the co-leaders. Keith will--another member 
  of the New Approaches committee will present.

Distinguishing Separable Domains of Cognition in Human and Animal Studies: What Separations are Optimal for Targeting Interventions

  DR. NUECHTERLEIN: So, our discussion was rather far-ranging. The topic was a 
  very broad one, in a way, starting with the seven domains that the MATRICS Neurocognition 
  Committee had earlier decided were the domains that we thought were relatively 
  separable ones at this time--at least in the psychometric statistical sense.

  [Slide.]

  The first of three major points that came up as recommendations was that--as 
  we might hear echoed in another group--that the major obstacle to incorporating 
  advances in cognitive neuroscience is not at this point the model development--although 
  that, of course, can use additional work; it's really the practical psychometric 
  issues of the major measures that employ new distinctions made prominent in 
  recent cognitive neuroscience so that it seemed like, to address this, the issue 
  was partially one of the individual investigators who typically do the studies 
  in human cognitive neuroscience to develop new paradigms, new measures, parsing 
  of processes don't put as their natural high priority like test-re-test reliability 
  and repeatability, practice effects--if there are practice effects, whether 
  you can use parallel measures to address those questions.

  And also, generally--although this might be a later stage--don't address the 
  sort of validity question ultimately of interest to the FDA of whether these 
  measures have any relationship to functional outcome.

  So one way of addressing that the group discussed was some sort of pairing of 
  psychometric experts with cognitive neuroscience investigators in a way that 
  doesn't naturally occur, so it would probably take some special incentives to 
  do such work.

  [Slide.]

  There was discussion about the fact that this might actually have the potential, 
  though, as you had to tweak measures, to compromise some of the basic distinctions 
  between cognitive processes that you were seeking to maintain, so that you'd 
  have to kind of go back and forth a bit if you had to actually change the paradigms 
  somewhat in order to get them to be psychometrically stronger, then you'd have 
  to go back and make sure that you're still parsing the same processes that you 
  intended to to begin with.

  So it seemed that this would take a bit of shift in, say, research emphasis. 
  There's not so much anything technically problematic with doing this; it's just 
  not what people would sort of naturally put as their first agenda.

  It was thought that if the cognitive neuroscience advances were coupled with 
  psychometric advances, that maybe 10 years from now, if we were

  doing a MATRIC consensus cognitive battery, you could then incorporate a lot 
  of the newer distinctions made by cognitive neuroscience and partition the cognitive 
  processes in a more fine-tuned way.

  Now, what do I have to do to move this? Is there a secret to this?

  [Discussion off mike.]

  [Slide.]

  Okay. So a second general area--there was discussion--there was a more general 
  discussion about beyond the kinds of studies that we use to decide which were 
  the separable cognitive domains, which were mainly factor analytic studies of 
  human performance with, of course, mainly existing well-known measures.

  What forms of evidence would be critical to decide on separability of cognitive 
  processes in a way that matters to drug discovery and drug development?

  Now, several were discussed, and all are relevant. We'll talk a little bit more 
  about one: more differentiation of the areas at a sort of neural substrate, 
  neural representation level was discussed--certainly relevant. The genetic covariation 
  are some of these domains--are portions of the domains inherited? Did it get 
  together in schizophrenia? Or, for that manner in normal subjects? Very relevant, 
  probably, to later intervention.

  But there was the most discussion about the issue of pharmacologic separation. 
  And there, the issue of whether a range of existing compounds, when you actually 
  intervene to try to change cognition, whether across several compounds--obviously, 
  some of these things will be very compound-specific--but we know relatively 
  little about whether the domains that we think of as separable in the statistical 
  sense, if you're measuring them at baseline in schizophrenic patients, whether 
  they--or, for that matter in normals, where the dimensions are rather parallel--whether 
  actually they'll tend to co-vary; that is, whether changes in some domains of 
  cognition will co-vary more than baseline measures, or whether they'll co-vary 
  differently.

  So, this can only come with intervention trials. It can't be done simply from 
  looking at baseline cognitive measures of even large groups of patients. It's 
  a different question for large groups of normal subjects, or large groups of 
  animals.

  [Slide.]

  Here, one obstacle to doing this was seen as the fact that the individuals who 
  tend to develop the relative cognitive paradigms--really, both at the animal 
  level and at the human level, although the specific examples that came up were 
  at the human level--don't necessarily have access to the compounds that they 
  would like to most use to differentiate, potentially, the processes. And on 
  the--because they tend to be in academic centers, focus mainly on paradigm development. 
  And industry sometimes doesn't, although they have the compounds--doesn't always 
  have ready access, through existing mechanisms to incorporation of the newest, 
  most promising cognitive paradigms.

  So there was a sense that maybe NIMH, through some process as yet unspecified, 
  might need to specifically figure out a way to encourage such sort of collaboration 
  and sharing, in a way that doesn't always exist right now.

  Then--last slide--

  [Slide.]

  --no, it's not the last, actually, but the third--we did a survey, then, of 
  the existing structure of domains that the MATRICS Neurocognition Committee 
  came up with and recommended, based on the existing factor-analytic literature, 
  which of those--this is really following up on a sort of a slide that Jim Hagan 
  used yesterday--which of those domains already have promising parallel animal 
  models to the human measures? And which did not--also beyond that--which domains 
  needed what specific kind of work to be refined for drug discovery and drug 
  development work.

  The domains that were seen as having been developed and refined the furthest 
  in relevant animal work were working memory, attention, vigilance, and speed 
  of processing. There, each of them--several prominent animal models were suggested. 
  It seems that we don't always know whether the animal models that are most prominently 
  used in those domains--and working memory was used asa particular example--actually 
  have the same sort of definition for working memory, or inclusion of sub-components 
  as the human models. For example, the relative emphasis on--in working memory--on 
  maintenance of information, versus manipulation of information differs across 
  working-memory tasks, badly--sort of slave-systems, versus executive. And there 
  are tasks in this domain at both the human and the animal level have a lot of 
  diversity in the extent to which they employ those two possibly quite different 
  sub-components. An there needs to be a more systematic look at that, and a look 
  at in to try to develop more parallelism between what's going to be used at 
  the human and the animal level if we're going to hope to get better prediction 
  from the animal models to later human tests of the same cognitive processes.

  This seemed--many things would help with this, but there was the sense that 
  one obstacle to this was that although there are interactions amongst the human 
  labs and the animal labs doing this work, it's not focused around this kind 
  of question. And more kind of directed interactions at this kind of question 
  from cognitive psychology, as we partition the processes more fully, from cognitive 
  neuroscience and from behavioral neuroscience, really needs to be encouraged.

  Again, it would be helped by some kind of formal rationale for doing this; some 
  incentive process.

  [Slide.]

  Finally, kind of coming from this, more attention needs to be paid, as we ask 
  questions like, "Okay, now we have an animal model of working memory that 
  seems to parallel, as best we can tell from a task analysis the human measures 
  that are going to be used for working memory to validate and later human trials, 
  what--is there really predictive validity from the animal early trials to the 
  human trials?" And, where there isn't, then there's got to be some tweaking 
  back and for so that there can be more assumptions about the validity of animal 
  models and human models tapping analogous processes.

  The final slide?

  [Slide.]

  Some areas are less well developed, but there's at least reasonable development, 
  like animal models for visual learning and memory. They could use a great deal 
  more development, specifically as they relate to drug screening tools, and some 
  of the demands of that kind of application.

  There, the issue of what seems to be first an intractable problem came up of 
  how do we model an animal's verbal learning? One suggestion was that, for many 
  compounds that might affect verbal learning, actually they might affect those 
  aspects of the circuit of learning and memory that are common to visual and 
  verbal learning, rather than the specific verbal ones. And there was--it's not 
  clear how strongly people believe that there really will be, at the pharmacologic 
  level, a need to develop verbal-learning-specific, content-specific drugs. If 
  there was that need, though, over time, then there was the feeling like we'd 
  have to develop better sort of left hemisphere animal models--even though those 
  wouldn't be verbal models, per se. It would have to be some other kind of precessing 
  specialization that would be left-hemisphere preferential.

  For reasoning and problem-solving, there were two issues. One, ironically, was 
  that we noted that in the selection of the MATRICS battery so far, some of the 
  tests that you'd most commonly think of as reasoning and problem-solving, or 
  high level executive function tasks--like the Wisconsin card-sorting test--were 
  actually eliminated from consideration for psychometric, and especially practice-effect 
  problems; that is, once you know how to solve the problem in the Wisconsin card 
  sort, you don't tend to--it's not the same task the second time around.

  Here, ironically, the feeling was that animal models of one critical element 
  of that--set-shifting--have actually progressed further, and are psychometrically 
  stronger and well validated than are the human models. The human models, although 
  they do exist, attempt to pull that part of the process apart. They actually--while, 
  in theory and in small study practice they exist, they haven't been validated 
  far enough to use them in clinical trials. So there's actually kind of a reverse 
  need in that domain.

  And then, finally, the issue came up about what species are--in the animal models--probably 
  to be preferred in these different domains? There was the feeling that in almost 
  all domains, you could model the cognitive components in some analogous way 
  in rats and/or mice, but probably in the area of reasoning and problem-solving 
  and the sort of higher level abstraction questions, you'd probably have to use 
  monkeys. And certain of the interesting new distinctions in declarative memory, 
  and somewhat in episodic memory, may also take monkey, rather than just rat 
  or mice work.

  So that was the summary. There were other key points, but not ones to bring 
  up at this point.

  Thank you, Keith.

  DR. GOLD: I was wondering, with regard to the pharmacological manipulations, 
  whether you discussed anything about whether you needed to see effects acutely, 
  or whether chronic effects. And it sort of feeds back into the test-re-test 
  reliability, and some of the properties of the tests.

  But how would those issues sort of be interwoven?

  DR. NUECHTERLEIN: That was amongst the questions in our long list of questions 
  that we had hoped to get to. There's recognition that it's important, but we 
  didn't end up discussing it.

  DR. GEYER: I'm wondering how much of at least the human part of the research 
  agenda implied by these needs could dovetail and capitalize on the recently 
  initiated TURNS project?

  DR. NUECHTERLEIN: There was discussion--without being presumptuous about how 
  it would work--that it might be possible to sort of do some tag-on parts to 
  using the first MATRICS cognitive battery. And when you wanted to especially 
  use one of the promising new paradigms from cognitive neuroscience and develop 
  it further psychometrically, and sort of do some initial trials of pharmacologic 
  sensitivity and pharmacologic covariation of cognitive processes, that TURNS 
  might be one way to do that.

  Again, you know, this would get into the practicalities of how much can you 
  add on--as in another session--it's easy to add on seven or eight minutes, if 
  your cognitive paradigm takes an hour, three hours--you know, a bit harder. 
  So--but that was thought of as one opportunity for immediately moving this along.

  DR. GEERTS: [Off mike.] [Inaudible.]--whole identity issue, how are you going 
  to address the issue of practice and learning with the patients. I mean, in 
  the field of Alzheimer's disease, we know that patients actually tend to learn 
  the test, so how does the process in the ongoing trial for schizophrenia being 
  addressed as this test-re-test specifics?

  DR. NUECHTERLEIN: Okay, there was some discussion about that, and it can also 
  be informed by the approach we're using in the MATRICS psychometric validation 
  study. In some domains--there's always evaluation of practice effects, but what 
  you do about practice effect sort of differs by domain.

  In some domains, the practice effects are either minimal or they're small enough 
  so that even if you got like a main practice effect, you're still in an area--and 
  you're not trying to measure, sort of learning per se, you're still in a domain 
  of psychometric measurement that you're not up against ceiling, you're not restricting 
  the range of scores and so on. And so as long as you're using a control-group 
  design, you actually don't have a psychometric problem with practice effects 
  in that situation.

  Now, there's other situations, though, where the practice effects are large, 
  especially, typically, in learning and memory, and in some reasoning and problem-solving 
  tasks, where the effects are so large you need probably parallel tasks; parallel 
  tasks, alternative forms. And that was a consideration in selecting potential 
  measures for the current MATRICS cognitive battery.

  If we move to cognitive neuroscience paradigms, where that hasn't traditionally 
  been a major issue, such parallel tasks and alternative forms would really need 
  to be developed, which is a major program of research unto itself.

  DR. JONIDES: John Jonides. Let me make a comment about developing parallel paradigms--not 
  parallel in the sense of what you were just saying, Keith, about alternative 
  forms, but parallel paradigms using human and animal models.

  I'll use spatial working memory as an example, because I know it well. The original 
  research on the ocular motor delayed response task--Jakobsen, going through 
  Golman or Keach, and more recently--used a very simply rarified version of this 
  task, where a single location is presented. The animal has to store it for a 
  period of time, and then it's probed. That works very well in the animal domain. 
  But porting that task over to humans turns out to be quite problematic. Humans, 
  clever as they are, and multi-modal as they are, will invariable use verbal 
  coding strategies to solve a problem of that sort that defeats probing the spatial 
  memory system itself. And so there is trickery tat's required in developing 
  a paradigm. And that trickery, in turn, doesn't port well back over to the animal 
  domain.

  So, that's one example of a subtlety.

  Another example is--again, using spatial [technical interruption] noticeable 
  asymmetries in the accuracy of spatial working memory, depending upon the visual 
  field in which the stimuli are presented--left versus right--probably corresponding 
  to the hemispheric specialization of the structures that are processing the 
  information, there is much less of that asymmetry, if any, that's noticeable 
  in the animal models in which these paradigms are tested.

  And so, once again, there's a subtlety there that needs to be thought about 
  seriously in developing parallel forms so that the essence of each of the tasks 
  can be brought to the fore, and serious consideration given to what variables 
  are important, and what variables really aren't important.

  DR. NUECHTERLEIN: Right. I think those are really superb examples of the dilemmas 
  that are faced, and especially, take the spatial working memory example you 
  gave, just to kind of reemphasize the key point: you really need to decide how, 
  in humans, you can keep it being a spatial working memory task and avoid the 
  cleverness of humans to redo the task on you, and do it by verbal encoding. 
  Ad as we both know, there are various strategies for doing that.

  So a question would become: once you use those other strategies to keep it a 
  spatial working memory task, has what you have done to accomplish that resulted 
  in a lack of parallelism at the pharmacologic sensitivity level. And--yes, I 
  completely agree. Very tricky question, and it has to be addressed very systematically.

  DR. MATHALON: [Off mike.] Nor should you underestimate the ability of animals 
  to use trickery--

  [Laughter.]

  --in working memory tasks. The classic delayed-response test is a very good 
  example, because of the use of mediating strategies. And actually, it's quite 
  interesting to consider the relative functionalism of mediating strategies in 
  animals and in humans, because, after all, the verbal encoding is simply a mediating 
  strategy.

  So there are parallels here which actually could be quite usefully compared.

  DR. GEYER: The issue of test-re-test reliability, I think, comes up in a lot 
  of different domains. So one question that I had that I don't think I've heard 
  as much of is: how reliable do the measures have to be to be useful for detecting 
  the kinds of changes or effects we're interested in? That is so we can make 
  assessments of test-re-test reliability and we get a number that's maybe, if 
  it's an intra-class correlation, somewhere in the range between .6 and, if we're 
  lucky, .8 or something like that.

  You know, what is a sufficiently high test-re-test reliability for the effects?

  And then I think a corollary is: what strategies are being discussed to improve 
  test-re-test reliability in the event that they are showing up in the lower 
  range of correlations?

  Dan Mathalon, from Yale?

  DR. MATHALON: The first one's the easier one. The second one I'm not sure I 
  would tackle at this point, there are such a range of strategies.

  The first one--to give you examples--when the MATRICS Narrow Cognition Committee 
  considered cognitive measures and screened them for what was really reasonable, 
  the very best cognitive measures get test-re-test reliability over, say, two 
  weeks or a month in the .7, .8 range. The ones that--even some of them from 
  really promising cognitive neuroscience distinctions, actually, when you look 
  at test-re-test reliability, some of them are reasonable, but some of them are 
  like .3, .4. That's a problem.

  You know, if you were saying intra-class correlations--yeah. Well, if you can 
  get those up to .7, .8, I don't think anybody would have a problem with it. 
  If they're--.6 isn't terrible. But, you know, intra-class correlations, as many 
  of you know, kind of penalize you for shifts in level, not just ordering, so 
  they tend to run a little lower than product moment correlations.

  But someplace in that .6 to .8 range for an intra-class; .7 to .8 would be better. 
  Things like starting to hit about .5 and lower, that would be considered a problem.

  DR. NUECHTERLEIN: One of the strategies, of course, is to increase the numbers 
  of items and measures, but I think that you raise the point--there's a limit 
  on how long a test session can be. But given the effort that's going into this 
  enterprise, I think it's worth considering building in averaging over measurement 
  occasions--as in measurement days--as one strategy of improving the reliability 
  of our baseline and follow-up assessments. So having a test session, let's say, 
  even on two consecutive days, and having that average provide the baseline against 
  which change is assessed is going to be more reliable.

  DR. MATHALON: Yes, certainly that would be a means of--you're right that you 
  grapple with the duration of the test. And industry has especially informed 
  us--there's sort of a compromise going on, you know, clinical trial specialists 
  would tend to--we started off with their preference for a cognitive battery 
  would be about 30 minutes. And neuropsychologists said, "How about four 
  hours?"

  [Laughter.]

  And we thought, "Huh. There's a little gap there." And so we ended 
  up with 90 minutes as sort of the upper end for measuring seven different cognitive 
  domains, with reliability and validity.

  So you've got sort of 10 minutes, maybe, per domain--which, for some existing 
  technologies, that's really short. And you might have to--if you can't do something 
  to compress them, make fewer trials adequate, then you might have to use strategies 
  like you're talking about: multiple baseline sessions or something like that; 
  or use them earlier in drug development for screening, but not in the major 
  Stage 3 clinical trial.

  DR. GEYER: I think your point here, as to one of your research agenda items, 
  was to put the psychometricians together with the cognitive neuroscientists--and 
  we don't need to map the details. Today, I think, we have a research agenda 
  that we're trying to define which would accomplish, I think, a lot of what this 
  discussion is.

  Unless it's critical, John, I really would like to keep on schedule. Is that 
  all right?

  Thank you, Keith.

  John Jonides will present now, out of order, the group: "Emerging Frontiers 
  in Behavioral Measures of Cognition" that Jonathan Cohen and he co-led. 
  John?

  

Emerging Frontiers in Behavioral Measures of Cognition

  DR. JONIDES: I'll report that our group also had, as Cam put it, quite a nice 
  spirited discussion that I think is best characterized with a funnel-like quality. 
  We started off with fairly broad strokes about the nature of cognitive tasks, 
  and impediments in the implementation of cognitive tasks for the study of elementary 
  cognitive processes. And we gradually funneled down. And one of the things that 
  helped us funnel down was a nice metaphor, which we altered a bit later on, 
  and that's the kind of sub-title of this: analyzing the cylinders in a combination 
  lock.

  So let me just introduce you to the metaphor, because I think it's a helpful 
  metaphor to understand the approach that we're taking, before I have the rubber 
  hit the road and tell you some details.

  How many of you have seen the kind of bicycle lock, with a chain attached to 
  it, that has rotating cylinders on it. And you can set the cylinders. And when 
  you have all four or five or them set--depending upon how many digits there 
  are in the correct solution to this combination--the chain pulls apart, and 
  then you can unlock your bicycle.

  This combination lock is better to have in mind than the one that you might 
  have on your locker at the gym, where you actually have to serially put in the 
  correct numbers, because these cylinders can be set in any order.

  Imagine that each of these cylinders is component cognitive process, and what 
  you're trying to do is to unlock the entire lock; that is, you're trying to 
  figure out the complex of cognitive processing, and your analysis technique 
  is by setting each of the cylinders so that it's in its appropriate position.

  Now, you have to kind of modify your image of this lock, because the way I've 
  presented it to you, it's an all-or-none solution; any one cylinder doesn't 
  get you any farther to the solution than any cylinder does. So, imagine that 
  each time you set one of the cylinders correctly, the lock opens up just a little 
  bit more, so that when you set the final cylinder, it will finally separate.

  Now, that's an interesting image to have in mind in considering elementary cognitive 
  processes that might contribute to an understanding of complex cognition, and 
  the deficits in complex cognition that attend pathologies like schizophrenia.

  So now, in fact, our research recommendations turn around a number of domains 
  of study, and we actually are making quite concrete proposals about what some 
  of these domains should be; and, in fact, if we wanted to we could, I think, 
  devolve this discussion--if we had the time to do it--into a discussion of individual 
  tasks that represent each of these domains quite well.

  [Slide.]

  There are--pardon my deviation from the three goals--there are here four goals. 
  They have to do with identifying perceptual processing, executive processing, 
  memory processes, and then what I'm kind of loosely referring to as "higher 
  level cognition."

  So let's go through each of these in turn, just a bit, because it's--and I think 
  once we identify what the processes are, many of you will know what some of 
  the appropriate tasks are to delve into these.

  [Slide.]

  So what we're doing here is trying to lay out a taxonomy of cognitive processes 
  that derives from years and years of basic cognitive psychology, before cognitive 
  neuroscience became a field, and then since cognitive neuroscience became a 
  field--using the study of lesioned patients and neuroimaging techniques of various 
  sorts; the development of tasks that have come under study once those fields 
  have matured, as well.

  [Slide.]

  So this is an enterprise quite different from the original MATRICS enterprise--as 
  I understand it. I wasn't part of that development routine. This is not a psychometric 
  enterprise, where you try to find tasks that are statistically independent of 
  one another and come up with a categorization of, say, seven cognitive abilities 
  that you're interested in investigating, and then finding tasks that measure 
  those cognitive abilities. Rather, this is a more principled analysis, based 
  on theoretical notions of how the cognitive system works, and then trying to 
  unpack what some of the elementary processes are that feed into this.

  This same kind of discussion, I must say, came up in the neuroimaging group, 
  which I participated in yesterday morning, that Cam reported on earlier, and 
  I think there is--if there's not consensus, there is at least widespread agreement 
  about the taxonomy that's being offered here--although at the edges, of course, 
  there are going to be important discussions about the right way to define some 
  of these processes.

  [Slide.]

  Certainly, selective attention, in the presence of distracting stimuli in the 
  perceptual world, is an important component to this. And the selective attention 
  need not be limited, of course, to the typical paradigm that's used to study 
  it in the visual world; it could have to do with auditory distraction, or tactile 
  stimulation, gustatory stimulation, as well. There is a general problem of selecting 
  out important objects for perceptual processing, and leaving behind other objects 
  that are irrelevant to the task at hand. And that's what's meant by this first 
  elementary processing operation.

  [Slide.]

  There was also a discussion--although, I must say, brief--in our group bout 
  even more elementary perceptual operations; ones that are earlier in the visual 
  processing stream: things like masking and feature-finding processes, which 
  might also be interesting domains of study, where there might be important differences, 
  as a function of pharmacological manipulations, as a function of certain disease 
  states, as well. And certainly those are worth looking into.

  I think there wasn't concentrated discussion of that, because I think there 
  was a general feeling that the higher level cognitive processes--including other 
  perceptual ones--might be deserving of more concentration.

  [Slide.]

  There is, of course, a great deal of attention being paid to executive processes. 
  And, as I said yesterday in my formal talk, there isn't any consensual agreement 
  about what constitutes executive processing--although there is a widely cited 
  list of processes that comes up in any discussion of executive processing. Certainly, 
  shifting of attention from one representation in memory to another representation, 
  or shifting from one task to another task that you're performing, or shifting 
  from one attribute that you're concentrating on to another attribute that you're 
  concentrating on. In the cognitive literature, this is typically identified 
  as "task switching" or "object switching." Certainly, sequencing 
  and planning operations, where you have to lay out some event from its beginning 
  to its end, and sequence the middle operations in the appropriate order.

  Resolving conflict between representations and responses--the last example that 
  I gave in my formal talk yesterday was about resolving conflict among representations 
  in working memory. But one can imagine resolving conflict in the sense of the 
  Stroop task, or resolving conflict between one decision that you want to implement 
  and another decision that you want to implement.

  So here's the case where resolving conflict is a kind of catch-all phase for 
  a number of underlying processes, and there has to be some discussion of the 
  separability of those--as well as resolving conflict between alternative responses; 
  the tendency to make one response, in the face of a situation where another 
  response might be appropriate.

  [Slide.]

  And, again, processing information under conditions of distraction, where, in 
  this case I'm not talking about perceptual distraction, as I was with perceptual 
  processes, but where there are other thoughts in mind as you're trying to move 
  toward some final thought that you have. There are other distracting thoughts 
  in mind, and you need to hold those down; you need to inhibit those.

  I'm trying studiously to avoid the use of the word "inhibition," because 
  inhibition, in the cognitive neuroscience literature is a theory-laden term. 
  It's a term referring to certain processes that might be in play, all in the 
  service of resolving conflict. And that's why I refer to "resolving conflict" 
  as the appropriate term.

  [Slide.]

  There are memory processes--certainly including working memory, but also including 
  long-term memory--that were the subject of discussion. These have come up before, 
  so there isn't a great deal of need to elaborate them. Both Cam and Keith brought 
  these up.

  Storage of information in working memory, independent of any other kinds of 
  operations that are applied to the contents of working memory; and then manipulation 
  of information in working memory, where you're not only storing data, but you're 
  manipulating those data in the service of some higher cognitive task.

  And then there are declarative memory processes--the ones that engage the medial 
  temporal lobe system; what are typically thought of as hippocampal memory tasks, 
  surely there are many excellent examples of such tasks. And, again, as I illustrated 
  in my second example yesterday, there was a possible partitioning among medial 
  temporal lobe tasks, depending upon the processes that are engaged--encoding 
  processes, or retrieval processes, or storage and transformation processes, 
  as well.

  [Slide.]

  And then there are the higher level cognitive processes I think probably the 
  least well-defined of these is language processing. It's not clear whether the 
  appropriate concentration here deserves to be on lexical access or on syntactic 
  processing; you know, interpreting center-imbedded sentences, versus right-branching 
  sentences; or on semantic processes, doing language processing for meaning. 
  And I think there needs to be a great deal of concentration on what aspects 
  of language processing need to be further studied in order to identify some 
  elementary aspects of those.

  [Slide.]

  Two other aspects of higher level cognitive that might be appropriate domains 
  for study are what I brought up as kind of source attribution but was, I think, 
  appropriately generalized as "meta-cognitive processes" as well; that 
  is, monitoring your own ongoing cognition, and the extent to which that might 
  differ as a function of pathology. And also changes in reinforcement-based learning 
  that engages the dopamine system, clearly an important system to think about 
  with diseases such as schizophrenia.

  [Slide.]

  So what are the requirements in order to implement a research program that looks 
  at these four domains of study? And what are the obstacles to doing that? Those 
  are the subject of the last two slides that I have.

  Of course, we need to develop criteria for task acceptability. And there are 
  physical constraints in doing that; the 10-minute constraint that Keith just 
  alluded to as part of putting together a neuropsychological battery. But there 
  are other operational constraints as well. I mean, you need to take a look at 
  reliability issues, and validity issues and so forth. And I'll get to those 
  in just a moment.

  There are issues having to do with duration of the test, as I said; the stability 
  of the test. The effect size of the test: in many cognitive neuroscience laboratories 
  out of 30-millisecond effects. In fact, in my laboratory, I've gone dancing 
  in the streets over a 70-millisecond effect, because that is good enough for 
  me to be able to do the kind of experimental work I need to do.

  But a 70-millisecond effect may not be the kind of effect size that is testable 
  in a 10-minute session; that has the kind of reliability--that meets the kind 
  of reliability criteria that you would like to meet, and so forth. And so there 
  needs to be some serious consideration of what can be tweaked in experimental 
  paradigms that will make the effects not only robust, but also sizable in magnitude, 
  and will show up in every individual that is being tested--or at least nearly 
  every individual in whom you're doing the test.

  I think one valuable technique that's been underused in helping to select tests 
  is meta-analysis. There is, by now, a very rich cognitive neuroscientific literature 
  that makes use of tests in all of the domains that I showed you on the previous 
  slide. Let's mine that literature and see whether we can pick out the tests 
  that have robustness, that not only appear frequently in the literature, but 
  that do work, in the sense of predictively--have good predictive validity for 
  other kinds of cognitive processes as well.

  Of course, you need to create a process to validate and test the reliability 
  of these tests. And in the next slide I'll tell you about one of the obstacles 
  to do that. You need to develop norms for the tests, in the same sense that 
  neuropsychology, over the course of many years, has normed a wide variety of 
  instruments so you can find out what standardized performance is.

  You need to establish test performance in not only patient groups such as schizophrenics, 
  but also sub-groups within those patient groups that might be of interest, as 
  well, because there may be important performance differences in the sub-groups.

  I think it's important--this is a point that I could have brought up as Cam 
  was making his presentation, but is well worth bringing up now, as well--I think 
  it's important to use imaging data--neuroimaging data--ERP data, and functional 
  MRI and PET data--as an important lever to help identify what tasks are separable 
  by way of brain mechanisms that underlie them, and what tasks are closely associated 
  by way of brain mechanisms, because that might help give you a menu--for example, 
  if you found that three or four tests were closely associated in the underlying 
  brain mechanism that mediated them, that might give you a kind of cafeteria 
  of tests among which you can choose, based on other criteria that you might 
  want to apply to those tests, like duration of administration, and stability, 
  and so forth. So I think imaging data can feed well into that. And, again, meta-analysis, 
  I think, is an important tool that might help make some headway there.

  And then--the point that I made before--it's not easy coming up with parallel 
  models in animals. I gave a couple of illustrations before, so I'm not going 
  to elaborate on that now.

  [Slide.]

  There are a number of obstacles to engage this research program: identifying 
  individual domains and operationalizing them.

  First of all, there isn't any generally agreed upon set of processes, in spite 
  of the fact that I came up with slide one. There is a matter of open discussion 
  here. And I think the discussion will largely occur around the edges, because 
  I think in the center of the mass here there is a good deal of agreement among 
  cognitive neuroscientists. But, nonetheless, the discussion is worth having 
  around the edges.

  And I think this discussion can be guided quite profitably by computational 
  models that have now been developed about some of the underlying elementary 
  processes.

  We need a consensus on the choice of tasks, once we've agreed upon the set of 
  processes. And, once again here, I think the literature is going to be an important 
  aid here, because there has been a good deal of replicated use of a common set 
  of tasks--albeit, with important tweaking from one to another.

  It's important, I think, to choose multiple tasks for any single construct that's 
  being investigated in order to get kind of converging measures--in the sense 
  of structural equation modeling, getting manifest variables to identify some 
  underlying latent variable.

  We need to, of course, assess how these tasks make contact with functional outcomes; 
  that is, the validity issue. And here's an important kind of logistical point: 
  I think Cam brought up this point and Keith brought up this point, and I just 
  want to underscore it again--there's a need for multidisciplinary participation 
  in the behavioral studies, of validity and reliability, to help develop these 
  underlying component tasks. No single laboratory will probably want to do this 
  heavy lifting. And, in fact, there is great value in having multiple laboratories 
  do this, and also having multiple--the participation of multiple disciplines: 
  certainly, cognitive neuroscientists; certainly, clinicians who are close t 
  the end product by way of pathology; and, of course, certainly those expert 
  in psychometrics to be able to develop tests that have the appropriate psychometric 
  property.

  The task creation, of course, ha to be guided by potential drug intervention 
  effects. You wouldn't want to choose effects that are immutable to drug interventions. 
  They're not going to serve the purpose here. I think large-scale imaging studies 
  might be necessary to establish individual differences in brain activations 
  for these tasks in order to show that there is mutability--again, by way of 
  the brain mechanisms that are involved.

  There are task parameter issues, by way of standardization, that are going to 
  involve locking some people in a room and not giving them food or water until 
  they've come out with a consensus of how the tasks ought to be organized. The 
  room probably needn't be locked for more than a couple of hours.

  And we need to determine where, in the compound development process, the tasks 
  should be used? Should they be used early in the development process, far away 
  from the functional outcomes that are of later concern? Or can the tasks be 
  developed close enough to the functional outcomes at the end-stage of drug development 
  so that they might actually be important behavioral markers late in the outcome.

  I think that kind of summarizes--at least from my perspective--a synthesis of 
  the discussion we had yesterday.

  DR. GEYER: Thank you, John. That was a very concrete and useful review, I think.

  We have some time for questions. Diane?

  DR. BARCH: John--wonderful.

  One point that came up in the meeting that I just kind of wanted to reiterate 
  was--I think Cam put it nicely as--that part of this process needs to be building 
  a bridge from the tasks that we're currently using to the new tasks, and so 
  that one sort of piece of the research agenda will need to be some sort of comparison 
  along a number of different levels. And we'll need to have some discussion about 
  the right way to do that. It might not be one task to one task; it might be--given 
  the lock analogy--it might be a constellation of tasks to a different task, 
  but to actually try to do some comparison to let us know--"Here's another 
  piece of the reason why we should be really moving forward, because we get better 
  discrimination, better sensitivity, better drug effects, so that we can help 
  the field sort of see how this might be a beneficial process."

  DR. JONIDES: Yes, thanks very much for emphasizing that, Deanna. I think one 
  important component of that process can be a componential analysis of the tasks, 
  for example, in the beta battery, to find out what elementary processes might 
  participate in those tasks, and whether there is a kind of common set that either 
  converges or, in important ways, might diverge from the set that I included 
  on my first slide.

  DR. ROBBINS: Trevor Robbins.

  That's a very nice taxonomy of the discussion we had yesterday. The only thing 
  I felt was left out a little bit was a discussion of the state-dependent processes 
  which affect, for example, executive function, like motivation. And you mentioned 
  reinforcement learning. That's actually a fairly low-level process. I mean, 
  obviously, the elaboration of reinforcement learning, with motivation, and goal 
  representation, and how that interacts with the cognitive modules in frontal 
  cortex is also extremely important.

  DR. JONIDES: Yes, Thank you, Trevor. I didn't mean to underplay the value of--the 
  discussion we had about motivational variables. I was hoping that Deanna would 
  take that up, because her breakout session concentrated on that issue, among 
  others.

  [Pause.]

  DR. GEYER: Other comments? It strikes me that there's a considerable amount 
  of integration, which you pointed out. I like the image of locking people in 
  a room to reach standardized task parameters. [Laughs.]

  DR. JONIDES: [Off mike.] I have some nominations of people I'd like in the room, 
  by the way.

  [Laughter.]

  DR. GEYER: In some other similar--especially in the toxicology arena--rather 
  than always mandate specific tasks, or equipment, or parameters, what has been 
  used is to mandate certain known, positive effect results, such that different 
  procedures establishing the same--addressing the same construct can be validated 
  by a known pharmacological effect or, in that toxicological effect.

  I'm not sure how much you want to take the standardization of the specifics 
  of the tests, versus the standardization of known effects in the positive control 
  sense.

  DR. JONIDES: Yes, I think that's an important comment. Standardizing the tests, 
  I think, isn't nearly as important a job as standardizing the constructs. And 
  so if there are multiple tasks that feed importantly into understanding a construct, 
  that behaves similarly, for example, with pharmacological intervention, that's 
  a very important thing to discover; and that behave, for example, similarly 
  when you subject all of them to the same experimental variable--like practice, 
  or like task-demand, or whatever--that's another very important consideration.

  DR. TAYLOR: Steve Taylor.

  John, I'm wondering if there was any discussion, or kind of a consensus that 
  would emerge from your breakout group as to how one knows we have gotten the 
  right subcomponents; that we've actually truly carved cognition at its joints. 
  And you mentioned a few things like meta-analysis--and I assume you meant also 
  some factor-analytic studies, or imaging studies. I just wondered if there was 
  any kind of feeling about that--to know when we've got the right carves?

  DR. JONIDES: We actually did have a discussion about that. I can't honestly 
  remember, now, whether it was in the session that Jonathan and I led, or in 
  Cam's session, because I think it came up in both contexts.

  And I think the answer is that 40 years ago we would have carved cognition differently 
  at its joints than we do today. Forty years from now we will carve cognition 
  differently than we do today--possibly. If I could predict that with certainty, 
  I would carve it now as we would in 40 years.

  And so I think it's time for the rubber to hit the road. And right now I think 
  there is good evidence from the separability of constructs by way of brain mechanisms, 
  and there's good evidence by way of the separability of constructs through double-dissociation 
  work behaviorally. And there's also good evidence about the separability of 
  constructs by way of lesion evidence--patients who have lesions. So that the 
  taxonomy I laid out in the first slide--as I say--is not going to be agreed 
  upon by everybody, but it probably captures a fair amount about what comes from 
  those various dissociations that appear in the literature.

  DR. TAYLOR: If I could follow up--because I noticed one thing that wasn't on 
  there was psychomotor speed. And that's certainly been--impaired reaction time, 
  or slowed reaction time, has been referred to as the North Start of schizophrenia 
  research several decades ago. And that's still, of course, true, that it's an 
  invariable finding. And I know also psychomotor processing speed has come out 
  as one of the factors in factor-analytic studies of, I believe, the intelligence 
  testing--just general cognition.

  Any--would you like to comment on the psychomotor speed as one of the possible 
  domains that you didn't include? Or that would be included?

  DR. JONIDES: Absolutely. I actually think psychomotor speed is well worth considering 
  as an elementary--you know, to call it a "process"--I don't know if 
  I would call it a process, but it seems to be a kind of skill, or a competence 
  that might run across a number of elementary processes. For example, work by 
  Tim Salthouse has shown that individual differences in working memory--at least 
  in some contexts, especially as a function of normal aging--can all be accounted 
  for by individual differences in processing speed. And so maybe that allows 
  you not to even consider the construct of working memory.

  Well, that doesn't seem appropriate to me, because working memory is an important 
  building block for any complex cognition.

  So it seems to me--at least on first pass--that considering processing speed 
  as an important skill or competence might be in some ways orthogonal to considering 
  some of these other specific cognitive domains. But--you know--who knows?

  DR. GEYER: Time for one more.

  Cam?

  DR. CARTER: Yes--you know, I just wanted to get back to this issue of the fact 
  that there probably are multiple tasks that can be used to measure a construct, 
  and one of the challenges will be using some criteria, such as the psychometric 
  properties of those tasks to distinguish between them. And I think there's the 
  potential for a very helpful integration of a sort of psychometric analysis 
  and componential analysis. My guess is that for some tasks--for instance, you 
  know, the Eurex and Flanker task is used to measure sort of conflict resolution. 
  It's probably--and a surface analysis, not a very reliable task. It's one reason 
  why it's probably not so popular in functional imaging circles.

  But it also might be that a componential analysis that pulled apart things like 
  repetition effects from Flanker effects might actually allow you to have a more 
  stable measure of the construct of interest.

  So I think, again, it's sort of an argument for a collaboration between psychometricians 
  and cognitive neuroscientists.

  DR. GEYER: John?

  DR. JONIDES: I might say, just as an addendum to Cam's comment, that merging 
  psychometricians and cognitive neuroscientists in an enterprise to develop a 
  more stable set of tasks to measure individual constructs that are somehow agreed 
  upon is going to have benefits, not only for the study of schizophrenia and 
  drug interventions, it's going to have enormous benefits for cognitive neuroscience, 
  because it's going to help us develop a set of tasks that I think will then 
  proliferate in the literature for even further analysis, that might get us to 
  the kind of analysis of components that we'll know about 40 years from now that 
  we don't know about now.

  DR. GEYER: Thank you very much, John. Have a good trip home.

  The next breakout group was led by Tony Grace and Lisa Gold, and is entitled 
  "Developing Predictive Models and Establishing a Preclinical Trials Network 
  for Assessing Treatment Effects on Cognition in Animals."

  

Developing Predictive Models and Establishing a Preclinical Trials Network for Assessing Treatment Effects on Cognition in Animals

  DR. GRACE: Thanks. As was mentioned by the other groups, we had a rather lively 
  discussion in our group that ranged everything from acute and clinical drug 
  effects, to possible interactions of the cognitive-enhancing drugs with other 
  drugs.

  In order to try to distinguish the goals of our group, of assessing animal models, 
  from things that were done in Group Two, we decided not to focus so much specifically 
  on the cognitive domains, but rather on--especially the individual tests associated 
  with the cognitive domains--but more on general issues that we think are more 
  needed in order to drive the field forward, and in order to try to assess what 
  can take us to the next stage of looking at animal models and how they can inform 
  us with respect to the clinical picture.

  Ne thing that was obvious through this discussion is that we really need to 
  advance the state of the clinical-basic interaction. We need to more from the 
  clinicians exactly what the pathophysiology is: what are the changes that we 
  can model, and then to go and look at the models and see how well they can predict 
  what we see in the clinical picture.

  Now, one of the things that we came across that we thought would be very valuable 
  is the possibility of implementing a database to test the data and drug effects 
  that contain essential information that's going to be relevant for looking at 
  these cognitive domains.

  One thing that became clear is that, in talking with the pharmaceutical company 
  representatives, is a number of them had mentioned that as soon as the cognitive 
  domains came out, everybody went through--got their tests and tried to match 
  the cognitive domains, ran through the standard screen of anti-psychotic drugs, 
  as well as different types of prototypical agents, in order to try to come up 
  with exactly what was available, and how these tests were affected by these 
  drugs. And this seemed to be--to us, and to them--a big duplication of effort, 
  and that we could manage this much more efficiently by creating a database that's 
  going to share this information.

  So, the pharmaceutical companies--and we have this written down, and I think 
  we'll try to get them to sign something afterwards--

  [Laughter.]

  --have volunteered to give us this information in helping us to set up this 
  database, both with respect to the tests that were used, the results with the 
  anti-psychotic drugs, as well as some of the prototypical non-proprietary agents; 
  the classic nicotinic agonists and antagonists; the D-1 agonists and antagonists--and 
  try to understand from that how they affect these tests and to help us evaluate 
  which are going to be the best models to go by.

  [Slide.]

  What we would want to do is to avoid this duplication of effort, to have this 
  non-proprietary information placed in a form of a general-access database; something 
  that's accessible, not only to pharmaceutical companies working on drug development, 
  but also academics that are working on trying to understand what type of test 
  are going to be most effective for evaluating drugs, and possibly to identify 
  what targets the companies may use in order to better address these types of 
  issues.

  The database is going to list the drugs that have been tested, as well as pharmacokinetic 
  data and receptor-specificity data, so that we can try to isolate what components 
  of these drugs are going to be the most essential in order for getting the types 
  of information that we need.

  [Slide.]

  In addition, there are certain types of proprietary drugs that aren't going 
  to be going into development, that some companies have suggested could be made 
  available to academics, along with pharmacokinetic and receptor-specificity 
  data, that are going to enable us to drive this field forward. There are certainly 
  drugs that have come out that have failed in tox tests, but that are still highly 
  specific, and are going to be very valuable for the basic scientist, because 
  of their selectivity and specificity, in order to try to understand how they're 
  going to affect these different cognitive domains.

  [Slide.]

  So what we hope is that we can get an initiative to construct and implement 
  this database in a manner that's easily accessible, easily searchable by an 
  individual or organization, and in this way more efficiently advance the field 
  of drug development in ameliorating cognitive dysfunction in

  schizophrenia.

  [Slide.]

  In order to do this, of course, we're going to need to do more than just have 
  a verbal agreement and plug it into a standard Excel spreadsheet. What we need 
  to do is to get some kind of a consensus development meeting in order to define 
  what we need in this type of database, how it's going to be set up; use a lot 
  of the information technology that's been developed at NIH and NIMH in order 
  to get this into the most useable form possible for driving the field.

  [Slide.]

  The second thing is that what we need to do is use some of the data gleaned 
  from the clinical studies--especially imaging studies, infarct and brain-damage 
  studies--to derive some of the animal models that can be mapped onto specific 
  cognitive domains of the schizophrenia pathophysiology. And just like with Group 
  Two, what we found is what we need to do is define what type of models we need.

  And for this there are a couple different approaches that you can use. You can 
  either use rather specific types of manipulations that you know what transmitter 
  system is being manipulated, hat brain system is being manipulated.

  In order to do this, what you would get is something that is going to give you 
  a lot of information with regard to mechanisms and transmitter systems. The 
  other possibility is to look at things that are more related to developmental 
  disruption models. Developmental disruption models have a lot of construct validity. 
  We think we may actually be modeling what's happening during the etiology of 
  schizophrenia based on some of the information such as what's happening early 
  on in development with respect to insult such as virus, malnutrition, and changing 
  the incidence of schizophrenia.

  The problem is is even though these developmental models may be more accurate 
  with respect to construct validity, they really leave us in a hole with respect 
  to trying to figure out what the exact mechanism of the changes are.

  So what we want to do is use approaches where we can understand the mechanisms 
  underlying some of these selective changes; for example, lesioning the dopamine 
  system within the prefrontal cortex, and what kind of manipulations can overcome 
  the working-memory deficits that are induced; and then to take this selective 
  information--which may be etiologically valid, or have a lot of construct validity, 
  but certainly a lot of face validity--plug this into the etiological models, 
  and then determine whether what we have with respect to specific changes is 
  something that can be paralleled in an etiological model.

  And with this type of give-and-take between the clinical and the two model systems, 
  what we hope to do is to get this to a stage where we can make rather strong 
  predictions.

  [Slide.]

  The obstacle, of course is going to be coming up with the means to evaluate 
  the predictive validity. Are the etiological models going to be something that 
  gives us predictive validity? Is this something that we're going to get from 
  the more specific type of face validity or constructs that we change here and 
  try to develop some type of an interaction in order to come up with what are 
  going to be the most effective models for individual domains, and how well this 
  is going to plug into a system that we think may be a better etiological model 
  of schizophrenia.

  [Slide.]

  The other is--and this is the Holy Grail of our group--to look at devising a 
  means that's going to effectively implement a preclinical trial's network to 
  facilitate this goal. Now, what this means is coming up with a means where we 
  can interact in a way that's going to give us information from a number of different 
  groups in a coordinated fashion--sort of a basic-science type of matrix--in 
  order to define what tests are going to be the most effective to use, and the 
  best way to implement them.

  One thing that a number of pharmaceutical companies do, and that we recognize 
  as a very valid approach, is to look at a hierarchy of drug screens; things 
  that can be used in order to screen very basic information with respect to whether 
  drugs are going to help cognition, then selectively screen them according to 
  which general domains and which specific domains are going to be most effectively 
  addressed by these different types of drug treatments.

  [Slide.]

  Now, because of the complexity, such approaches are likely not going to be amenable 
  to high throughput screening; at least not at this stage. What we're going to 
  have to do is, rather, a long slow process through the academic institutions, 
  collaborating with pharmaceutical companies to define the tests, and then perhaps 
  take that to the next stage to define markers that may be more amenable to high 
  throughput screening, so that we can get a way to more efficiently get at drugs 
  that are going to be effective for treating the different cognitive domains.

  [Slide.]

  And, of course, a major obstacle to achieving this is that we are going to need 
  a preclinical type of MATRICS interaction in order to most efficiently implement 
  these goals. A lot of discussion was made about Alzheimer's and the fact that 
  what we're doing is taking drugs that are effective in Alzheimer's and plugging 
  them into the schizophrenia model.

  This may or may not be accurate. Because, of course, the pathology in Alzheimer's 
  is very different than what you see in schizophrenia. You have rather extensive 
  cortical cell loss in Alzheimer's before you start getting a lot of the dementia. 
  In schizophrenia, earlier on, you're not even going to have a lot of structural 
  damage to the extent that you have in Alzheimer's when you start to get the 
  deficits that you see in schizophrenia. It's obvious that in schizophrenia there 
  are some deficits in cognition that are present very early on in childhood, 
  and they're stable throughout the course of the illness. There are other types 
  of deficits that seem to be more prominent whenever the individual approaches 
  and goes through the first break. And it's likely that these types of pathologies 
  are represented by very different mechanisms, and are going to like respond 
  to very different types of treatments.

  We also recognize that whenever we start to use some of these cognitive-enhancing 
  drugs, we're going to be using them in patients that are treated with anti-psychotic 
  drugs. And, of course, there's always the potential for interference. Just as 
  a very general type of example, amphetamine is going to improve attentional 
  focus. But, of course, if you give a catecholamine blocker, you're going to 
  prevent the effects of amphetamine on a lot of the attentional focus.

  There are likely going to be a lot of more specific and selective types of interactions 
  with cognitive things that we develop; things that are going to be effective 
  in a non-treated patient, may be interfered with whenever we start to look at 
  it under the condition of a drug-treated patient.

  However, we also know that there is a very large battery of drugs that we can 
  use. Certainly, something like Salperide is going to give us a very different 
  type of interaction then Eripiprosol or Clozapine. So testing these types of 
  cognitive enhancers, in animals, in the presence of chronic anti-psychotic drug 
  treatment, is going to give us some indication that we may be able to translate 
  into the clinic about which types of drug combinations are going to be most 
  effective for addressing these types of issues.

  [Slide.]

  And this is a basic summary of the information that we went over in order to 
  try to identify the things that we need to drive the field.

  As I mentioned, the most critical part is understanding better the clinical 
  pathophysiology of schizophrenia so that we know what to model, and then being 
  able to take these models and make very strong predictions about what we should 
  see in the clinical picture. And, from there, apply it to the development of 
  these types of cognition-enhancing drugs.

  Thank you.

  [Applause.]

  DR. GEYER: Thank you, Tony.

  DR. MARKOU: Can we put this--my name is Athena Markou, from the Scripps Research 
  Institute.

  Can we put the second slide up, because I think it will be easier for people 
  to understand?

  Yes--about these two different types of models that you have there, I think 
  that, in my opinion, this distinction is rather artificial, because I don't 
  really see much difference between-- etiologically--between developmental models 
  and, let's say, pharmacological models or lesion models, in the sense that even 
  a pharmacological model can help us assess etiological processes.

  And the construct validity comes more from the measure--from what we measure--and 
  how that may map on a cognitive domain. So, yes, you can say that pharmacological 
  manipulations are different than lesion manipulations different than developmental 
  manipulations, but I don't see them as being so far distinct as they're presented 
  there. I think all of them can assess aspects of etiology.

  DR. GRACE: That's certainly true, but there are certain advantages and disadvantages 
  of each model. For example, whenever you're talking about an etiological model--like 
  a ventro-hippocampal lesion that's in a developing animal, a developmental neurotoxin 
  given during gestation, extra radiation given during gestation--when you look 
  at the adult animal, what you get are a very large number of subtle pathological 
  changes. You have changes in dopamine in the frontal cortex, you get cortical 
  thinning, you get changes in dopamine dynamic sub-cortically, you get changes 
  in the migdela volume. You get changes in hippocampal volume.

  Trying to identify which of these are going to be responsible for working memory; 
  which of these are going to be responsible for social interactions is something 
  that's going to be very difficult to do--although the model itself may have 
  a lot more resemblance to the process that occurs in schizophrenia as a result 
  of this early intervention.

  On the other hand, whenever we have a very selective model--for example, lesioning 
  the dopamine within the prefrontal cortex and seeing how it affects working 
  memory, we have a very specific site; we have a very specific mechanism. We 
  can identify an intervention in order to reverse the deficit that's being introduced 
  by the specific lesion.

  So, mechanistically, systems-oriented, we have a very precise deficit that we 
  can overcome, that may or may not be what is occurring in a schizophrenia brain 
  as a result of genetic and developmental changes that occur early on.

  And what I'm proposing--as we talked in the session--is a give-and-take between 
  the two; looking at specific mechanisms to identify cognitive domains, plugging 
  them back into the etiological model to see whether they make sense in a model, 
  and as a way of testing to the model to see whether it's going to accurately 
  represent deficits within the cognitive domain.

  DR. MARKOU: I agree with what you say. I think a part of the confusion comes 
  from something that we have to make very clear, and which we discussed in our 
  group, which is that a model consists both of the manipulation and the dependent 
  measure. So you are making arguments about both aspects, but we have to make 
  that clear: that the model consists of both.

  And another thing that we discussed in our group, and I want to bring up, is 
  that we discussed whether we should be trying the effects of putative cognitive 
  enhances or anti-psychotics on perturbed animals that a manipulation has been 
  made, or in normal animals. And that's something that we didn't really have 
  the answer, but I think it's an important issue to bring up: that it might be, 
  in some cases, that tests will be able to give us answers in normal animals, 
  and in other cases, where it will be important that we do do one of these supposedly 
  ecological manipulations to get to some of the answers.

  DR. GRACE: Yes, thank you for bringing that up. That really is a critical issue. 
  Cognitive enhancers may have very little effect whenever you're looking at a 
  normal animal, but have very profound effect when you're looking at a disturbed 
  animal. I mean, a classic example would be something like anti-depressants. 
  Anti-depressants don't have strong effects in normal people, but when people 
  are very depressed, you have a very strong normalizing effect.

  And in trying to determine whether we need to test this in a normal animal, 
  or in an animal with specific disruptions is really going to be critical in 
  evaluating the efficacy of these drugs.

  DR. GEYER: A parallel thing, of course, occurs in humans. And the same question, 
  I think, will need to be addressed in the human proof-of-concept studies. And 
  there's a lot of opportunity for translational benefit by having those two approaches 
  and species studies interact.

  Lisa Gold?

  DR. GOLD: My question isn't so much for Tony for some of the people involved 
  in the MATRICS and the TURNS effort.

  And one of the things we talked about that's absolutely key for the translational 
  piece is that we have feedback from the clinic to the animal models, and vice 
  versa.

  And so I was just wondering, within the TURNS mechanism, is there going to be 
  the opportunity, ultimately, to take whatever compounds have gone into the treatment 
  units, and then test them in the animal models in whatever battery evolves, 
  regardless of whether the outcome is a positive or a negative, or some mix of 
  that, in the clinic.

  And I don't know whether that is explicitly not going to be possible, or whether 
  there might be the opportunity to do that. But, at the end of the day, the difficulty 
  for the development of the animal models is, of course, right now we have no 
  positive controls; no agents that we know work, or we know don't work in certain 
  aspects of the clinical battery that's being proposed.

  DR. GEYER: I think that's a very important point, Lisa--as one who will continue 
  from the MATRICS, and also be involved in the TURNS program, I can address the 
  hope, at least, that as the TURNS program matures, and begins to generate data, 
  as was recognized and acknowledged yesterday, most of the initial tests of compounds 
  may well fail. The odds are in favor--more in favor of initial failures than 
  initial successes. But, certainly, it will be extremely advantageous to all 
  of us, both preclinical and clinical, as soon as we can identify any positive 
  effects--because, as you point out, normally, in animal predicative modeling 
  and screening tests, one likes to have a positive control.

  We don't know, if we get the right answer, how will we recognize it? So that's 
  something that just has to evolve. The science is young, and the purpose here 
  is to advance that.

  Trevor Robbins?

  DR. ROBBINS: Yes, I just wanted to pick up this interesting issue of cognitive 
  enhancers in normal animals, or normal humans, and in disordered animals or 
  disordered humans.

  Essentially this is a fundamental question, because, for example, you may have 
  a drug which improves cognition in normal subjects. I'll give you an example--Ritalin. 
  Ritalin can improve many aspects of cognition in normal subjects. And, of course, 
  it's also used clinically in ADHD to improve almost the same functions.

  So that poses some real questions about how Ritalin is interactive with any 
  putative pathology in ADHD. And it seems as though it just improves performance 
  anyway. And obviously, one has to take advantage of that.

  On the other hand, you may have a drug which improves performance in the disordered 
  subject, but actually may impair performance in normal subjects. And what one's 
  actually doing is repleting some abnormality, or dampening down some pathology. 
  And the example would be, from the animal model that I showed yesterday, Sulferide. 
  Sulferide actually can improve performance in frontal-lesioned rats, while impairing 
  performance in normal rats.

  So I think these are two very different aspects of cognitive enhancement, and 
  they need to be considered quite carefully, because they interact powerfully 
  with our notion of the mechanism of disease.

  DR. GRACE: Yes, exactly.

  DR. HIGGINS: My name's Guy Higgins, and I think the idea of a database to summary 
  compounds, both--particularly non-proprietary--you can see would be a very easy 
  think to achieve from a PhaRMA incentive. I think proprietary compounds, that's 
  good in principle. In practice, you know, then we're--then the situation's a 
  lot more complicated.

  But just, in addition to having a database to summarize effects of various drugs 
  as a means of validating models, I think would be also useful is to have, on 
  the same database, a summary of people's experience with the models themselves, 
  and, you know, particularly the various non-pharmacological models which probably 
  hold the greatest interest in terms of identifying novel anti-psychotics.

  And I just--you know, over the last 10 yeas I've worked with models such as 
  isolation rearing, and ventral hippocampal lesion models. And I know it first 
  hand, the difficulty in trying to establish these models, and the endpoints--certain 
  endpoints are more robust than others.

  However, there are obviously other models I'd be interested in, such as man, 
  and so on. But, you know, before investing all that effort in terms of setting 
  it up, it would be very useful to have, you know, other people's experiences 
  with those models on such a database.

  So, you know, in addition to a compound database, a model database would be, 
  I think, a useful inclusion.

  DR. GRACE: That's a very good point. I mean, we all know that schizophrenia 
  is a very heterogeneous disease. And there are probably a number of different 
  etiological ways that you can get to a schizophrenic adult. In the same token, 
  there are probably a number of manipulations we can do in order to model different 
  aspects of schizophrenia, and some of them are going to model certain cognitive 
  aspects better than better cognitive aspects.

  Having that type of database, where we have lists of deficits, and lists of 
  drugs that have been tried in order to ameliorate these deficits could be really 
  important whenever you're trying to find drugs that are going to selectively 
  address one pathology over another.

  DR. GEERTS: Hugo Geerts.

  An exciting issue of your presentation was that you were going to pursue animal 
  models in which you have co-medication of, let's say, chronic medication of 
  an epileptic, on top of a cognitive enhancer, which is similar to the clinical 
  situation.

  The other point, of course, is: what about the dose of the agents used, so that 
  you get at least a dosage which is range of the clinical situation.

  Are there provisions in the program to really keep that in mind, as well?

  DR. GRACE: Well, one thing--that's certainly an issue that's a very complex 
  issue, given--especially looking at mice and rats versus humans, and the differences 
  in metabolism, blood-brain barrier and a number of different issues.

  But one thing that we wanted to be able to include in this is pharmacokinetic 
  data. Having a handle on the pharmacokinetics, especially in animals versus 
  humans, is going to allow us at least at first outset, to try to understand 
  what kind of differences we're going to need as far as dosing regimen, as far 
  as drug interactions, and things along those lines.

  It's something that everybody pays attention to--at least everybody should pay 
  attention to right now whenever we're looking at chronic treatment of animals 
  with different types of test drugs. And it's important to keep this in mind 
  whenever we start to do these types of cognitive enhancer, anti-psychotic drug 
  combinations. Because, certainly, overdosing with an anti-psychotic drug is 
  going to impair even the--I would think--a very strong cognitive enhancer. And 
  selecting the correct doses is going to be critical.

  DR. MARDER: I just wanted to respond to what Lisa Gold said about TURNS, and 
  sort of addressing this potential translational barrier from going from clinical 
  research that will come out from TURNS, back to animal studies, and of course 
  the other way around.

  MATRICS has addressed it by, you know, meetings like this, and having basic 
  clinical scientists involved at every--at virtually every meeting. As TURNS 
  evolves into a clinical network, an advantage to having a preclinical network 
  would be that it would be very easy to organize communication among the two 
  to sort of address this translational barrier which always has been sort of 
  a weight on the field; that sort of has prevented that kind of communication. 
  So that could actually be part of any preclinical network.

  DR. GRACE: Well, that's certainly true. I mean, schizophrenia centers, to some 
  extent, have addressed some of the basic clinical interaction, as far as looking 
  at the pathophysiology, and having some type of a multi-site mechanism in order 
  to maintain an interaction with respect to models, pharmacological agents, and 
  treating cognitive deficits it could do a lot for advancing the field.

  DR. GEYER: One of the things that we might want to discuss later, in the afternoon 
  session--in the integrative session, because this is, of course, a very major 
  integrative issue, of the translational aspects of this enterprise--is from 
  the preclinical perspective, you mentioned the possibility of a preclinical 
  version of MATRICS, and also a preclinical version of TURNS, I'm wondering what 
  the group will feel as to whether it would require a two year consensus-building 
  evaluative discussion process, such as MATRICS has been, before one could implement--as 
  was the case clinically--a preclinical TURNS, if you will, or could that be 
  more expediently accomplished by a workshop or two, rather than--do we need 
  a two-year process?

  I don't know the answer, but I thought maybe we could come back to that in the 
  afternoon discussion.

  Lisa has an opinion right now.

  DR. GOLD: I do. Well--my observation was that the clinical effort was a two-year 
  effort, and we were being given two hours. So we thought something in between 
  two hours and two years we could probably do it.

  [Laughter.]

  And I propose two days. But--I mean, it would be great to have a discussion.

  DR. GEYER: We'll get a room--John Jonides' locked room--and try to have such 
  a meeting.

  DR. GRACE: We were thinking Caribbean island, actually.

  [Laughter.]

  DR. GEYER: Not this week. The hurricane is hitting Jamaica as we're sitting 
  here.

  Time for a coffee break. Come back, on that clock--which may not match what's 
  on your wrist--at--can we make it 10:35 on that clock?

  [Off the record.]

  DR. GEYER: We're going to pick up where we left off. This will be Group Four: 
  "The Impact of Emotion and Stress on Cognition"--no small topic. And 
  the co-leaders were Deanna Barch and Bruce Cuthbert.

  Deanna?
  

The Impact of Emotion and Stress on Cognition

  DR. BARCH: Okay.

  So the goal of this group was to start to think about and integrate some components 
  that we--those of us who work with patients with schizophrenia know how important 
  to the illness, and might be important to understanding cognitive, but really 
  have not been address in the MATRICS process to this point; and actually address 
  a lot of the issues that people have started to raise in the discussion: and 
  that's the degree to which cognitive function in individuals with schizophrenia 
  is influenced by, moderated by, or actually leads to disturbances in emotional 
  regulation, motivational disturbances, and are influenced by stress reactions.

  [Slide.]

  So one very concrete suggestion that came out of our discussion group was one 
  provided by Milt Strauss, and he actually pointed out that there's at least 
  some literature on understanding emotion, stress, motivation in schizophrenia 
  that tends not to be appreciated by folks these days because of the way psych 
  lit works. And his concrete suggestion was for the NIH to purchase a copy of 
  a book called Experimental Studies in Schizophrenia, which is an abstract book 
  that apparently covers abstracts of all studies done on schizophrenia before 
  1967. I admit I was unaware of this book, but Milt thinks that this would be 
  a boon to many of us who study cognition, emotion, motivation. So a very concrete 
  suggestion there.

  I think we can probably summarize the most important point of discussion in 
  our group yesterday by this slide.

  [Slide.]

  Motivation. Motivation. Motivation.

  It was highly apparent that many people, from many perspectives, interested 
  in the interplay of motivational processes and cognitive function and functional 
  outcome in patients with schizophrenia. This came up in Cam's talk. He mentioned 
  this as a possible factor contributing to a generalized deficit, and that if 
  this was, indeed, important in understanding cognitive function in schizophrenia 
  then we needed to better understand and isolate and measure motivational influences 
  on functioning.

  I think it will be apparent that this is going to relate likely to what the 
  topic of Group Seven is, in terms of social cognition, and it's going to potentially 
  relate to some of the fundamental pathophysiological processes involved in the 
  illness.
 [Slide.]

  But I think it raises a number of questions or issues that we've tried to outline 
  on this slide.

  One important goal, we believe, for the research agenda is that we need to determine 
  the neural circuits that support different components of motivation. I don't 
  think we can say at all that that's clear currently in the literature. For example, 
  one thing that came up is that it's not yet clear how the construct or constructs 
  of motivation relate to kind of fundamental systems involved in appetitive or 
  aversive processing. We use the terms "appetitive and aversive processing" 
  as opposed to, say, "reward and punishment" to be a little bit more 
  general about the systems that may be involved there. But one important question 
  came up was to determine these sort of systems are the building blocks of various 
  motivational processes and, if so, in what way and how?

  [Slide.]

  A second question--or sort of a goal for the research agenda--was to determine 
  how variations in motivational states influence cognitive processes such as 
  working memory, episodic memory, social cognition. Everybody has sort of an 
  intuitive sense that this occurs, but we don't have a ton of empirical data 
  as to how this exactly operates, or how this may be influencing cognition in 
  schizophrenia.

  A subpoint to this was the degree to which we might need to develop what was 
  termed more "ecologically valid" test of cognition; meaning that in 
  our everyday life we use cognition in an active way to modulate, regulate, determine 
  our social lives, our emotional lives, and that the tasks we use in the laboratory 
  tend to be very gold tasks that use stimuli that have no particular personal 
  salience for the individual; and that we may potentially increase our ability 
  to predict life function by using cognitive tasks that use stimuli that do more 
  clearly tap into the kinds of salient information that people have to manipulate 
  in their everyday lives.

  [Slide.]

  Another kind of goal, or research agenda, was to determine in some ways the 
  opposite: how cognitive processing might be influencing motivation, emotion, 
  stress reaction. So there's also growing evidence that variations in ability 
  such as working memory, episodic memory and so on may actually, in turn, influence 
  motivational states, or influence one's ability to regulate one's motivational 
  state. So it's not the case that this relationship is likely to be unidirectional, 
  but more bidirectional.

  [Slide.]

  Now, in doing this, there are number of obstacles, though. One of those obstacles 
  is that it's really not clear exactly how we should define motivation--or, briefly, 
  the topic of stress came up too, and the influence of stress. So we need to 
  figure out how would we define motivation? What exactly do we mean by motivation?

  For example, how is that different from effort? Are there multiple components 
  of motivation that are relevant?

  Second: how are we going to measure this? This came up repeatedly. If we want 
  to be able to measure motivation, how are we going to do it? We need measures 
  that are well validated, easy to use, and that capture the phenomena that we 
  think are relevant to the processes of interest. These could be self-report, 
  behavioral markers, neurophysiological markers. Perhaps we need to look to the 
  existing literature, such as organizational psychology, which has paid more 
  attention to trying to measure and think about motivational processes to see 
  if we can use, adapt, modify some of those measures for the purposes that we 
  need them in this field.

  [Slide.]

  A second point that came up is that more work needs to be done on understanding 
  the basic function of the appetitive and aversive processing systems in individuals 
  with schizophrenia. There's just starting to be a more extensive literature 
  on really understanding the sort of basic emotional processing mechanisms. I 
  want to point out that I think this is going to be--and the group, I think, 
  thinks this is going to be highly relevant to what was mentioned on the first 
  day: the sort of next goal of trying to understand or tactical negative symptoms 
  in schizophrenia. Because right now, we have a way or characterizing or describing 
  negative symptoms in schizophrenia that assume we understand the emotional, 
  motivational life of individuals with schizophrenia. And I don't actually think 
  that's correct. I think we have a lot of empirical work to do to determine, 
  for example, to what degree do patients with schizophrenia have an intact experience 
  of reward or pleasure, when given appetitive kinds of stimuli? The group argued 
  that we should do this with a range of different stimuli. We need to look at 
  stimuli that have very sort of basic rewarding properties, to more abstract 
  social stimuli. We need to understand is that multiple levels, in terms of behavioral 
  levels, self-report. WE need to understand it at the level of peripheral physiology. 
  WE need to understand it at the level of brain activation.

  We need to try, as much as possible, to tie this work as closely as we can to 
  the animal models that have been developed because, again, this may give us 
  better traction at being able to develop pharmacological treatments that might 
  be able to remediate or ameliorate these kinds of disturbances if they're contributing 
  to functional outcome in patients with schizophrenia.

  Another important point that came up is that we need to have a better understanding 
  of the influence of our current treatments on motivational, appetitve reward 
  processing. For example, you know, anti-psychotics often influence the dopamine 
  system. We know the dopamine system is involved in reward processing. To what 
  extent might we actually be impairing the ability of our patients to experience 
  reward or pleasure, impairing motivational processes, which, in turn, influence 
  life function by our treatments.

  [Slide.]

  And a last point that came up was that in this domain in particular, you may 
  get very different answers, depending on what level of analysis you're operating 
  at; meaning if you're asking individuals their self-reports about how they feel, 
  if you're looking at action tendencies or behavioral indicators, if you're looking 
  at peripheral physiology or brain activation--and that we need to understand 
  better why we may see these dysjunctions. I mean, a classic issue in the domain 
  of schizophrenia is that we have a number of patients who appear not to be expressing 
  emotion. And we sometimes assume that that means they're not experiencing emotion. 
  And for the most part, that's a very incorrect assumption--at least if you ask 
  them to tell you what they're feeling. And we don't understand exactly why there 
  is this dysjunction.

  And this dysjunction, to a certain extent, occurs even in healthy individuals. 
  We know that all indicators of emotional processing don't covary across individuals, 
  across time, perfectly. We need to maybe understand better what are the mechanisms 
  or factors that are driving these kinds of dysjunctions, and how might they 
  be contributing, again, to emotional disturbances in patients with schizophrenia.

  Okay. We're going to be quick and easy. And there we are.

  DR. GEYER: Thank you, Deanna. That was great.

  Do you have some comments from the group?

  Trevor Robbins?

  DR. ROBBINS: It sounded like a really interesting group, and thanks for that 
  summary.

  One thing I think's left out, which we think about a lot in the animal context--obviously 
  it has human concomitants as well, in other domains--but the concept of control. 
  So, of crucial importance in the animal domain,--for example, is the distinction 
  between various contingencies where events happen uncontrollably--although they 
  may be predictable--versus events over which one has control--for example, via 
  instrumental responding.

  And I think this could be a really important issue in schizophrenia, because 
  it gets at the whole issue of volition, as well as control.

  DR. BARCH: And, you know, that's in part what we were trying to capture by emphasizing 
  that we need to look at both the influence of motivation, emotion and such, 
  on cognition, and vice versa. Because one critical piece of this might be that 
  because of cognitive disturbances that lead to disturbances in volition or self-generation 
  of action, or the ability to regulate or control one's emotional state, you 
  might actually see that it's not always the case; that it's, you know one direction 
  motivation-emotion influencing cognition, but maybe what we're seeing in part 
  is an inability to regulate one's emotional state.

  DR. SILVERSTEIN: Steve Silverstein.

  Deanna, first of all, that was a really good summary of the very complicated 
  discussion yesterday.

  I guess the point I wanted to make was--the point you made at the beginning 
  about motivation, emotion, and all that--I just wanted to emphasize that it's 
  not just a kind of theoretical point or a suggestion because it seems like a 
  good idea--even though it is a good idea--but it's actually based on a lot of 
  evidence we already have on how the use of incentives can really improve schizophrenic 
  patients' performance already, and how changing the affective valence or intensity 
  of stimuli--positive or negative--can improve or worsen performance. So that 
  we already know that kind of a static assessment of cognition with neutral stimuli 
  in a relatively unmotivated state is not giving us the kind of true score of 
  the patient, you know, that we really want.

  So just to reiterate something I think I said yesterday, which is that it may 
  take a while before we can come up with a consensus on how we define motivation 
  or effort or a lot of these terms, but in the meantime it's still possible to 
  explore how performance changes as a function of affective salience of incentives 
  and other things which might bring us closer to understanding what people are 
  actually capable of. And there is that distinction between kind of what the 
  neural circuit can do, and what it actually does and when it does it and when 
  it doesn't do it. And those are important.

  So it wasn't just a theoretical distinction.

  DR. JAIN: Right, and I think that one of the things that came up is that that 
  might have some really important consequences for this particulary process, 
  in the sense of what's a better predictor of functional outcome? So if it's 
  the case that in everyday life--real world--we have incentives or salient information 
  that drive cognition, it might be that cognitive tasks that include that kind 
  of information might be better predictors of how people are actually operating 
  in the real world.

  On the other hand, if for some reason they're not picking up on the salience 
  of information in the environment, or somehow in their everyday life, without 
  us providing incentives, that's not occurring, it could be that cold cognition 
  is a better predictor. But we don't know. And so that is critical, to look at 
  that in a very practical way.

  DR. GOLD: Lisa Gold.

  I was wondering, in the beginning, about the--what's unknown about the neural 
  circuits of motivation, and appetitive versus aversive reinforcement and so 
  on--pertaining--I'm imagining you're speaking of in the human situation, and 
  perhaps in particular in the schizophrenic patients. But clearly there's a very 
  exhaustive literature in animals examining neural circuits of motivation, appetitive 
  versus aversive reinforcement, and some of the world's experts are sitting in 
  this room.

  So I guess, in terms of a research agenda, perhaps you might consider somehow 
  incorporating some of that information into where the next step needs to be 
  was determining whether there's a parallelism with the humans.

  DR. BARCH: Yes, absolutely. I mean, that came up repeatedly in the session. 
  And the question--I think that relates back to how are we defining motivation 
  as it's relevant to schizophrenia. Because much of the animal literature has 
  been done with sort of basic, you know, food, water--things that the animal 
  needs to survive, and you can measure motivational or appetitive drives to engage 
  in social or affiliative kinds of behaviors.

  And often when people use the term "motivation" in humans, or in patients 
  with schizophrenia, we sometimes mean something a little bit more abstract. 
  You know, we have our, you know, undergraduate subjects come in and do tasks, 
  and they seem somehow to just want to do well; you know, they want to--for whatever 
  reason. And there's a growing literature on issues related, for example, to 
  intrinsic versus extrinsic motivational states; the role that different personality 
  factors have in driving motivation; how this may--motivational states may prioritize 
  goal representations. And that needs to be brought to bear, I think, as well 
  as the animal literature, to understanding.

  But I think that's the level at which we weren't ready to say: "We understand 
  what the neural circuits are driving motivation." And we don't know whether 
  we can equate those with appetitive and aversive processing. That's sort of 
  the open question in humans.

  DR. GEYER: John Krystal?

  DR. KRYSTAL: I think one of the nice areas that gets opened up as you start 
  thinking about emotion and reward and motivation is substance abuse, and the 
  comorbidity of substance abuse and schizophrenia. And we think of roads to poor 
  functional outcome in patients with schizophrenia, one of the common disturbances 
  are the ways that substance abuse gets in--can interfere with life.

  I think it's a very exciting and productive area, and maybe the substance abuse-related 
  issues can remain a part of it.

  UNIDENTIFIED SPEAKER: Yes, I just wanted to comment on the issue of what are 
  we going to do about studying treatment effects? Given that this is an area, 
  in particular, where there is potential cause for concern that our treatments 
  might be mucking up this process in patients, and that we're paradoxically, 
  you know, doing some harm while we're doing good.

  It seems to me that the onus is to develop animal models that people believe 
  are credible for studying these processes in humans, because I don't think, 
  as a field, we're ready to embrace the possibility of needing to study medication-free 
  patients to address this process on a wide scale. It simply can't be done in 
  this country anymore.

  DR. BARCH: Although one point came up--I'm not sure it was there. It might have 
  been at the bar afterwards [laughs]--that this is another reason why, you know--given 
  how hard it is to study medication-free patients, that it's another reason to 
  try to study some of these processes as early as we can in the illness, before 
  maybe--in at-risk individuals, or individuals who don't yet have manifest illness 
  and need treatment, to try to understand whether maybe some of the processes 
  may be going awry and actually contributing to the development of, or serving 
  as a risk factor for the development of schizophrenia.

  DR. SPAULDING: Will Spaulding.

  I'm sorry I wasn't able to come to the discussion yesterday. It was hard choices 
  to make, between the discussions.

  I was curious about what role the concept of stress turned out to have, especially 
  since it was part of the title of the discussion.

  DR. BARCH: We hardly discussed it at all. Mostly that was a time constraint, 
  I think. And that's too bad in many ways, because there is such a large literature 
  suggesting that stress may play a role. And there's--but there's much more that 
  could be done in terms of characterizing, you know, how that might be influencing 
  it. And, again, what do we mean by "stress?" You know, there's a lot 
  of evidence about how that might alter brain function in animals who've been 
  subjected to chronic stress and so on.

  So I think there's a rich field there, but we ran out of time to have a more 
  detailed discussion about it.

  DR. SPAULDING: Is there something we can do to re-integrate--stress leads to 
  a lot of neuro-endocrine--

  DR. BARCH: Right.

  DR. SPAULDING: --considerations, leading, in turn, as you say, to cognitive 
  activation factors.

  DR. BARCH: Right.

  DR. SPAULDING: How can we get this back into the dialogue?

  DR. BARCH: Well, we actually tried to include it a little bit by saying "or 
  stress," because it did come up at the very end. But I think the group 
  did agree, in general, that this would be another important factor to look at. 
  Because, again, we've got paradigms for measuring stress in many was at sort 
  of a behavioral level, and we have some paradigms for looking at neuro-endocrine 
  effects. And there's other fields that have been developing additional paradigms 
  that have yet to be brought to play in schizophrenia.

  It does raise the issue, again, of studying medicated patients, because we know 
  that some of the medications they take influence the function of this system. 
  So we would agree that it's important to look at, as well.

  DR. GEYER: Cam?

  DR. CARTER: A comment on Lisa's question, and also Jim's point.

  I often say that there are two different kinds of FMRI studies: there are studies 
  that are used to test the hypotheses about the human cognitive architecture, 
  and then there are studies that test the hypothesis that human brains are like 
  monkey brains.

  And there certainly have been FMRI studies done that have looked at appetitive 
  learning and aversive learning. These sort of classic paradigms have begun to 
  be implemented in FMRI, and the technical limitations associated with imaging 
  places like the amigdulla, and the orbital frontal cortex have largely been 
  overcome.

  And I think there are results that are either on the way to print--and there 
  are some published results--that suggest that there's a lot of conservation 
  from rats to monkeys to humans, in terms of those circuits. And so that does 
  open up the possibility that these things can now begin to be explored in patients.

  The second comment, about the negative effects of meds on these systems, and 
  the difficulty of studying unmedicated patients in the U.S.--and in the rest 
  of the world at this point--I think this highlights the importance of the first-episode 
  strategy. This is, I think, one model where it's completely ethically acceptable 
  to study people before and after they're treated with medication. And perhaps 
  a need, in terms of the kinds of research support that are available for building 
  and funding networks that recruit at this stage of the illness. It's a lot of 
  work to find these patients, and it's certainly hard to get large n's in a short 
  period of time. So the only way that it can really be done is in a multi-center 
  study. And so this assumes that we have mechanisms that would allow us to build 
  in finding what works for early schizophrenia patients, and also that we work 
  out all of these other issues regarding doing multi-center imaging studies.

  DR. GEYER: Daniel?

  DR. MATHALON: One of the things that also came up in the meeting yesterday that 
  I want to emphasize is that deficiencies in reward processing or motivational 
  deficiencies may also impact on the capacity of patients with schizophrenia 
  to learn. And one of the things I think I haven't heard brought up that I think 
  it's important to raise--both in the context of emotion-motivation, but cognition 
  more generally--is that as we identify promising agents and test them in patients, 
  that we have to think beyond just looking for changes in our cognitive matrix 
  measure, but consider the possibility that these agents may work by altering 
  experience-mediated neuro-plasticity; that is, it may create a new situation 
  in the brain for schizophrenic patients to take advantage of cognitive remediation 
  training, social skills training, and that in the absence of those, we may not 
  see much change, and we may end up throwing out some very good agents because 
  of that.

  DR. GEYER: One other point that I wondered if your group discussed: you, of 
  course, mentioned--talked a lot about the influences of motivation and abnormalities 
  in motivation on cognition, per se. Did you address the complication of abnormalities 
  in motivation on the performance of cognitive tasks?

  DR. BARCH: Umm--I think it's not clear--that's an interesting and subtle distinction 
  [laughs] and I think right now we don't have a way of distinguishing between 
  the two, because that assumes we would have some measure of how that cognitive 
  process was operating in the absence of any performance measure.

  We could maybe begin to--I mean, it raises an interesting question, and it's 
  not clear, right? Whether--I mean, I think the concern is that what we're getting, 
  then, by our performance measures, is being contaminated by motivation and is 
  not actually a true reflection of the underlying cognitive process.

  DR. GEYER: The kind of dimension that I had in mind relates to the Chapman and 
  Chapman old notion of "matched task difficulty."

  DR. BARCH: Right.

  DR. GEYER: And I admit my thinking here is rather vague. I don't--

  DR. BARCH: Well--so, in theory--right?--you can deal with this to a certain 
  extent by showing that you have a differential deficit in one measure versus 
  the other.

  Some would point to literature, though, and say it's been actually very difficult 
  to do that, suggesting that perhaps these motivational processes do play an 
  important role.

  So if it was the case that selective deficits abounded, and we had clear evidence 
  of problems in one domain and not others, one perhaps would be less worried 
  about motivation as, you know, a critical driving factor. But given that it 
  is often difficult to do that, and that we often find results that sometimes 
  appear to be as consistent with a generalized deficit as anything else, I think 
  it continues to raise this issue.

  DR. GEYER: Will Carpenter?

  DR. CARPENTER: Mark, I just--because it's come up a couple of times about off-medication 
  research--I wanted to make a quick comment on that.

  First of all, I think it's very important that the field not collapse in the 
  face of the problems and the criticisms of off-medication research. The clarification 
  for the Helsinki agreement--and the Helsinki is probably the poorest of the 
  international standards that have been set--the clarification in 2000 made it 
  clear that you have to evaluate protocols on a case-by-case basis; no longer 
  a categorical rejection of off-medication research in circumstances where there 
  is an effective medication.

  The first episode, which--I agree completely with Cam, except with one proviso, 
  that if a person has started to become psychotic, then you're going to have 
  formidable safety issues to address if your plan is to withhold medication while 
  you complete certain studies. And it may be relatively easy, if you're talking 
  about the same day. It may be a very different question if you're talking about 
  two days.

  There's enormous amount of safety data that would suggest that it's perfectly 
  plausible to do off-medication research without any long-term adverse consequence. 
  And people have to take very seriously, if the study design requires it, how 
  strong is the scientific merit, and how strong are the safety provisions. But 
  there are methods for doing it.

  It may be that we also need to learn more about certain possible new advantages. 
  I don't know the answer to this, but it may be that if you need an off-medication 
  study of a reward system, with imaging, that three days off a drug like Protilapene 
  may not be the same as three days off a drug like Haloperidol. There may be 
  much shorter time periods. It may be that partial agonists, with a brief withdrawal, 
  have different implications. I don't think we're informed on these things yet.

  But I would encourage--the same thing came up in the last meeting, at the clinical 
  session, where just getting an off-medication baseline cognitive measure, to 
  be able to tell whether the comparator drug made it worse, suggesting a difference, 
  or whether the experimental drug made it better, was kind of publicly denounced 
  on the face of it.

  And it will be a great mistake if the field collapses in the face of this criticism. 
  Which is not to deny that all of us are facing, now, enormous obstacles when 
  we try to plan medication-free research.

  DR. SEIDMAN: Larry Seidman.

  I just wanted to add one point to build on the issue of how motivation and reward 
  can be built into cognitive paradigms. Because this was under some discussion 
  at the meeting yesterday.

  And that was--I think we agreed that this is a very old problem in schizophrenia 
  research, and that sort of plagued--you know, are the results of any performance 
  measure valid, given the either psychotic interference or motivation? And there's 
  actually a long history of sporadic attempts to--but not systematic attempts--to 
  solve it. For example, when Phil Holtzman had first reported eye-tracking deficits 
  in patients with schizophrenia in the early ‘70s, this question came up. 
  And Charlie Shaggas put more meaningful stimuli on the pendulum, which was swinging 
  back and forth, which was being tracked, and it turned out that patients with 
  schizophrenia did improve--somewhat normalize their eye-tracking--but normals 
  also improved. So the differential deficit remained--the deficit remained the 
  same.

  But I think it's instructive that we don't' really know what the true--what 
  the difference is between a person's capacity at any one point in time, and 
  their performance. And one strategy that Milton Strauss had talked about was 
  that we could manipulate reward more routinely, with incentives--whether it's 
  monetary or otherwise--within any experiment; and that if we can simply build 
  that in as a paradigm more routinely, we may be able to solve this problem more 
  clearly.

  DR. GEYER: Okay. Thank you.

  We're one minute ahead, and I have been looking for a free minute to rectify 
  one error. We haven't introduced Dr. Robert Heinssen--my co-chair--who all of 
  you have seen working dutifully and hard, and he's been working dutifully and 
  hard on this program for the last several months.

  And I just wanted to publicly acknowledge him.

  Bob?

  [Applause.]

  DR. GEYER: Moving right along--let's see. We're at Number Five: "Biomarkers 
  for Clinical Trials II--Psychophysiology and Electrophysiology." Bruce 
  Cuthbert--another NIMH person who's been working diligently on this program 
  for a long time--and Judy Ford will be presenting.

  

Biomarkers for Clinical Trials II - Psychophysiology and Electrophysiology

  DR. CUTHBERT: I'm technically challenged today.

  [Laughter.]

  DR. CUTHBERT: The left button where? By the microphone? Sorry.

  [Slide.]

  It's perhaps unfortunate that our group's presentation did not happen right 
  after the Biomarkers I--the Hemodynamic Neural Imaging group--since obviously 
  these are complementary techniques, and it would have been nice to have all 
  of this in mind as we went along.

  It's mindful of a few years ago, when Steve Heinssen first came to NIMH in the 
  Event-Related Potential--or ERP--people felt as though their area were being 
  relegated to back-tier status and complained. And Steve's reply was, "What? 
  You know, that would be crazy, not to get the ‘where' guys together with 
  the ‘when' guys." So, of course, we need them both.

  Left button.

  [Discussion off mike.]

  [Slide.]

  We wanted to go over some of the advantages and disadvantages of these techniques. 
  For the most part we will be talking about ERPs, although there are also techniques 
  like pre-pulse inhibition which, of course, is a startle measure. We didn't 
  talk much in our group about the magneto-encephalogram--or MEG--but, of course, 
  this technique has many comparable advantages and disadvantages to traditional 
  ERPs.

  [Slide.]

  So, one advantage is that, obviously, as I just said, event-related potentials 
  reflect various processes that can be precisely distinguished in time, so we 
  can directly measure neural processes. Just as John Jonides said, he's out dancing 
  in the streets if he gets a 70 millisecond difference, as well as reaction times. 
  This is something, of course, for which ERPs are particularly well suited.

  Also, as Judy Ford pointed out, these are measures that we can measure processing 
  speed. And if you want to measure when a response is occurring centrally, as 
  in a decision to make a motor response, this is something that's not confounded 
  by tardive dyskinesia or other peripheral motor problems, because you're actually 
  measuring the response centrally.

  [Slide.]

 Some of the components--many of them are elicited relatively automatically, 
  such as n-100, mismatched negativity, or MMN, which is an early component that 
  is elicited if two stimuli are slightly different. There is also an error-related 
  negativity, which is the brain's flagging that you made a mistake when you are 
  responding. And, as it says up here on the PowerPoint, these may not directly 
  be probes of neural cognition in the sense that we are talking about it with 
  the test battery.

  However, these may be, in fact, probes of the fundamental building blocks upon 
  which cognition is built in some higher order sense, and so in that sense we 
  may be probing the fundamental structural integrity of the system. And this 
  is very potentially useful.

  [Slide.]

  Going on--these measures are something that can be acquired rather quickly. 
  So, in terms of an overall test battery, it takes up a relatively short amount 
  of the session. You can either use what Judy likes to refer to as the "sparse 
  EEO array" of one or two or three electrodes. On the other hand, there 
  are modern techniques--as you saw illustrated in the talk yesterday--that you 
  can put on 128 or 256 channels in five or 10 minutes, with some of the modern 
  technologies.

  Another advantage, of course, is that you have a more realistic recording environment; 
  something that more looks like a living room or an office, as opposed to the 
  rather unusual situation of the magnet; no concerns about patients, that they 
  might have metals in their head, or that you'll put metals into their head while 
  they're in the scanner. And, of course, with a lot of patients, in schizophrenia 
  and other disorders, you may have claustrophobia and other concerns. I know 
  in our lab in Florida, we actually had a whole simulated magnet, and we'd have 
  people come there for the first session, just to give them a chance to have 
  sort of a desensitization run to get used to the environment.

  [Slide.]

  For many of these components there's also much less of a concern with the practice 
  effects that have been discussed so much this morning. These are relatively 
  automatic components, and so they're juste going to be there. And you do see 
  effects over time, say, with habituation, but they are more immune, relatively 
  speaking, from the practice effects that we have to worry about. And this is 
  particularly true for some of the earlier components that I've mentioned.

  [Slide.]

  Finally, you can see we were taking notes this morning as other people were 
  talking earlier. These techniques are readily translatable in many cases--although 
  not all--to animal models. And we don't have to worry about trickery in the 
  subjects, and strategies and so forth again, because many of these responses 
  are relatively automatic.

  [Slide.]

  Also, obviously, these measures are not subject to the concerns--as in bold 
  imaging--with the vascular effects changing, and the base level of blood flow, 
  and so forth.

  [Slide.]

  Finally, these measures are relatively low cost. Even a 256-channel array is 
  only, probably, $80,000 or $100,000, and the per-session cost, after that, is 
  very low. As one of our group pointed out, in the overall cost of a clinical 
  trial, where it's so expensive to ascertain and screen patients, this may not 
  be a very large consideration, in terms of the comparable costs for an FMRI 
  session. However, in cases where you are trying to screen a large number of 
  compounds, this may be a significant factor. And, certainly, from the point 
  of view of a funding institute, we like to fund lower cost wherever we can get 
  the job done.

  [Slide.]

  Of course--this is not a perfect set of measures, either. There are disadvantages 
  here, as well. Of course the most obvious is the low spatial resolution compared 
  to FMRI.

  One strategy that our group talked about was that in some cases, of course, 
  people are using ERPs and FMRIs simultaneously, where labs have the technical 
  capability to do the very difficult job of recording event-related potentials 
  in a magnet environment.

  Another strategy we discussed is to use FMRI to localize effects, particularly 
  where they can be localized in a relatively gross sense; that is, left versus 
  right hemisphere, or parietal versus frontal effects. And then once you know 
  that's pretty much where you are, then you can use a series of ERP studies to 
  answer a lot of questions about various aspects of a task. And then when you 
  finally bump up against the place where you need more spatial information, you 
  run another MRI study, and then you go back. So you sort of iterate sequentially, 
  as a cost-efficient strategy to more along rapidly.

  [Slide.]

  Another disadvantage, obviously, is that most ERP components have been developed 
  in the course of relatively specific paradigms. And they were not necessarily 
  conceived of to measure the kinds of higher order cognitive functions that we 
  have been talking about in our test battery. So the P-300, for instance, is 
  generally considered to be a response that is associated with context updating, 
  if you will; sort of flagging the fact that something has changed in your overall 
  context, and the frequency of events. The mismatched negativity is something 
  very specific, to a difference in two stimuli, that you've had one pattern of 
  stimuli when you're really not paying attention, and then something is slightly 
  different and the brain flags that.

  So these generally are sort of, to some extent, constrained by paradigms, and 
  that's a task for future research is to find ways of sort of equilibrating these 
  kinds of tasks to the elements of the test battery.

  [Slide.]

  Finally, it's been pointed out that, unlike FMRI where you can measure over 
  a fairly sustained time period, for the most part ERPs and other peripheral 
  measures have a relatively limited ability to measure sustained processing. 
  There are cortical slow waves in which you can measure for up to six, maybe 
  eight seconds, but this is a trickier recording environment, because you have 
  to look at DC potentials and all the things that can affect these slow DC potentials, 
  like DeSkal potentials and so forth.

  So, although you can get there for the most part, this is an inferior measure 
  in terms of looking at sustained time responding.

  So, over all, I think the message there is that these techniques and FMRI and 
  other hemodynamic imaging techniques--are complementary techniques, each with 
  their own strengths and disadvantages and, to a large extent, we want to use 
  the appropriate technique in order to get at the answers to the experimental 
  questions that we have.

  [Slide.]

  Going on to the priorities for research, the first would be that our group, 
  at least, felt that there still was a lack of a lot of basic information about 
  many of these ERP components, just in terms of sort of the basic science of 
  the pharmacology of schizophrenia; what are the effects of various agonists 
  and antagonists on these measures? And the group members were not aware of a 
  really very large literature there, which obviously hurts their application 
  to looking at cognitive remediation.

  Likewise, we don't know, really, what happens with the various cognitive enhancing 
  drugs that have been tried to date. So there's just not very much data on this 
  yet. "PE" is just short for "psychophysiological" and "electrophysiological."

  [Slide.]

  Another problem, similarly, is that there is relatively little information that 
  would be relevant for an overall clinical trial or screening setting, or things 
  like effects of different populations, gender effects, age effects; effects 
  of different environments, and so forth. And, again, this is just the kind of 
  basic information that would be useful to have.

  [Slide.]

  Finally, as with the other measures that have been discussed this morning, we 
  need to look at the reliability of these measures on a test-re-test basis. And, 
  again, the data here are somewhat limited.

  One group member commented that it could be useful to look at measures taken 
  over two successive days--which I think was a strategy mentioned earlier for 
  other areas. And this has been proved useful in some contexts in getting more 
  reliable effects related to experimental measures.

  [Slide.]

  Our second priority--and these all overlap to some extent--is to actually look 
  at these measures to detect the relatively subtle changes in cognition that 
  we might affect; look at when we're trying to screen a large number of drugs.

  First of all, again, as I've mentioned, the earlier components--pre-pulse inhibition, 
  the P-50 gating that we've talked about; mismatched negativity; error-related 
  negativity--are relatively automatic. And the advantage of these, as I mentioned 
  before, is that these can be used as probes of basic neural integrity. So there's 
  a real advantage there to looking at just the fundamental aspects of the system.

  [Slide.]

  The later components are more influenced by cognitive processes; various kinds 
  of task variables--difficulty, demand; the kinds of stimulus contents and so 
  forth. Obviously, then, this is a plus, in that it's more related to the kinds 
  of cognitive batteries that we may be interested in. The minus is that it's 
  not so automatic, and is not perhaps as good a probe of just the fundamental 
  neural architecture.

  [Slide.]

  Finally, we felt that there still needs to be a lot of research to determine 
  what paradigms would be best used, and what components are best to look at this 
  basic neural circuit integrity. For instance, Judy Ford commented that in the 
  P-300, there's just a fundamental effect with aging that, on the average, we 
  lose about 1.2 milliseconds per year in P-300 latency, juste reflecting, on 
  the average, the system slowing down over time.

  Similarly, with the onset of Alzheimer's disease, you see successive reductions 
  in P-300 amplitude. So this is a case where we have some indication that this 
  can be a useful measure. But over across all measures, we still need a lot more 
  basic data of this type.

  [Slide.]

  Finally, we think that there is a good potential, but we need a lot more information 
  about the use of these measures in model systems and in screening large varieties 
  of compounds. So, in a drug screening protocol, for instance, one potential 
  use of these compounds--going along with the points that I've made earlier--is 
  to look at model system, at what the protective effects of different compounds 
  could have, using ERPs as dependent measures. For instance, it's well known 
  that ketamine reduces P-300 amplitude, and so this could be a useful model system 
  just to screen a large number of drugs to see which of them might restore, in 
  part or altogether, P-300 amplitude. So this would be a potential screening 
  technique.

  [Slide.]

  Similarly, there is a question about which of these paradigms, and which components 
  could best be used in primate models and other animal models. And unfortunately 
  Dan Javitt was not able to be with us, as originally planned, yesterday. But 
  he is here today, and we may hope to have some of his comments about this in 
  the discussion period.

  So, overall, that was where we felt the priorities are. WE think that these 
  are very useful techniques and have a lot of advantages as described, but there 
  still is a lot of fundamental work for them to be applied to this area. But 
  we think that it would be worth the effort to do so.

  [Applause.]

  DR. GEYER: Topic's open for discussion.

  Trevor Robbins?

  DR. ROBBINS: Sorry--I may have missed this, but was there any discussion on 
  the utility and potential of MEG?

  DR. CUTHBERT: [Off mike.]Pardon me?

  DR. ROBBINS: Was there any discussion on the utility and potential of MEG--magneto-encephalography?

  DR. FORD: We didn't actually discuss that in our group. We could have. We don't 
  personally have experience with it [laughs] so it seemed to be a less relevant 
  thing to discuss.

  But it certainly is something that we actually are hoping to get into--recently 
  attended a conference on it, and are making a move toward getting one, because 
  it has a lovely spatial--lovely spatial information, as well as providing the 
  temporal information that we already enjoy with the ERP.

  And we actually entered some discussions with somebody who has a monkey-MEG 
  and so, again, in our attempt to have translational studies, then we would have 
  our monkey-MEG and our human-MEG, and we would be able to do the same sorts 
  of things we're talking about with ERPs.

  So--we'd like to go that direction, but haven't managed it.

  DR. GEYER: Cam Carter?

  DR. CARTER: So, Bruce, you commented that there is a paucity of data with regard 
  to pharmacological effects on EFPs. And I think that's true.

  The paper that I showed yesterday, from Enoch Calloway, from the 1980s, actually 
  the focus of that paper--in addition to making a pitch for experimental cognitive 
  approaches to measuring cognition in psychopharmacology--was to describe a whole 
  program of research that he did for quite a long time, looking at things like 
  stimulant effects on P-300 and what have you. I think that was a big part of 
  Noch’s research career. I think he was a little bit of a--you know, I 
  don't think there was a large movement, but I think Noch was clearly a pioneer 
  in this area.

  With regard to the issue of ERP correlates of sort of higher cognitive functions, 
  you mentioned the ERN and the P-3 and the N-4. What do you think the potential 
  is for things like slow waves--slow-wave analysis--which you mentioned is technically 
  challenging? Or perhaps wavelet analyses that look at various spectral domains 
  associated with invoked--cognitive-invoked brain activity, for looking a little 
  bit more closely at what might be localizable surface electrophysiology signals 
  that could be useful targets for studying the neural correlates of higher cognitive 
  functioning?

  DR. FORD: One of our hopes, actually, is that we might be able to use something 
  like Faurier transform--perhaps after a wavelet analysis of Gamma-driving data, 
  and hope that it could, in fact, be a reflection of the integrity of a neural 
  circuit. You know, in the best of all worlds, you know, this ability of the 
  system to flicker at a 40 hertz rate may be an essential building-block to your 
  ability to do the Wisconsin card-sort task, or an interdimensional shift, or 
  something like that. You know, this is all a very much pie-in-the-sky. It would 
  be lovely if we could do that, because these measures are--although they are 
  subject to attention, you can get them without attention; you can them without 
  motivation.

  So it would be--we're going that direction. Of course the need then is to relate 
  it to the measures that we're interested in it being a reflection of, and that 
  is the power to know what your ability to do these complex cognitive tasks.

  And if it could, it would be a wonderful way of probing that system.

  DR. CUTHBERT: With respect--if I could just do the first part of that, actually, 
  with respect to slow waves, there is an older literature on the contingent negative 
  variation, which was probably the best known slow-wave, or CNB--this is a wave 
  in which, classically, is evoked in a reaction-time task, when you have the 
  "ready" signal, and them the imperative stimulus, and in the interval 
  between, which can be six to eight seconds, you see this steadily increasing 
  cortical negativity across the surface of the cortex, which is very reliable. 
  It's one of the most reliable effects, and has been shown to respond to many 
  attentional variables. And if I remember right, it's been shown to be somewhat 
  diminished in schizophrenia--probably related to problems with attention and 
  so forth.

  So this is something, I think, because it's such a reliable phenomenon would 
  be a good candidate to look at cognitive enhancing drugs.

  The other thing is that recently there's been a lot of work on slow waves--either 
  a late positive potential or even more extended slow waves, with a variety of 
  stimuli that have motivational properties. So, going back to this issue, the 
  more motivationally relevant the stimuli are, the larger the late positive potential. 
  So this is something that can be looked at.

  Also, John Cassiopo has shown that you can do this as a sort of an oddball thing. 
  You can look at how people parse stimuli by an oddball. If you show a series 
  of happy faces and an angry face, you'll see a P-300 actually, or a late--it's 
  longer than P-3, actually. It's sort of a late positive complex, related to 
  the fact that the subject categorizes that as "different" face. So 
  this is something, using a variety of motivational stimuli, or stimuli with 
  motivational properties, could be used to assess how people are changing sets 
  and that kind of thing.

  DR. FORD: And let me just add something about the slow waves, also, which I 
  think may be what you're getting at.

  One of the things about the contingent negative variation that Bruce mentioned--and 
  maybe that's where you were going--is it's a lovely way, actually, to parse 
  what is typically considered a working-memory task. You know, you can independently 
  look at the encode phase, and then you've got this long period of waiting--of 
  holding these items in memory before you act on them. And we can look at that 
  at that period.

  DR. CARTER: Yes, and I guess maybe the message for industry is that this is 
  not your father's pharmaco-EEG. I think that's probably--particularly when you 
  start talking about spectral analysis, I think it harkens back to an effort 
  that really didn't bear much fruit in the 1970s and ‘80s with regard to 
  EEG. But I think this is also--not your father's EEG, is I guess the point.

  DR. FORD: Thank you, Cam. I'm glad you made that point.

  DR. MEYER: I was going to make a similar question, because I think one of the 
  issues that you didn't mention, either as an advantage or a disadvantage, is 
  the lack of specificity, as with the P-3, which you find in schizophrenia, in 
  children at the risk of alcoholism, and a variety of other issues.

  And it raises the question--and it raised the question for me, as we think about 
  applying the more expensive technologies, of the dangers of our small-end studies. 
  And I think that's very instructive, in terms of the history of both this field 
  and, potentially, for the application of the more expensive technologies.

  And I guess the issue is the potential feasibility for now moving some of this 
  technology, which is less costly, into some of the necessary population-based 
  studies that might either give us greater specificity, or related to some diagnostic 
  issues, or greater specificity with regard to cognitive issues.

  DR. FORD: One point about specificity--if what you're talking about is diagnosis, 
  and whether you could use the P-300 to help you make a diagnosis, that's one 
  question. Okay.

  I just want to clarify one thing, because I think this is a misunderstanding 
  by some people who are slightly--who don't do this every day, and that is that 
  it's really the visual P-300 that seems to be more sensitive to whether you 
  have alcoholism in your family. And it's the auditory P-300 which seems to be 
  more sensitive to schizophrenia.

  And--Bruce, you had--

  DR. CUTHBERT: [Off mike.][INAUDIBLE].

  DR. GEYER: Dan Javitt?

  DR. JAVITT: I apologize for not being here yesterday, but it sounds like you 
  did a great job without me.

  The only comments I would make--I mean, first of all, I think all the measures 
  you have there are great. I didn't see a lot about visual processing. You mentioned 
  visual P-300, but actually, you know, as we go along in other groups, there's 
  actually a lot of deficits early on in the visual system. We're going back to 
  doing SSVP, which people haven't done for years. But if you do that you have 
  magni-cellular deficits that even seem to be driving some of the emotion-recognition 
  deficits, with faces and things like that.

  DR. FORD: Mm-hmm.

  DR. JAVITT: So I wouldn't leave the visual system out. And there's a lot to 
  be mined, and I think you can get very great differential deficits, in terms 
  of cognition. I mean, P-3 is, of course, best known, but you can take a task 
  like the AXEPT task that we stole from Cam way back when, and actually you can 
  get EP to each of the individual stimuli in it, and exploit the temporal resolution 
  that you have, and marry that to the FMRI that gives you the spatial extent. 
  And it can actually show, for example, juste as Cam would have predicted, that 
  you don't' get responses to the invalid cue the way you should.

  So there are ways of taking some of these--not all cognitive tasks lend themselves 
  to EP. But if you have a cognitive task that you think is really informative, 
  EP really can go in and try to look at the sequence of activation and see where 
  it's breaking down.

  So it's great for sensory, but I wouldn't discount it at all for cognition.

  And in terms of animal models, I think you're absolutely right, and the only 
  thing that I would add there is: a lot of the animal models are basically one 
  shot. You know, you give your psychotomimetic drug; you give them amphetamine. 
  You see what it does. They're not really developmental models. And I think a 
  lot of these things--and it may be where the strengths of ERP are in animals, 
  is that you can implant electrodes, and then have the animals live for a long 
  period of time afterwards, and you can really follow, sequentially. And now 
  that some of the knock-out mice are coming out that have genes of interest to 
  schizophrenia, or some of the perinatal or prenatal treatments are really being 
  worked out, that would be an ideal place to use ERP, to try to model the prodrome 
  and the neurodevelopmental aspects that go into development of schizophrenia, 
  rather than just focusing on established illness and what can you do to reverse 
  the established illness.

  I think a lot of what we need to be starting to do is take some of these treatments 
  that we really care about here, and move them as early as possible into the 
  course of the illness, and try to find medications that prevent the onset of 
  the illness. Because I think it's going to be very hard--much harder to reverse 
  the symptoms.

  Danny mentioned the possibility of restoring plasticity, but adults have just 
  such poor plasticity to begin with. If we can restore plasticity to the adolescents 
  we'll be much better off.

  So those would be my only additional comments.

  DR. CUTHBERT: Thank you. Those are excellent comments, and we certainly didn't 
  mean to neglect the role of cognition in ERPs, but we wanted to focus on some 
  of the more automatic components, if you will. But thank you. I would totally 
  agree.

  DR. FORD: And it's good that you highlighted the visual system. Certainly--I 
  think that the more you study schizophrenia, and the more you look at other 
  places in the brain, you're going to continue to find deficits, basically, almost 
  everywhere you look.

  I think the visual system is sort of the last bastion of what we thought was 
  going to be normal. It turns out it's not. And it may underlie a lot of the 
  more complex cognitive processing and emotional.

  DR. GEYER: Wim?

  DR. RIEDEL: Wim Riedel. I've got two additional topics. One is, very bluntly 
  put, how many electrodes? And the other is in relation to that, what about studies 
  of coherencies? And maybe also in relation to that, have you any views on how 
  they can be translated to animal models?

  I'll elaborate a bit on "how many electrodes." I think you did address 
  it in terms of mentioning the lack of regional specificity, but on the other 
  hand, in the same presentation, stressing that you can hook up 256 electrodes 
  in 10 minutes.

  So I think it's important to formulate a clear message which direction to go. 
  Because that's often very confusing in the pharmaceutical industry. Some people 
  say the more electrodes the better the method. I tend to advocate the opposite 
  standpoint.

  DR. CUTHBERT: Yes, I think that issue, or the issue of the number of electrodes--and 
  I'll let Judy speak to the coherence issue--really goes to the issue of the 
  experimental question that you have. For instance, with P-300, it's now pretty 
  much generally agreed that there are a lot of generators all over the brain, 
  and that you may--if you're just looking at a fairly straightforward P-300 effect, 
  and you want to see if, for instance, some drug has an overall effect, you may 
  only need one--you know, just something at the vertex; or three--you know, FZ, 
  CZ, PZ, and that's sufficient to answer the experimental question.

  On the other hand, there may be cases where you want better localization than 
  that. Obviously, with ERPs, you know, you're not going to, say, know that you're 
  in the amigdula versus, you know, somewhere else very close by. You know, you 
  don't have that specificity. But particularly for cortical locations, you may 
  do very well.

  As Trevor mentioned, MEG is going to be even better than that, and we simply 
  didn't have a lot of people that wanted to talk about that. But that's also 
  very relevant.

  And it should be--people should be aware that with these new large arrays, such 
  as the 256-lead systems--and the math models are getting a lot better too, to 
  work in the inverse problem. So we can get much better specificity than we used 
  to.

  So I think it's a matter of the experimental question. If you need, you know, 
  the most sensitive spatial resolution, at this point you're still looking at 
  FMRI. But if you have a question where, you know, fairly close is good enough; 
  and particularly if you could be bounded by prior work in FMRI, then a large 
  array EG may be ideal to answer your question. But I think the last point--the 
  experimental question at hand is what you need to bear in mind.

  DR. FORD: And your second question, regarding coherence--this is something I'm 
  just beginning to get involved in, so I'm probably not the best person to talk 
  about it. But, certainly, if what you're interested in is something that we've 
  been interested in, which is frontal temporal, or frontal parietal coherence, 
  you would want to put electrodes over these areas.

  So in our coherence studies, we're sure to rather densely sample these areas. 
  We may not have samples--electrodes at the other places, but in order to calculate 
  the coherence between the frontal and the temporal area, we use these electrodes.

  And as far as the effects on cognition--of cognition on coherence--I don't know. 
  We have one paper out on coherence during talking and listening in schizophrenia 
  that seems to be one of our best reflections of whether people are hallucinaters 
  or non-hallucinaters. And there was no attempt whatsoever to try to tie that 
  to any sort of cognitive profile. And we have another paper that is in revision 
  right now, also about coherence--Gamma coherence--related to the perception 
  of when you say something, whether what you hear is what you said. We've distorted 
  the speech, and we see that the coherence between frontal and temporal areas, 
  when we distort your speech as you are saying it, is disturbed, suggesting that 
  it's--this coherence is a measure of binding of experience with expectation.

  But as far as, again, relating it to cognition, we haven't attempted that.

  Maybe other people here who--

  DR. RIEDEL: The point I was about to make is why wasn't it on the priority list?

  DR. FORD: Well, it's interesting, actually. I think Cam mentioned that this 
  is not your father's EEG. And people who do ERPs sort of grew up in an ERP tradition. 
  And although we use EEG, we didn't actually analyze the EEO in its time-frequency--it's 
  frequency way, like the typical EEG people. But I think more and more of us 
  are now beginning to do some of that.

  So, basically, very few of us have had a lot of experience with it.

  Dan?

  DR. GEYER: We need to move along. We're over time.

  DR. CUTHBERT: I'd just say it should be on the list.

  DR. FORD: Yes.

  DR. GEYER: Thank you.

  The next breakout group was entitled "Social Cognition: Human and Animal 
  Paradigms." This is a field that's relatively in its infancy. It was in 
  this room, April more than a year ago, that social cognition was added by acclamation 
  from the audience as a seventh domain of interest.

  Michael Green took the charge of speaking to this, and organizing this, and 
  we've brought a good friend, Berend Olivier, from the University of Utrecht, 
  to contribute some of the animal perspective on this, and Jackie Crawley and 
  Tom Insel also contributed. So, I look forward to your presentation, Michael.

  [Discussion off mike.]

  DR. GEYER: There's another group of workers who have been contributing to this 
  effort, and will continue to contribute to this effort, as we try to bring the 
  products of the breakout groups into a publication format. And these are the 
  raporteurs, who have volunteered--some of them from disparate fields--but to 
  come in and help with note taking and recordkeeping and contributing to the 
  breakout groups.

  I just wanted to name them. They're listed in your MATRICS detailed program 
  about the breakout groups, but--Raymond Cho, Stan Floresco, Sharon Engels--who 
  I think just walked out of the room; Raymond Cho did double duty; Haim Einat 
  also did double duty, as did Stan Floresco. And then Jamie Driscoll, as well.

  So, it's like--we were recently at the Olympics, and they had 160,000 volunteers 
  that they kept thanking every chance they got. And these are also volunteers--on 
  a smaller scale. But very important, and I want to thank them.

  

Social Cognition: Human and Animal Paradigms

  DR. GREEN: Thanks very much.

  We had a rather lively discussion about social cognition in schizophrenia. And 
  what I'll do in this report from the breakout group is to first of all give 
  a brief background as to why we're even considering social cognition; secondly, 
  to mention some of the themes that came up during that session that interface 
  very directly with the themes that have been mentioned so far in the groups 
  that have presented so far; and then, third, I'll present the research agenda 
  in three specific areas.

  As Mark mentioned in his introduction, the area of social cognition was, in 
  fact, not one of the cognitive domains that was initially reviewed by the sub-committee 
  that was looking at separable factors that was chaired by Keith Nuechterlein. 
  And it was not originally proposed to be included in the MATRICS battery. It 
  was in the first meeting, held in April of ‘03 in this very room that 
  there was a rather strong sentiment to indicate that social cognition should 
  be included.

  And the arguments ranged from a couple different topics. One is that there was 
  the argument that when social cognition measures are added to batteries, they 
  are reasonably good at predicting functional outcome, and even beyond the degree 
  of prediction from non-social measures of cognition; in other words, there is 
  an added value to having measures of social cognition as part of the battery 
  in terms of predicting outcome. Not only that, there are data--largely unpublished 
  at this point--to suggest that social cognition may be a mediator that acts 
  between non-social cognition and functional outcome.

  The second line of argument came from individuals involved in cognitive neuroscience 
  that suggested that the neural substrates of social cognition were at least 
  partially different from those substrates that subserve non-social cognitive 
  processes. And based on discussions following this meeting there was a feeling 
  that we, as a committee, did not want to omit this potentially important area. 
  We didn't want the battery to miss out on what would be a potentially useful 
  area. And so it was included in sort of a future-looking effort.

  The disadvantages of including social cognition are that there was an added 
  domain to the cognitive battery that adds something like 10 to 12 minutes to 
  the length of the battery. And, of course, whether or not that's a large or 
  small amount of time depends on whether or not you're a neuropsychologist or 
  the one conducting the clinical trials. But that was seen as sort of the cost 
  of adding another domain.

  And the reason in the first place why the measures were not evaluated was the 
  feeling that, first of all, they haven't been included all that much in schizophrenia 
  research, and they haven't reached a state of maturity, compared to the neurocognitive 
  or non-social cognitive measures.

  So that is, by way of background, as to why we're even discussing social cognition 
  in the context of a research agenda.

  During our discussion yesterday there were several themes that came up that 
  overlap extremely well with the presentations you've already heard from the 
  other breakout groups. So I'll try to enumerate some of these key themes.

  First of all, one theme was the need for consensus; in other words there needs 
  to be some agreement on definitions; there needs to some gathering together 
  and some agreement--that, basically, in the absence of the consensus, in the 
  absence of this agreement, it's a non-starter. You don't know how to move forward 
  from that point of time.

  And so there was a suggestion for a mini-MATRICS process that would reach some 
  agreement, for example, on definitions of terms, where the territory begins 
  and ends in the area of social cognition. That's a point that's been mentioned 
  in reference to other groups; that some agreement on exactly where the territory 
  is.

  Secondly--and related to that--is the active involvement of other disciplines; 
  that this is not something that psychopathologists should be doing in isolation. 
  There are people who do this sort of thing for a living, and that these are 
  the people whose opinions we want to have, including experts in social psychology.

  Thirdly--and this is, I think, a recurring theme throughout this meeting--is 
  a tension, and a useful tension, between the need to have something as good 
  as we can right now, and the need to have something better later.

  The need to have something as good as we can right now is driven by practical 
  concerns that here are going to be, in the near future, clinical trials. These 
  trials need to contain some type of outcome measure, as imperfect as they might 
  be, so they need to have something as good as we can right now. I think Steve 
  Marder referred to as a snapshot; something frozen in time; something as good 
  as we can come up with right now.

  The need to have something better later has been articulated well by the cognitive 
  neuroscientists to use additional measures to sort of parse these into subcomponent 
  processes. That certainly was actively expressed in our session.

  And, lastly, the need for validation of animal models. In our discussion, much 
  of the discussion was from human constructs downward translation to animal models, 
  so there were discussions about whether or not, for example, some domains of 
  social cognition, like theory of mind, just do not translate well, whereas others, 
  such as social cue perception, might translate better.

  So, with those as an enumeration of the themes, let me give you the specific 
  research agendas.

  One way that we tried to resolve this tension between what's good enough and 
  what's better later, is to have two different human research agendas; one a 
  short-term human agenda, and one a long-term human agenda. And the third agenda 
  is an animal translation agenda.

  So the first one--let's see. I'm not getting it.

  [Discussion off mike.]

  This has been cursed by Bruce.

  Okay. Got it.

  So the first agenda is the shorter term human studies, which is agreement eon 
  definition, measures and factors of social cognition.

  And what this breaks down to, more specifically--okay, there we go--

  [Slide.]

  --is the conduct of a sort of mini-MATRICS process to agree on which areas should 
  be included. Discussions yesterday included social cue perception, affect perception--affect 
  perception often is identifying emotions and faces; theory of mind, or inferring 
  inferences from other people; attributional style--how we explain certain events. 
  But exactly what should and should not be included in social cognition in schizophrenia 
  is a matter of, you know, some debate and some honest differences of opinion, 
  and requires some gathering of individuals to agree on what it is.

  In the absence of this, it's extremely hard to move forward.

  It's extremely hard to move forward.

  [Laughter.]

  Would you do this for me?

  [Slide.]

  Great.

  The discussion of increased need for input from experimental social psychology 
  was--this need for interdisciplinary input; the idea that there are experts 
  that we can benefit from. This in many ways is a reflection of what's been going 
  on at this meeting, where we're getting input from cognitive neuroscientists.

  Could I have the next one, please?

  [Slide.]

  However--as I think Keith Nuechterlein articulated quite well, as did others--that 
  as we borrow from people who do not work with psychopathology, we have certain 
  problems with the measurement. We've talked a lot about the issues of reliability. 
  But, when it comes to something like social cognition, the issues really are 
  scaling problems. If we borrow from autism, the scales are geared towards low. 
  If we borrow from normal cognition, they're just not practical to administer 
  to our patients. So we've got a need to refine these measures, so that would 
  be part of the agenda.

  Next, please?

  [Slide.]

  There was a suggestion that it would be helpful to collect interpretable data 
  on drug effects. That would include various types of drugs, but also anti-psychotic 
  medications. This is informative from a number of points of view. One of the 
  points of view of interest would be to help with the validation of animal models, 
  if we knew exactly which medications and which drugs affected social cognition.

  And could I have the next one, please?

  [Slide.]

  And, there's a need for data collection. It's surprising that we don't have 
  answers to these questions. What about data collection that involves rather 
  comprehensive neuro-cognitive batteries, as well as social cognition so that 
  we can examine structure within and between.

  Social cognition is in a sort of unique area, in reference to the other cognitive 
  areas; in other words, we don't know how well social cognition separates out 
  from the other areas. We've already made the argument there's some added value, 
  at least when it comes to predicting outcome. But we don't know if it factors 
  separately, because it was not part of the factor analyses that were reviewed. 
  And so you can look at the difference between social cognition and other areas, 
  and then, of course, you can look within social cognition. How broadly you'd 
  want to define it will almost certainly yield some factor structure within social 
  cognition.

  This is really unexplored territory, at least in psychopathology, we don't know 
  the structure well enough.

  Next, please.

  [Slide.]

  That was research agenda number one.

  Research agenda number two is the longer-term human research agenda, which would 
  look for an increased influence of cognitive neuroscience and social cognition, 
  or affective neuroscience.

  Next.

  [Slide.]

  And these would be: encouragement to support studies of neuroimaging; electrophys; 
  human brain damage; pharmacology to identify the neural substrates of social 
  cognition.

  Next.

  [Slide.]

  And that this would be used to inform the development of more refined methods 
  of social cognition that adequately parse it into subcomponent processes. So 
  we were very much thinking about the examples that have come--in yesterday's 
  presentation, by John Jonides and Cam Carter and Trevor--as to how those can 
  be parsed into subcomponents.

  And--next.

  [Slide.]

  And that there would be a need, then, to assess the similarities and differences 
  between identified neural substrates to basic non-social cognition.

  This is another way of addressing the overlap question; in other words, are 
  the neural substrates that subserve the non-social cognition entirely overlapping? 
  Partially overlapping? Independent from those that subserve social cognition? 
  These would be basic questions of definition and overlap, but at the neural 
  substrate level.

  So, so far, what you see is our attempt to resolve this tension, is to have 
  a shorter term, more product-oriented human agenda, and a longer term process 
  that would inform us more about the underlying subcomponents.

  And the last research agenda, on the next slide--

  [Slide.]

  --is the translation of social cognition into animal models.

  In many ways this was the most vexing of the topis that we approached. I should 
  mention that my co-chair, Berend Olivier, is here, and is an expert in this 
  area. And so questions can be directed to him about this.

  We struggled quite a bit with how one validates models.

  Next slide, please.

  [Slide.]

  And so there was an overall, rather general goal, to develop new tasks for rodents 
  with conceptual links to social cognition. And, as I mentioned earlier, there 
  are some aspects of social cognition--like theory of mind--that just don't seem 
  like they're going to translate terribly easily to rodents; maybe to non-human 
  primates. And there's other areas--and Jackie Crawley gave a number of paradigms 
  that sounded to have some pretty direct similarities between things we would 
  call social cue perception in humans.

  The tasks should be able to measure both social cognition and social motivation. 
  This is an interesting topic, interfaces nicely with Deanna's group, in that 
  there was initially an opinion that, well, what we're interested in is not social 
  motivation because, after all, if we define things too broadly they become hard 
  to measure and they become hard to validate. But at the end of the session there 
  was a feeling that we shouldn't be hasty about this, and we really don't know 
  if social motivation should be included, excluded; that that should be a topic 
  for discussion.

  And so we need measures to sort of consider these things separately.

  Next.

  [Slide.]

  To explore with lesion, pharmacology; genetic approaches--we heard some very 
  elegant examples of genetic approaches; the neural substrates of social learning. 
  We're using "social learning" and "social cognition" in 
  this context, because some of the examples come from social learning in the 
  animal models that Jackie Crawley uses.

  And, lastly--and this applies equally to humans and animals--to explore the 
  overlap between social cognition and negative symptoms in animals and in humans. 
  You can see how we've been led into this, sort of inevitably, because not only 
  do we start talking about social cognition, we start wondering about social 
  motivation. And before you know it, we're sort of right in the sort of interface 
  area of negative symptoms.

  And, for the most part, the group, I think, felt comfortable with this. We didn't 
  feel terribly defensive about the fact that social cognition might be, in fact, 
  a reasonably good overlap area between cognition and negative symptoms. And 
  that's something that we don't need to worry about. It just simply might be 
  the way that we're being led, and that might be the sort of boundary we're getting 
  between cognition and negative symptoms.

  So that's, then, first of all, a reason for why we're looking at social cognition 
  at all in these breakout groups; secondly, some of the themes that our group 
  touched on that relate rather well to the other groups; and then a specific--not 
  so specific, but moderately specific research agendas for short-term human, 
  long-term human, and then animal models.

  And I want to thank, certainly, the members of the breakout group for the discussion. 
  And I'll stop here and take any questions.

  Thank you.

  [Applause.]

  DR. GEYER: Thank you, Michael.

  We knew, going in, that that was going to be a daunting task you undertook. 
  I think you did a good job.

  Deanna?

  DR. BARCH: Michael, great summary.

  You spent a lot of time talking about distinguishing the boundaries of social 
  cognition and cognition. And I just wanted to suggest that it may be also important 
  for us to understand the relationships--the influence of deficits in sort of 
  an emotional experience in relationship to social cognition in patients with 
  schizophrenia.

  DR. GREEN: I should have said that, Deanna.

  The definition of social cognition--I sort of listed the ones that have most 
  commonly been looked at in schizophrenia. But clearly, if we were to have a 
  category of emotion processing, it very well might fall in social cognition. 
  We've been leaning in that direction anyway.

  So that's a fair point, and that was just an omission on my part.

  UNIDENTIFIED SPEAKER: I think one of the ideas that came out in our meeting, 
  that I often think about in relation to my patients is the obstacles that they 
  face to social affiliation, which is often that the beliefs that they articulate, 
  and the behaviors that they engage in often lead to their marginalization, rather 
  than their affiliation with groups.

  And I struggle with what the connection is between basic social cognition and 
  these more complex social behaviors that seem actually quite profoundly impaired 
  in schizophrenics. And I wonder if the discussion in your group went in that 
  direction at all.

  DR. GREEN: I'm not clear on the question. Is it the relationship--there was 
  a fair amount of discussion about social cognition and functional outcome, generally 
  defined. But you have a more specific--

  UNIDENTIFIED SPEAKER 1: Well, I think functional outcomes--we often mean occupational 
  outcomes. And I guess what I'm getting at more is, I mean, I think one of the 
  reasons there's a struggle is I don't think that we have a good outcome measure 
  that reflects this aspect of schizophrenic behavior deficits.

  DR. GREEN: I think I understand your question. All right.

  Many of us--and David Penn or Steve Silverstein might want to chime in--many 
  of us do assess things like richness of social networks as an outcome, and that 
  becomes a separate social outcome. I guess, in this case, one wants to be cautious 
  about face validity, because sometimes these measures predict vocational outcome 
  at least as well as what you would think they would predict, in terms of social 
  networks.

  DR. GEYER: Other comments? Yes?

  DR. PENN: David Penn. I just want to break this mike.

  DR. GEYER: We missed your name.

  DR. PENN: All right--David Penn.

  I just wanted to follow up--can you hear me?

  DR. GEYER: Yep.

  DR. PENN: I just wanted to follow up to your point. That was really good.

  DR. GEYER: No, not now.

  DR. PENN: I just wanted to follow up--

  [Laughter.]

  I just wanted to follow up to your point--I mean, I think you speak to a couple 
  issues. One is, I don't know if we have really good measures of social competence 
  and social functioning. I don't know if Alan is here. He could talk about that 
  more, because he's developed some of the best.

  I think also, here, you're talking about complex social behaviors--and I don't 
  think there's going to be a one-on-one correspondence between social cognition 
  and a complex behavior like work functioning, just like you wouldn't expect 
  the same sort of correspondence between performance on a card-sorting test and 
  work behavior.

  So I think the way we think about it is in terms of its adding another building-block 
  to the model.

  And you talked about sort of what people's experiences are. You know, I read 
  a therapy transcript recently of a client who became psychotic while at work. 
  And a lot of the things she talked about, I mean, were sort of theory-of-mind 
  attributional problems. She talked about people having meetings, and her wondering 
  were they meeting without her and talking about her. She literally said, "I 
  do not understand the intentions of my co-workers"--which many of us can 
  probably relate to.

  [Laughter.]

  But I think--you know, we wouldn't want to--I don't want this to be a Pollyanna-type 
  thing, like social cognition's going to fill an important gap and answer all 
  the questions, but I do see it as an important building-block. And I think that's 
  what got people like me and Michael and Steve and Will Spaulding and others 
  involved in this field.

  And, by the way, you know, many of probably know this--I know the social cognition 
  people know this--I mean, there is a very long history of social cognition research 
  in schizophrenia. If you go back--longer than 20 years, 25 years--if you go 
  to the work of Richard Bentol and colleagues in the United Kingdom, and they've 
  developed a pretty elegant model, I think, of understanding how social cognition 
  relates to persecutorial delusions. And that may be one reason why we've had 
  trouble grasping this in this country because I think social cognition is more 
  of a psychological model, based more on specific symptoms, perhaps, than neurocognitive 
  models, and it doesn't lend itself--I think at least at face value--to pharmacological 
  interventions.

  But, you know, if we had this meeting in the U.K.--I know there's some people 
  from the U.K. here--and we had Richard Bentol and Max Birchwood and Peter Kinderman 
  here, I think they would have a different read on sort of how young research 
  in this area is.

  UNIDENTIFIED SPEAKER: I think one assumption that's often made in social cognition 
  research is that social cognition is somehow halfway between cognition and behavior, 
  and it somehow causes the behavior. And I understand the previous point: it's 
  that patients often behave in bizarre ways, which then cause people to react 
  to them, and cause some of their social impairment and the ostracism and stigmatization 
  and all that.

  So I think it is important, in just broader modeling of the role of social cognition, 
  to not only understand how social cognition affects behavior but how, you know, 
  behavior and--how it works the other way around, too.

  And so--which may be important for understanding performance on certain kinds 
  of tests a function of learning history of the person, or even what their current 
  life is. And it might be important to try to develop tasks--or new tasks--that 
  can try to tease that apart from performance that may be linked to learning 
  history versus that that isn't.

  DR. ROBBINS: Trevor Robbins.

  I think it's interesting to take some lessons from the area of autism research, 
  particularly in the theory of mind domain.

  We have had a workshop recently in the U.K.--I suspect you have in America, 
  too--kind of MATRICS-like in the sense of trying to get consensus on batteries 
  for epidemiological purposes over the lifespan; measuring the development of 
  theory of mind, for example. I had to chair this thing and I was astonished, 
  actually, given the large number of experts in this area in the U.K.--Richard 
  Freethal, Leslie Hustaby, and Song Barakan and so forth--on the lack of consensus 
  about what would be good measures for theory of mind that could be used, you 
  know, quickly in the type of situation that we're interested in for clinical 
  trials in this project.

  So, I do think a lot of work has to be done in that area, but you might want 
  to, in this case, bootstrap on--it's certainly been approached elsewhere.

  DR. GREEN: It's a good point. If our breakout group had a mantra, it would be 
  something like, "Agreement, refinement and data collection"--where 
  "agreement" is the sort of, you know, what constructs and domains 
  we're interested in. And this is the sort of mini-MATRICS discussion came out 
  of ours. But it sounds like that's already ongoing in the U.K.

  "Refinement"--there the problems are, you know, the sort of adapting 
  the measures for the right level, because a number of us had the experience 
  of trying to borrow from the autism literature, and just not having a lot of 
  success in doing so. And so then that would be the--you know, once you have 
  the constructs, then you could make the adjustments. And then, of course, the 
  data collection.

  But your comments are certainly encouraging, in that the efforts to reach the 
  agreement are already underway. Once those agreements are reached, those of 
  us in schizophrenia would still have to make considerable adjustments.

  DR. GEYER: Thank you, Michael.

  Lest anybody doubt that we were serious about calling this "new approaches" 
  committee and conferences, we had a breakout group that is really cutting-edge 
  and forward looking--although, certainly, biomarkers for clinical trials involving 
  genetics and genomics is not an entirely new areas, its specific application 
  to cognitive deficits in schizophrenia is certainly a forward-looking adventure 
  that Danny Weinberger and Linda Brady chaired.

  And, Linda, maybe you can present the summary for us. Thank you.

Biomarkers for Clinical Trials III - Genetics and Genomics

  DR. BRADY: Right. So we had an interesting discussion in our group. And anyone 
  that's there that would like to participate in the presentation, go right ahead.

  But we started out by talking what was the state of genetic biomarkers for schizophrenia 
  and, you know, what can they do for us. And, you know, that led to some interesting 
  discussions. And there's a lot needs to be done, in terms of validating genetic 
  markers to be useful in clinical trials, and also as a tool to stratify patient 
  populations.

  But we sort of moved away from that discussion and tried to focus on what are 
  the candidate genetic markers that we know now, and which of these markers relate 
  to cognition, and how can we go about doing that?

  So the research areas that we cam up with try to reflect the discussions that 
  we had.

  [Discussion off mike.]

  It doesn't want to work.

  [Pause.]

  All rightee.

  [Slide.]

  So, one of the research priorities that the group came up with: to collect the 
  genotypes of schizophrenic subjects in studies that would be ongoing in the 
  MATRICS cognitive battery studies, and within those studies, to select a panel 
  of candidate genes of interest to be surveyed to see what sort of correlations 
  and data came up from those. And one of the things that this could do for us 
  was to look at how the variance in the battery of the cognitive measures related 
  to the genes, or whether these genes were predictive for any individual elements 
  of the battery.

  And we talked about a number of candidate genes that might be ready to look 
  at in these types of studies.

  The next one.

  [Slide.]

  So one of them was really to try again to use the candidate genes to parse out 
  the individual variations in the cognitive domains.

  The next one.

  [Slide.]

  Another sort of variation on this was to really look at unaffected family members 
  of schizophrenics also in these studies to look at the MATRICS battery. And 
  the thought of this was that the genotyping could be used, then, as looking 
  as biomarkers in the trials, and that these particular studies would then be 
  complicated by, you know, symptom domains and anti-psychotic drug treatments. 
  And we actually have another, you know, further elaboration of that for the 
  run.

  But we thought that the genotypes could also be used to serve as markers to 
  look at specific pharmacological treatments, and the marker on which they would 
  be presumed to act. So, examples of sorts of trials that would target these 
  gene functions would be looking at D-serine or glycine; looking at the G72 marker. 
  We could be looking at alpha-7 as a marker, with alpha-7 targeted; nicotinic 
  cholinergic compounds.

  Some other targets that were felt to be ready are the COMT marker, with COMT 
  inhibitors; some of the metabotropic glutamate receptor compounds, looking at 
  GRM-3.

  And then there's a number of genetic markers that are associating in signaling 
  pathways that are connected to some of the main neural transmitter systems involved 
  in schizophrenia, and also in the systems that mediate plasticity in a number 
  of different mood disorders, and these include the growth factors such as brain-drug 
  inotropic factor; could be even the serotonin transporter and other markers 
  that come to mind.

  So that there needed to be some systematic way in which to examine the pharmacology 
  and its relation to the genotype, along with cognition.

  [Slide.]

  The second priority was also to do more genome-wide association studies for 
  genes that might be related to cognitive deficits, using phenotypes that were 
  based on the cognitive domains that were talked about during this meeting. And, 
  again, the idea was to conduct studies in unaffected siblings which were felt 
  to have a concentration of risk alleles that would help you associate genetics 
  with the cognitive domains and the deficits that were specifically thought to 
  run in schizophrenia and their relatives.

  Some of the questions that came up in sort of interpreting and thinking about 
  these data were: are the cognitive deficits that are seen in schizophrenia something 
  that is unique or different from deficits that you see in other disorders? And 
  some studies that will be ongoing, supported by other NIH institutes, could 
  be really informative in trying to figure this out.

  And another question came up was whether the cognitive domains themselves are 
  heritable. And this was a good approach to apply. We wanted to know if the genetics 
  involved were similar in schizophrenia to other disorders, and whether the gene 
  sets that you would find that lend to risk or susceptibility for schizophrenia 
  would overlap at all with the cognitive domains.

  And there was some discussion regarding that; that the cognitive markers themselves 
  could be good biomarkers for looking at schizophrenia.

  [Slide.]

  So this was really an elaboration of the first point that we talked about, but 
  again emphasizing the treatment of unaffected family members as a strategy for 
  conducting proof-of-concept or efficacy studies for the cognitive enhancing 
  drugs; even, you know, things that are being considered in the TURNS network. 
  But this would allow you to look at relatives of schizophrenics that have specific 
  cognitive dysfunction, and also have a heavy loading for risk alleles, and compare 
  the drug effects on these individuals, where you didn't have the confounds associated 
  with anti-psychotic drug treatment affecting some of the cognitive domains you're 
  looking out. You didn't have to try to dissect out the drug effects on improvements 
  in symptom domains that might overlap or interfere with dissecting the pure 
  improvement in cognitive domains, and that this might be a nice way to sort 
  of give you an idea of is this a reasonable drug candidate to move into studies 
  with schizophrenic subjects.

  [Slide.]

  This could also--so another point associated with this is that this could be 
  another way to use cognitive markers and genetic risk factors for even identifying 
  people that have--prodromal individuals that might have the potential risk of 
  developing schizophrenia, and perhaps consider intervening with therapeutic 
  intervention earlier on in the process.

  [Slide.]

  The last--there was a lot of discussion about gene expression profiling in post-mortem 
  brain and other tissues. And one of the points that was made is that there's 
  really a need to take the micro-array studies that have been done in post-mortem 
  brain into living tissues, and how is the best way to do that.

  And there was quite a bit of discussion of looking at circulating cells as potentially 
  being a way to relate micro-array profiles of gene expression changes with what's 
  going on in the brain--something that would be more readily amenable to introduction 
  into a clinical trial setting.

  And the use of these expression profiles could also be used for high throughput 
  screening. So if you were able to get a profile that indicated a cognitive enhancing 
  effect, or if you were able to get a profile that indicated risk to schizophrenia, 
  that you might be able to use that profile itself as a high throughput screening 
  approach to try to identify drugs that could counteract that profile.

  [Slide.]

  So these were sort of--I wasn't quite done--

  [Slide.]

  --these were sort of ideas that are obviously in the future, but things that 
  people are thinking about, and approaches that are being undertaken.

  And then it was thought that if you could find a treatment that did alter the 
  expression pattern of the particular gene array, that you could move into animal 
  models to try to test it; or, alternatively, you could, again, take a potential 
  candidate--pharmacologic treatment--into the clinic and see if this really is 
  a good way of predicting cognitive enhancing effects.

  And again, there was talk about trying to monitor treatment effects by looking 
  at peripheral cells such as lymphocytes.

  [Slide.]

  So, in terms of methods that were needed, there were three areas that were identified. 
  One was referred to as "biobank" which, again, is really looking to 
  accumulating a data bank, or a repository of relevant tissues from subjects 
  with schizophrenia that have gone through the MATRICS studies that could be 
  used for further studies--gene array studies, association studies; that we needed 
  DNA for genotyping subjects that has been in the clinical trials--again, looking 
  at, or having data available from lymphocytes for doing expression profiling 
  studies where you could, again, compare those effects with what's going on in 
  the brain.

  And then, again, having tissue and brain--peripheral tissues, as well, from 
  patients that have very well characterized cognitive deficits, so you could 
  try to understand what the overlap is between genetic factors there and the 
  disease itself.

  [Slide.]

  It was thought there needed to be a core sequencing facility, and that this 
  would really be a major advance to bring the field forward, because there's 
  a very big need for sequencing of the genes, as well as identification of snips 
  that are associated with schizophrenia. And there just isn't enough resources 
  out there, or enough centralized facilities to do this.

  It also would be valuable to have a standard set of DNA from control subjects 
  that could be used as a comparison for smaller studies that are done by various 
  labs.

  [Slide.]

  the last method that really is needed has to do with the area of bio-informatics 
  and statistical analysis or algorithms. And it was felt that there's really 
  a lack of tools that are readily available to begin to really look at or explore 
  the very complex gene-gene interactions that underlie the pathophysiology of 
  schizophrenia, and in even the gene-environment interactions that underlie schizophrenia, 
  and also the cognitive deficits as well.

  And so that if there were tools available to do this, then there would be able 
  to be making more associations for how does the disorder develop, and how can 
  you identify targets for treatment and cognitive enhancement.

  So that was a summary of the discussion that our group had.

  [Applause.]

  DR. GEYER: Thank you, Linda.

  Ming Tsuang?

  DR. TSUANG: Linda, I think it was a very good summary. I was there from the 
  beginning, and it looked like a rather enthusiastic discussion, not focused. 
  But now listening to your summary, it's so cohesive.

  [Laughter.]

  It's amazing. NIMH train you well.

  [Laughter.]

  In the council I listen to all these people presenting, and it's such a short 
  period of time. And you guys are very well trained. I was very impressed.

  Now--I'd like to highlight, or underline what you have said.

  In this meeting, with regard to measures of cognitive function, the heritability 
  of the measures are not stressed. Even in the MATRICS, yesterday I learned that. 
  Each measure seems to be not taken--seems to not--if I'm wrong--heritability 
  is not a major criteria--one of the major criteria to select the measures. So 
  this area needs to be really strengthened.

  However, to get the heritability, you need a twin sample. And it is rather difficult--or 
  from population genetics point of view, this may be private company cannot spend 
  so much money in this type of basic work. And this day at the NIMH--unless it 
  has immediate impact on the treatment of the patients, the funding is not very, 
  very encouraging.

  So this is the more academic issue.

  Now, get into real issues of gene expression. The gene expression, too, is very, 
  very important because gene expression is actually influenced very much by environment.--with 
  gene and external. And one of them, the drugs modify the gene expressions. And 
  I'd like to caution that gene expression seems to be a fashion, but that without 
  the good bioinformatics technique, so many gene express, it is difficult to 
  pin down this gene-gene interaction on genes. So this is actually I like to 
  caution.

  But saying that, I think this is a very important tool because using the medication 
  actually affecting the gene expression, and knowing which gene is upregulated, 
  which gene is the downregulated, where one would be able to test the cognitive 
  function and the responsible genes.

  Talking about the peripheral or circulating cells, that is the different from 
  post-mortem in its--or it's the live tissues. However, to what extent this reflects 
  the CNS needs to be tested. So, using post-mortem brain or gene expression, 
  or from there to cultured lymphocytes, to validate the peripheral blood or gene 
  expression would be very important for the future study of cognitive function 
  and gene expression. And that is the very, very important, because it's so much 
  influenced by drugs, and also the timing. Gene expression is really influenced 
  not just by one gene expression, it's two now. So these areas methodologically 
  need to be really tackled before we can really say something.

  But just like to highlight what we talked about. And this is the--you already 
  did it. I just--as addendum to this.

  And, finally, I'd like to say is this: so far we are talking about, from epidemiological 
  point of view, if I wear my epidemiological hat, seems to not meet the epidemiological 
  criteria. WE don't talk about the descriptive statistics of age, sex, social 
  and the racial/ethnic differences. Those are very important to measure the cognitive 
  functions. Of course these need a lot of money to really pursue those basic. 
  And n so far is relatively small to really come up with a significance level.

  So this is for the future, we need to worry about.

  Thank you for giving this time for me to give a relatively long speech.

  [Laughter.]

  Thank you.

  DR. GEYER: Somehow I knew it was coming, Ming. Thank you.

  Yes?

  DR. GEERTS: I'm Bill Geerts.

  Interesting conversation, but I think something which we didn't touch upon in 
  this meeting is the field of proteomics, which is an emerging field. And, again, 
  there's a lot of activity in the area of schizophrenia. And I think some of 
  the proposals which have set out there can also be translated into the field 
  of proteomics, which is the kind of side comment.

  DR. GEYER: Keith Nuechterlein

  DR. NUECHTERLEIN: Keith Nuechterlein--yes.

  Just to follow up on one of the points Ming was making, that you were making, 
  and sort of make one distinction--first of all, Ming's absolutely right: heritability 
  was not really taken into account in selecting measures or, for that matter, 
  domains, for the MATRICS first generation cognitive battery, shall we say.

  What is known about that--to the extent we do know, but we need a lot more data--is 
  probably almost every one of those seven domains. Maybe we don't know so much 
  about social cognition--is partially heritable, at least in the general population. 
  Sometimes we know that in schizophrenia families, sometimes we don't. We need 
  more either twin or family studies to address that.

  I think, though, that even more than that awe need to know the genetic covariation 
  amongst cognitive traits. WE need to know how many of the domains that are apparently 
  somewhat related in individuals, that relationship is because of common genes' 
  driving those things versus how much of the covariation is due to other within-individual 
  sources that aren't genetic, because that will give us a real leg-up on to what 
  extent we can hit several of these areas, driven by kind of a common genetic 
  model.

  In some cases we have--you know, like COMT or something--reasonable genes to 
  look at. In other cases we don't really yet, but it would help us to know just 
  within families--and especially within schizophrenic families--which of these 
  cognitive variables travel together across individuals. It's sort of the cross-trait, 
  cross-relative question.

  DR. BRADY: Yes, that's a very good point.

  UNIDENTIFIED SPEAKER: [Off mike.] William Hadrays [phonetic]--to follow up on 
  that point about genes for cognition. Keith mentioned COMT. But I think one 
  of the more striking things that came out in the session was the fact that many 
  of the schizophrenia susceptibility genes that have been detected so far, and 
  seem to be replicating in a number of studies, almost invariably seem to have 
  a cognitive effect, when we look at these genes either in healthy controls or 
  in schizophrenia patients, for the most part G72, GRM3, BDNF, dispindine, we're 
  all seeing, I think, different particular tests, perhaps, but mostly very similar 
  kinds of results that actually are having cognitive effects.

  So I think there's a reasonably robust literature now to suggest there are genes 
  to go after vis-a-vis these cognitive studies.

  I think the problem with the MATRICS using these kinds of things in turn, of 
  course, is sample size. None of these genes or alleles are that common, and 
  thus we'd have to think about strategies--what we stratify--perhaps by genotype. 
  And that's, of course, a relatively difficult thing to do in the context of 
  a standard clinical trial. And for any sort of power calculation, the numbers 
  usually are not really consistent with standard clinical trial power calculations. 
  And so I think it's a problem we will have to deal with.

  DR. GEYER: I know one person who's been involved in the MATRICS all along, who 
  would have liked to have been here, is Ed Scolnick, who couldn't join us. But 
  I think it would be his position that we should at least collect DNA on all 
  of these trials that are using the MATRICS battery, in the hopes that gradually 
  we will accrue a sufficient sample size to begin to address questions that we 
  can't even predict right now.

  With that--oh--if we can keep it brief, which is not characteristic.

  [Laughter.]

  DR. TSUANG: This time is not a speech.

  I think we talk about the potential of candidate genes approach for the future 
  drug development is the genes controlling the enzyme for glucose metabolism. 
  And, as you know, in schizophrenia, actually, the cognition seems to be improved 
  by glucose administration. And we are testing on the relatives. And therefore, 
  we use the NIMH Genetic Initiatives data, to look into the linkage of any genes 
  controlling the enzymes for the glucose metabolism.

  And we found that the chromosome 1 and chromosome 5 seems to have the linkage--those 
  enzymes seems to be are related. So, for the future drug development for the 
  cognition, I think that zeroing in on this potention--glucose metabolism enzyme, 
  and which is related to obesity, it relates to lipid metabolism and so one--will 
  be very important. That is one area which I forgot to highlight.

  Thank you.

  DR. GEYER: Thank you. That segues nicely into lunch.

  [Laughter.]

  Some of us would love to metabolize some glucose about now.

  So we'll see you back here at 1:35 on that clock. I don't know how that relates 
  to the clock in the cafeteria.

  [Off the record.]

  DR. GEYER: So, if breakout group co-leaders and speakers from yesterday morning 
  can join in the chairs up in the front, as a panel--

  [Pause.]

  Okay, one last request for the breakout group leaders and the speakers from 
  yesterday morning to join us at the front. Linda, I think that includes you. 
  [Laughs.]

Panel Discussion

  DR. GEYER: We had some behind-the-scenes discussion over lunch, indicating that 
  our sense from the morning session is that there is considerable consensus built 
  already--summed up in "We need more research." "We need more 
  research of a preclinical nature; we need more research of a clinical nature. 
  And we need more research to examine each of the domains, their separability 
  and their interactions."

  Given the sense that we hope the rest of you share that there is some--if not 
  a consensus already, we had taken the executive action of shortening this afternoon 
  session.

  The plan I think we can still continue with, which is to invite any of the--kind 
  of the consumer representatives, the speakers from the first morning--and we'll 
  need some floating hand-held microphones, if we can get them--to share their 
  opinions about the consensus research agenda that they may or may not have heard, 
  especially if there are notable gaps that they feel we still need to identify 
  and work toward filling.

  So--and then we're basically going to open it up to comments from the breakout 
  group leaders about interactions--further interactions that haven't been already 
  identified, or comments from the floor as to gaps in the research agenda that 
  have been described and that you can identify.

  So, Wayne, do you want to lead off, as some initial impressions from--or do 
  you want to--we don't have a rigid agenda here.

  DR. FENTON: [Off mike.] [INAUDIBLE] I guess it goes without saying that it has 
  been a very rich discussion over the last day-and-a-half. And I think, you know, 
  from the perspective of NIMH, a very generative one, in terms of ideas for initiatives, 
  and how we might play a role going forward.

  For me, I guess there was some doubt going into this whether the net effect 
  of our having created, along with the UCLA team, a MATRICS battery; will it 
  be positive or negative on the field? And I think one of the tensions coming 
  into this meeting was the idea of whether or not we're freezing progress in 
  time in creating this battery; whether that would have a stifling effect on 
  innovation, and therefore be net negative.

  And I think by the end of the day-and-a-half, my own impression is that, in 
  fact, that is not the case; that the seven domains that had been identified 
  through the MATRICS process--they may be imprecise--it's become clear that they 
  can be decomposed into more finer-grained cognitive dimensions that probably 
  map on neuroimaging, and map on certain physiological processes with greater 
  specificity, but at the same time, it seems to me, at least, that they represent 
  anchors that should be useful going forward in designing research, and in designing 
  that does, working in a basic direction, try to decompose, and try to understand 
  the underlying cognitive processes with greater specificity.

  I think, for me, Dr. Hagan's slide mapping the domains onto a putative animal 
  model was an integrative concept that's important coming out of this meeting, 
  and I think will be a potentially promising area.

  I guess, as well, Alan Breier's comments, in looking at the drug discovery process 
  frames, for me, what the ultimate goal of this research is--which, of course, 
  is really to ameliorate the cognitive deficits that plague patients and that 
  interfere with their functioning. And, in that, I think that our aim going forward 
  ought to be--as Alan showed in his slides--both in terms of development of animal 
  models and human assays, is to increase the probability of success in developing 
  new compounds or, alternatively, determine more quickly whether something's 
  going to end up being a failure.

  I had asked, actually, Steve Zalcman and Bruce Cuthbert and Bob Heinssen, who 
  are program staff branch chiefs would be responsible, from the NIMH side, for 
  implementing some of these recommendations, to come up and if they could--

  DR. ZALCMAN: Sure. Well, I want to echo what Wayne said, generally, about the 
  past day-and-a-half being extremely productive and informative. And as probably 
  one of the NIMH people who was, you know, least optimistic, shall we say, about 
  this undertaking, it seems to me that based on both the presentations yesterday 
  morning, and more on the breaks--still on the breakout groups yesterday afternoon, 
  that a picture has now emerged in my mind, at least, of how the TURNS--no matter--whether 
  or not any compound that's a hit is identified. And, obviously, the odds of 
  that happening are not very favorable. But even if that doesn't happen, having 
  gotten this far, and viewing the TURNS, perhaps, as an opportunity to segue 
  into what may get called "MATRICS II," and thinking about how to put 
  into effect the research agendas that were so nicely outlined this morning is 
  going to be a very positive outcome for the field, and for pushing the science 
  forward.

  DR. CUTHBERT: I would also agree with Wayne's comments that this has been a 
  very productive set of discussions.

  From our point of view at NIMH, what we always hope to do with meetings of this 
  type is bring together people who might not have otherwise sat in the same room, 
  and hear lots of different perspectives, and hope that we get some new synthesis 
  out of that.

  Speaking for myself, I found that very useful over the last couple days. I think 
  that started for me with--right off the bat--with Cam Carter's talk yesterday, 
  talking about all the different domains of thinking about cognition, just in 
  a behavioral sense, and what the fundamental elements of cognition are; how 
  we can measure that through neuroimaging, and the limitations--but also the 
  psychometric challenges: that we need to pay attention to floor effects, ceiling 
  effects, item-response theory, getting the intervals right and so forth. And 
  we have these constructs about cognition, but we also have to think about how 
  to measure them properly, then.

  This is actually something that Bob Heinssen and I, and others in our branch 
  have been worrying about across all areas with mental disorders, is just trying 
  to figure out better measurement of the constructs that we have about these 
  various dimensions and elements of disorders.

  So I think that this conference has been very useful for us in thinking about 
  the kinds of groups that we will need to form, and whatever initiatives may 
  come out of this--however formal or informal--I'm sure that they will specify 
  the need to have this kin of very multi-disciplinary approach. And this will, 
  I think, help us in specifying what those might look like.

  As all of you know, we are currently in the process of reorganizing, and the 
  major goal of our reorganization and our program staff is to foster translational 
  and interdisciplinary research. So we will look forward to applying our new 
  organization to this task as we go ahead.

  DR. HEINSSEN: Well, it goes without saying that the scientific quality was just 
  extraordinary in this meeting, and I really appreciated being able to listen 
  and learn from all of the speakers, and benefit from the discussions.

  From my perspective as a program officer, you know, I've got a lot to think 
  about, about how we--you know, choices that we can make that might help this 
  initiative continue to move forward.

  I heard a number of very interesting ideas. We're going to have to mull these 
  over internally to make some choices about strategizing about what we should 
  do first. But one thing that became very clear--to echo what Bruce was saying--when 
  you bring together very bright people from closely allied, but not necessarily 
  oversecting areas, you very often discover areas where there would be profitable 
  collaborations. And this idea of bridging disciplines came up in a number of 
  ways in the different meetings.

  It seems like everybody wants a good psychometrician. That was something that 
  came up very frequently. And, you know, I think that that is certainly something 
  that we can help foster--the kind of interaction that would bring those skills 
  to bear to some of the demands, now, for the cognitive tests, both for the beta 
  battery and for emerging cognitive science tests, in a way that both areas benefit.

  There were other bridges that were suggested. I certainly was very, very impressed 
  with the kind of collaborative spirit between the people from academia and industry, 
  and really would like to see what we can do to further that momentum. Some areas 
  that were mentioned that are exciting are the ideas of these databases that 
  would provide conceptual tools that might further the work for people developing 
  models, and people who have compounds that they would like to see tested within 
  these models is something that's very intriguing, and is the kind of thing, 
  again, that I think NIMH--we can do things to facilitate that kind of product, 
  and produce a tool that will be useful to the broad research community.

  So, for my part, I'm planning on leaving this meeting with a lot of ideas to 
  mull over and digest, and looking forward, in the succeeding weeks, to be prioritizing 
  these things with my colleagues, and continuing this dialogue with the people 
  who participated today, to make some choices about, you know, how we move forward.

  DR. GEYER: One bit of housekeeping I neglected to mention--Tonya is coming in 
  the room. She can confirm or deny this--Tonya? Hi.

  I gather that some of the 3:30 taxi appointments have been moved earlier in 
  time. Is that correct?

  VOICE: [Off mike.]

  DR. GEYER: Okay. So they can wait a little bit, if need be. But, at any rate--with 
  that bit of housekeeping out of the way, I'd like to turn to our two industry 
  representatives from yesterday morning. Either of you--I guess my curiosity 
  is whether you feel you've heard, as a research agenda, either what you wanted 
  to hear, or you didn't want to hear, or whether you can identify still missing 
  gaps in what we've identified as the key areas to prioritize?

  DR. HAGAN: [Off mike.] Okay. Well, I've been very heartened by the degree of 
  openness to the ideas that have been discussed. And I think some of the themes 
  that have been picked up, both preclinically and clinically, are recurring themes 
  [INAUDIBLE]--and wanting to get our techniques frozen in time, locked, too early.

  Okay. Clearly, there's a need to consider that, and to do some further work 
  on validating the key techniques.

  I don't want the meeting to degenerate into violent agreement and harmony, and 
  so I'd pick out a couple of areas. I think one thing which came up in the discussions 
  within Group Two, which was chaired by Keith and Trevor, was, I think, as the 
  discussion evolved it became clear that the nature of the process that had been 
  used to define the seven domains identified some anomalies. And some of the 
  anomalies were, for example, the Wisconsin and the PPI. And so if you move--if 
  you now move into the animal literature, what you find, if we don't' address 
  that discrepancy, or those issues, then we'll be in a position where actually 
  some of the best validated preclinical models don't line up particularly well 
  with the tests being used for the seven domains.

  So I think that's an area where we need to pay some attention.

  I'm very taken with the idea of preclinical MATRICS. I think that would be ideal. 
  I'm a little bit nervous that the straw man slide that appeared in my pack, 
  which was put together fairly hurriedly on an airplane, should be taken as gospel. 
  I think the intention was that that's a starting point. It was really to trigger 
  people thinking about what we need to do here. And I think the proposal of taking 
  that up in a preclinical MATRICS workshop of some sort to hammer those details 
  out a little bit further, I think we should do that.

  The idea of a data base for exchange of information, I think that's also an 
  excellent idea. I think my view on this is that much of this methodological 
  development is what you could call "pre-competitive." And so sharing 
  information on procedures, sharing tips, and sharing data on standard compounds, 
  I think that's probably not going to be an issue, and I think it's achievable, 
  and could accelerate development in the field quite substantially.

  I think if you start, then, to think about transfer of compounds--proprietary 
  compounds--into that database, even tool compounds--I think with tool compounds--by 
  which I mean compounds which are no longer in development--I think there are 
  still some IP issues to work through. I don't know how we solve that, but clearly 
  within the mechanism behind the TURNS initiative, that issue has been addressed. 
  And so we should be able to learn what we can achieve, preclinically, from the 
  TURNS initiative.

  So I look forward to that workshop--that preclinical workshop--and trying to 
  get these tests married up, and start to generate more comprehensive and more 
  consistent databases so that we can really understand, you know, which models 
  are going to be useful for us, and where the issues lie.

  DR. GEYER: Okay.

  Alan Breier.

  DR. BREIER: A couple of comments. I think one on sort of process.

  I think, again, to reiterate, that our progress in development new treatments 
  for schizophrenia has not been good. And we spend an awful lot of time at it. 
  And I think taking a Manhattan Project mindset approach to that problem of developing 
  new drugs is important. And when you do that, I think you then get a sense of 
  urgency, and I sense of having to move forward in a different way.

  And, to me, MATRICS has been sort of that Manhattan Project, of saying "Let's 
  operate in a new way, with a different set of rules, and move with a different 
  sense of urgency." And I think that's what we need.

  When what that was was bringing diverse groups together that have common interests, 
  really demanding consensus. And I think for many people in science that's very 
  difficult; to compromise on things that they feel strongly about for some greater 
  good. And you need that greater good to rally around the cause of saying, "Yes. 
  Okay. I'm willing to do that." So it's that demanding of consensus, clear 
  timelines, clear deliverables in moving it forward which has been so beautiful 
  about MATRICS.

  I'd also note and say--and this was just reflections from the last day-and-a-half--I 
  don't think every problem should be MATRICS-ized. I don't think every problem 
  should be one where we seek consensus on. I don't think science works best, 
  necessarily, for certain problems, in that way; that raw independence, and violent 
  disagreement, and those kinds of things can be very healthy. And I think that's 
  particularly healthy when you don't have that sense of time urgency, and where 
  the depth of knowledge is not that extensive. Because consensus will close off 
  new ideas and new thought. And if it's a very nascent, immature area, I think 
  the last thing you want to do is drive to consensus.

  So I think separating the two different bodies of work, and applying the right 
  process and methodology is important. What I would say is that it's important, 
  then, to get clear in one's mind: when are we in a Manhattan-Project-MATRICS 
  frame of mind, and then work by those rules, and when are we not? And when we 
  are, what I would ask is that when you're doing your work, you're always linking 
  it back to "how is this going to help discover a drug?" And if that 
  link isn't there, then you're in the other conversation--a very important conversation, 
  but it's a different conversation.

  And to maybe get some discipline into separating those two conversations--not 
  that they're not somehow going to be mutually complementary, but they are very 
  different.

  And so what you need, I think, in discovering new drugs is a drug-hunting culture; 
  a drug-developing culture that has this different sort of sense of "We'll 
  take 80 percent right. We maybe won't get 100 percent, but we'll take 80 percent." 
  You must compromise. We cannot go forward if people don't come together and 
  work in that way. So I think it's a different kind of an approach.

  I want to state a real strong sense of enthusiasm for TURNS. And in my presentation, 
  I hope I hope I wasn't too pessimistic about the likelihood of success. I just 
  want to reinforce, I think, the importance of TURNS. But at the same token, 
  TURNS right now is an experiment waiting to happen. If the experiment is successful, 
  there's no doubt in my mind that TURNS could be self-funding; something that 
  is so valuable that dug companies are going to be lining up and pleading with 
  you to take their best molecule. But that means delivering, hitting your milestones, 
  high-quality data, doing it efficiently, having more the drug-hunting culture 
  in TURNS, and kind of moving that sense of urgency that "we've got to--patients 
  are waiting, we've got to get these drugs through--"--that kind of mindset.

  But I don't think that's there right now. So I don't know that all the absolute 
  best molecules, you know, have come forward. But I do think that will come, 
  as TURNS proves itself to be an effective network.

  So that's what I would put forward as sort of the goal of TURNS. And I'm very 
  confident that that could be achieved.

  You know, there were just so many interesting ideas that came up over the last 
  day-and-a-half. And I could spend an awful lot of time talking about the one-off's. 
  And I'll just point to a couple.

  I thought Anil Malhotra's idea of doing some mono-therapy, cognitive testing 
  of cognitive-enhancing agents--potentially cognitive-enhancing agents--in unaffected 
  relatives with cognitive deficits was really a neat idea; that by having a certain 
  criterion of cognitive deficits, you can take some of these agents--which have 
  fairly favorable side-effect profiles. You don't have the confounds of an add-on 
  anti-psychotic drug. You've solved pseudo-specificity--in my book--because you 
  don't have the psychosis or other potential confounds.

  Now, to me, that's sort of a surrogate study--we talked about surrogate endpoints. 
  Would that impress the FDA as part of a registration package? In my view, the 
  answer is absolutely yes. Would be a stand-alone? No. It would be a surrogate. 
  It would be a piece of the puzzle, but when the questions about pseudo-specificity 
  came up, if there was one study in that package, along with a compelling phase 
  three package, that would be pretty impressive.

  And the very first cognitive enhancer for schizophrenia will go to an advisory 
  committee. And people like you will be looking at that data. And I think, if 
  you think about it from that perspective, you're going to have these questions, 
  and you're going to look for convergence of evidence to prove it. So I thought 
  that was kind of an interesting idea.

  I do want to come back to the importance of identifying targets. The best is 
  going to be PET tracers for your targets. You can get those, and get there, 
  so you know the drug's getting in the brain; it's going to its target. And then, 
  through saturation curves, you can come very close to telling what the dose 
  should be. That is extraordinarily valuable information.

  You can't get there with FMRI. You can get part of the way there, but you'll 
  still do experiments where you don't have a clue what the dose is. And when 
  you fail, you don't want to be back scratching your head and wondering, "Did 
  we just test the hypothesis? Very valuable. Let's move on." Or, "Did 
  we have the dose right?" And start over.

  So it's things like that that I think will make it or break it, and become very 
  important.

  I'll stop there. I could keep going. But I won't.

  DR. GEYER: Thank you, Alan and Jim. Both were very helpful comments.

  I'd like to turn briefly to our representatives of academia. You don't have 
  to elaborate much, if you don't wish to. But--Cam Carter?

  DR. CARTER: This has been a very exciting couple of days, and I've been impressed, 
  too, with how much consensus there is--even though that might be a bad thing 
  among scientists.

  I also think it was very interesting to get a sense of how, within this Manhattan 
  Project-like mentality, there seemed to be some fairly obvious sort of timeframes 
  for things that on our to-do list. And it's a very interesting process to be 
  part, not just of something that's happening now, but something that can be 
  happening in a few years.

  DR. GEYER: Trevor? Trevor Robbins?

  DR. ROBBINS: Well, I have very little to add. I was obviously excited to come 
  to a conference of what I regard to be the most challenging issue in the whole 
  of neuroscience--actually. This--solving schizophrenia really is the big one. 
  And it is the most difficult. And I think all of what we've heard over the last 
  couple of days is consistent with that view.

  My own perspective on MATRICS, I have to say--coming as an outsider, as it were, 
  from U.K. and from another domain--has been slightly skeptical, in the sense 
  of we're quite suspicious of consensus in the U.K.

  [Laughter.]

  Because of our funding system we've been trained to be very competitive, as 
  it were--critical and skeptical of ideas. And I've been reassured, I think, 
  a bit, in the last couple of days, but the trend to inject some innovation into 
  what could otherwise be a very conservative approach, I think, to assessing 
  drugs. And I think--I'm not only thinking of DSM-III and IV, was rather conservative 
  approaches, but also the Hamilton, ADIS, CALC. I feel as though for much of 
  my career I've struggled against these dinosaurs, as instruments which have 
  been, in clinical terms, very successful, but with the disadvantage of not really 
  being at the cutting edge of new discovery.

  But my fears on this have been somewhat allayed, I think, in the last couple 
  of days.

  Together with the intellectual euphoria of coming together with such a distinguished 
  group, and hearing about very exciting ideas in different domains, I have to 
  say I still have a slight sense of trepidation about the enormity of the problem 
  that still faces us. It's useful identifying all those problems. Now we've got 
  to solve them. And we have to find the resources to do that.

  DR. GEYER: I'd like to turn next to Steve Marder and Michael Green, if he has 
  anything to say, as representatives of TURNS. Jeff Lieberman is at the back 
  of the room, as well.

  Have you learned things from this that will help you in the conduct of that?

  DR. MARDER: Yes, well a couple of things.

  I just want to say one thing regarding MATRICS, since the curtain is gradually 
  falling on MATRICS, and people are talking about other MATRICS activities. Just 
  two very brief points.

  One--the process of reaching consensus moved the field forward, I believe, as 
  much as the product of the consensus.

  And, secondly, if other people are looking for using methods of sort of reaching 
  consensus, one thing that we did that I think was helpful is we set out with 
  the idea that we couldn't fail to reach consensus, so we looked at the path 
  before we actually got into the content, so that it was sure to happen. And 
  I think people found that reassuring.

  Regarding TURNS--I came here with the sort of naive hope that I would hear people 
  in the sessions--for example, on biomarkers, for imaging, and for psychophysiology--sort 
  of arguing about why their measure would be the best to use as a biomarker. 
  And what I heard was exactly the opposite; that the people who were studying 
  each method seemed to be overly--not overly--seemed to be talking mostly about 
  the limitations of their methods as a biomarker. And I think that that was a 
  clash of two different approaches; one of the very careful scientist who moves 
  along very, very gradually, based on accumulating knowledge, and interventionists, 
  like myself, who says, you know, "Here's a drug. Let's give it a whirl. 
  We're not sure what it does, but it seems to be reasonably safe."

  And--now the problem is that we would really like those biomarkers, and we may 
  have to settle for things that really are not as good, as well developed, as 
  they could be, but the best around at this current time. And I also realize 
  that we have to make maximal use of the TURNS as a laboratory for improving--for 
  a number of the activities here, like refining what the domains of cognition 
  are, and the usefulness of some of these biomarkers.

  DR. GEYER: I would just interject one thought from--my personal thought is--and 
  Martin Paulus actually helped crystalize it for me at lunchtime--and that is 
  this whole enterprise has engendered some real dialogue between industry and 
  academia and NIH. And MATRICS officially concludes its contractual existence, 
  I think, at the end of this month? Or something like that?

  DR. MARDER: [Off mike.] Not really. We have a little bit more time because of 
  the PASS study.

  DR. GEYER: Yes, PASS is extending its lifetime a bit. But it is a finite enterprise, 
  and it's clear, from my experience, that this is a fruitful process of engaging 
  industry, academia and NIH--NIMH, in particular--in an ongoing dialogue.

  And so I would hope that, if nothing else comes out of it vis-a-vis cognition 
  in schizophrenia, or depression or other issues of clinical importance, that 
  this process and this activity of engendering serious dialogue between these 
  communities that have different needs and different goals, to some degree, but 
  very overlapping interests, would continue.

  I'd like to open it up, in case there are any suggestions from either the panelists 
  or anybody in the audience that has anything useful to add--or not useful.

  UNIDENTIFIED SPEAKER: One thing that I think came out from all of the discussions, 
  and everybody saying, is that there is more need--it was said again and again, 
  and I just want to emphasize again--there is a large need for interdisciplinary 
  research, and there are very few groups that can do it alone. So it requires 
  a lot of collaborations between laboratories, wherever they're found, in academia 
  or in industry. But something like that also came up in terms of the conversations 
  that we had, in terms of NIMH basic science priority. People were bringing up 
  again and again interdisciplinary work.

  DR. GEYER: If there's nothing--Jim? Jim Gold?

  DR. GOLD: First, I'd sort of just like to congratulate Wayne on this effort. 
  Because when this first came out, there were probably few people more skeptical 
  than me that it could be done without a great deal of rancor and pie-throwing 
  and name-calling. And there's been both a civility to this whole thing--which 
  is remarkable, given the caste of characters involved--

  [Laughter.]

  --and other than the people at U.C.L.A.--which is very laid back. I mean, it's 
  a real thing. Steve has a very light touch.

  But I think this has been remarkably successful. And I think there is--of the 
  people who, in the neurocognition community, it could be that our end product--no 
  one agrees with it, but there's a consensus that it was the right thing to do 
  now. Which I think is--and I think people are all fairly placid about that. 
  And even when we get called names.

  But I think you got a lot of bang for your buck out of this. I really do--in 
  terms of moving the field forward. And I think there are other opportunities 
  that are very similar to this--or issues that we've discussed. And I think they're 
  worthy of sort of pursing.

  The issue of biomarkers seems like--you know, is totally ripe. The issue of 
  how to move cognitive neuroscience out of the lab and into the clinic successfully, 
  rather than just being sort of an export business--I think that there's room 
  to try and, you know, spend money to get to a product. That would move the field 
  forward.

  I think that's true in social cognition, too, where--even if that's a field 
  that isn't ripe for plucking, it's clear that there are fruit on the tree, and 
  we can't quite agree whether they're--you know, is it an orange tree, is it 
  an apple tree--and that there are probably ways that NIMH could, with relatively 
  little money, move the field far, fast, and then let the competition begin.

  DR. GEYER: Yes, I'd like to thank also, not only Wayne, but Ellen Stover and, 
  of course, Michael and Steve Marder at U.C.L.A. Great job.

  I think we're done.

  [Applause.]

  DR. MARDER: Let me, if I may, just grab the very last word here to say that 
  Mark Geyer has done an incredible amount of work for this conference, even joining 
  us on conference calls from Athens during the Olympics, at who knows what local 
  time.

  So, Mark, thank you very much for making this such a success.

  [Applause.]

  [Whereupon, at 2:26 p.m. the conference adjourned.]