Email Icon Facebook icon Twitter Icon GooglePlus Icon Contact

User Top Menu

SRF Interviews David Lewis

Posted on 6 Apr 2006

Across neuroscience, there remains a wide gap between the systems-level study of neuronal function and the molecular genetic neuroscience that is now possible in rodent models. David Lewis has made his mark in part by drawing on a wide variety of approaches in an effort to begin to build an arch to connect these different levels of analysis. As a result, his work has given rise to a hypothesis that connects disturbances at the molecular level of neurotrophin signal transduction through findings at the cellular level of certain types of inhibitory prefrontal neurons, to a disturbance of network properties necessary for working memory, one of the cognitive domains affected in people with schizophrenia. While there is no single straight line, let alone proof, that this science truly explains a part of schizophrenia, Lewis's research has gained currency as an attractive attempt to study schizophrenia with a combination of modern and classic tools.

"I was asked to mention a little about my background and how I became involved in research. As the youngest member of a large extended family, I was always the kid left out at family gatherings, unable to keep up with the older kids. But I had an aunt, who was unmarried and lived with my grandmother, who always spent time with me. She played the ukulele for me and made me think I was the best poker player in the world because I always took home more pennies than I brought when I played with her. But, as a very young child, I realized there were times when Aunt Jessie disappeared. Weeks or months would pass, and then she'd come back, clearly not the same. I didn't understand it, but gradually she'd get better, we'd spend time together, and then the same thing would happen again." [Excerpt from speech by David Lewis while accepting 2005 Lieber Prize for Schizophrenia Research, from NARSAD Research Newsletter, vol.17, iss 3, Fall 2005]

SRF: What is the hypothesis driving research in your laboratory? What, at least at this point in time, do you think are the steps leading to schizophrenia?

DL: One perspective on schizophrenia is that it is a disease of the brain. I'll explain this view in terms of one model, which is illustrated in the first figure in a recent review (Lewis et al., 2005). First, some overview and definition of terms. A standard, well-exploited model of the disease process can serve as a strategy to study schizophrenia. Conceptually, this model assumes that a set of causes, that is, a polygenetic inheritance in the context of some adverse environmental events, unleash pathogenetic mechanisms that then produce a pathological entity. That would be a set of disturbances in the brain that so alter its function that the resulting pathophysiology gives rise to the properties that we recognize as the clinical syndrome. Understanding this process is critical from a clinical perspective because the key to the development of rational treatment interventions is knowledge of pathophysiology. With treatment, we are seeking to reverse the functional disturbances in the brain that give rise to the clinical syndrome. Likewise, understanding pathogenesis is the key to secondary forms of prevention because what we seek to do is to interrupt the connection between the underlying causes and the appearance of a brain abnormality. It's unlikely that we will accomplish primary prevention, that is, removing the causal factors, given how complex the genetics are.

The traditional starting point in the study of brain disorders has been a clinical-pathological correlation. For example, Alzheimer disease had that at its inception. Alois Alzheimer cared for this woman with a progressive dementia, Auguste D., and when she died he looked at her brain, found plaques and tangles, and associated the clinical syndrome of dementia with this pathological entity. Things sat rather quietly for the next 70 or so years, but when George Glenner in the early 1980s identified a 14-amino-acid peptide in amyloid, that set off what has become a very exciting two decades of research into this disease.

Historically, schizophrenia has an interesting relationship to Alzheimer disease, because Alzheimer was working with Emil Kraepelin. Shortly before Alzheimer made his discovery, Kraepelin had formulated his view of dementia praecox as a distinctive disorder, but he did so purely on the basis of the clinical phenomena. He didn't have a clinical-pathological correlation, and we still don't have a robust one today.
So our strategy has been to ask: Can we extract from the multifaceted clinical syndrome of schizophrenia an element that seems to be a core feature and look for a potential pathological entity? (I emphasize two words: "Core," as in of major importance, and "a" meaning it is not the only aspect of the illness. We have to focus our studies on specific aspects of the clinical phenotype, yet avoid being overly reductionistic.) With that clinical-pathological correlation, you can then begin to dissect other aspects of the disease process in order to inform novel strategies for treatment intervention.

That's the overview. Twenty years ago, I became interested in the observations that executive processes, which include working memory, are characteristically disturbed in schizophrenia, and that these disturbances are associated with dysfunction of the dorsolateral prefrontal cortex, raising the question of what is wrong in this brain area that could underlie those impairments. So a straightforward hypothesis would be: There is a discernable disturbance in the neural circuitry, or the functional architecture, of the prefrontal cortex that is responsible for the disturbances in executive processing and working memory.

SRF: Put simply, scientists break the clinical presentation into different aspects and then pursue those toward establishing a pathophysiology and pathogenesis—basically breaking up a big problem into a set of smaller ones. Given that, will there ever be one or a couple of key pathways in schizophrenia that will explain the disease and make good targets, or will there always be a very complex picture with many, many different aspects but none dominant?

DL: Both etiologically and clinically, schizophrenia clearly is heterogeneous. Different people get the illness for different combinations of reasons, and those who meet diagnostic criteria for it have different clinical pictures. But there are also elements that are relatively conserved despite people's clinical heterogeneity, and which may represent a common downstream abnormality. One such conserved clinical feature is the disturbance in the cognitive processes that depend upon the prefrontal cortex. That's not to say that everybody who meets diagnostic criteria has that, but it is a common enough feature that it's worth looking for a conserved underlying biological abnormality. A hypothesis we are pursuing is that there may be "hub" of molecules or circuits that are sufficiently downstream from the etiological processes and proximal to the clinical syndrome that they're found in a large percentage of individuals. That is what we've been looking for. If, in fact, they exist, then it would suggest that there may be a set of targets that, provided they are druggable, could be helpful. By that I mean neither a cure nor improving all aspects of the disorder, but helping a substantial percentage of people.

SRF: …with their working memory, presumably?

DL: Right.

SRF: Let's talk about the dorsolateral prefrontal cortex. Your Web site outlines a translational research program that goes from functional anatomy all the way through to future treatments. Let's lay out your approach by using it as a guide and summarizing in a nutshell the principal findings to date for each step. That will provide a foundation and offer more detail for the reader on your respective Web pages at the same time. Start by summarizing the relevant functional architecture of the neocortex. What are the players?

DL: Whatever I tell you, I'll offend people, because I've left out their favorite player.

SRF: This hour is all about your favorites. We can't cover the whole field today.

DL: Okay. You can look at cortical circuitry as having three types of components. Firstly, the excitatory pyramidal neurons are the principal conveyers of information, both within the prefrontal cortex and from the prefrontal cortex to other brain areas. Secondly, the different classes of GABA-containing neurons provide not only inhibitory control of these excitatory neurons, but also play a very fundamental role in the timing of neuronal activity within the prefrontal cortex. Thirdly, various inputs come in from external regions. They fall into two categories: There are excitatory inputs from other cortical regions and from the thalamus that convey information to the prefrontal cortex, and there are modulatory inputs, which tend to shape the activity within the prefrontal cortex. These include inputs such as dopamine, norepinephrine, serotonin, and acetylcholine.

Each of these three categories is absolutely critical for the normal function of the prefrontal cortex; the question really isn't which is the most important.

For each of these, there is some evidence of disturbance in schizophrenia. For us, the critical question has been: Is there a component of the circuitry that is robustly and consistently shown to be disturbed in schizophrenia, and the dysfunction of which would plausibly be associated with the nature of the prefrontal-mediated cognitive abnormalities observed in the illness?

SRF: What led to this approach?

DL: I became convinced of this way to study in the mid-1980s. I had just finished my clinical training and wanted to move from the clinical research I'd been doing to a more neurobiologically-based research of schizophrenia, but was unsure for a while how to approach it. I heard Danny Weinberger talk at a meeting of the American College of Neuropsychopharmacology about evidence for cognitive task-evoked dysfunction of the prefrontal cortex in schizophrenia. In another session, the late Patricia Goldman-Rakic talked about the critical role of dopamine in prefrontal function. The synthesis in my mind of those two presentations suggested that understanding just how the primate prefrontal cortex was built to mediate those processes was going to be essential in figuring out how those processes were disturbed in schizophrenia. With the advice and guidance of Floyd Bloom, John Morrison, and Steve Foote, and later Jenny Lund, I focused the rest of my training and then the first years of my faculty appointment just on studies in the nonhuman primate prefrontal cortex, looking at who the different players were and how they were connected to each other. Many other investigators were doing this also. I was merely contributing to building an initial research base so that later, once we had the capability of looking at postmortem human brain, we would understand who the players were and how we could interpret disturbances in them in schizophrenia.

SRF: The second page on your Web site concerns cortical development. There is a fascinating story developing of how various players that we haven't introduced yet—the chandelier neurons, their axon cartridges, parvalbumin varicosities, etc.—how they change during childhood and adolescence. Can you summarize that for us?

DL: A theme developing in the schizophrenia literature was that there must be critical processes during adolescence that allow for the clinical manifestations of the illness to appear. Scientists like Robin Murray and Weinberger were saying that one of the things we must understand about schizophrenia is why it appears typically during that time frame, and Irwin Feinberg had written a paper (Feinberg, 1982) arguing that the changes that occurred in the brain during adolescence could be critical for the appearance of schizophrenia. There were two views. One held that adolescent-related refinements in brain circuitry went awry and themselves gave rise to schizophrenia; the other view was that those developmental processes unmasked another, earlier developmental disturbance that happened in utero or perinatally, or that was genetically endowed. Different as they were, both views converged upon the idea that understanding which neural circuitry refinements actually occurred during adolescence would provide insight into the elements of neural circuitry that were disturbed in the illness. And so, in parallel to our mapping of the circuitry of the adult primate human cortex, we started a series of studies looking at how that circuitry changed during adolescence. We found a number of interesting changes in the dopamine innervation, and in the pruning of excitatory inputs to pyramidal cells. What really struck us were the marked changes that occurred in markers of chandelier cells, especially the magnitude of the changes during adolescence, and the coordinated refinements in different markers of chandelier neuron synapses with pyramidal cells.

SRF: The chandelier neurons are special. Before we talk about the changes, briefly describe them for us.

DL: Chandelier neurons, so named because of the distinctive morphology of their axon arbor, are a subset of cortical GABA neurons that provide synaptic inputs exclusively to the axon initial segment of pyramidal cells, near the site where the action potential is generated. Chandelier cells are the only source of synaptic inputs to this location, and as a consequence they exert powerful control over the output of pyramidal neurons. In vivo data indicate that chandelier neurons serve as a type of veto cell, preventing pyramidal cells from firing (Klausberger et al., 2003), although a recent in vitro study suggests that GABA released from chandelier cells could actually be excitatory under certain conditions (Szabadics et al., 2006; see SRF related news story and comment on Szabadics et al., 2006).

SRF: What were the changes you observed?

DL: On the presynaptic side, we observed reductions in two proteins, that is, parvalbumin and the GABA membrane transporter (GAT1), both of which contribute in different ways to regulating GABA neurotransmission. On the postsynaptic side, we noticed a change in a subunit of the GABA(A) receptor that is selectively localized to the axon initial segments of pyramidal cells where the chandelier neurons have their input.

SRF: Levels of these two presynaptic markers, and also of the receptor, drop off precipitously in the period just before what in monkeys corresponds to adolescence. How could that be physiological? On the face of it, wouldn't it mean that it is normal for inhibitory GABA transmission onto those pyramidal neurons to decline as teenagers hit adolescence? How does that relate to schizophrenia?

DL: We are still working to figure out exactly what these changes mean functionally. The findings clearly suggest that the nature of inhibitory control at the axon initial segment is different pre- and post-adolescence. We can draw inferences from what we know about the function of these different molecules.

For example, parvalbumin serves as a sink for intraterminal calcium; it slowly binds calcium that is in the axon terminal. Typically between electrical signaling down the axon, some residual calcium is in the terminal, and if a series of action potentials arrive in close proximity, these so-called calcium transients can sum up. Its buildup will augment neurotransmitter release, since release depends upon the amount of calcium in axon terminals. So when you reduce parvalbumin levels you reduce this calcium trap, and this permits the transients to sum up and actually increase the amount of GABA released in the face of repeated firing. If you consider only the parvalbumin change, it would suggest that the inhibitory control over pyramidal cells at their axon initial segments goes up in adolescence, because repetitive firing will lead to more GABA release. That's exactly what has been observed in parvalbumin knockout mice—increased GABA release.

The transporter decline is interesting in light of Gary Westbrook's work. He showed that blocking GABA transporters has little effect on postsynaptic inhibitory currents in the face of single impulses, but when you have synapses that are in close proximity, fire in synchrony, and release transmitter simultaneously, blocking the transporter prolongs the duration of the inhibitory postsynaptic currents. The synapses formed by chandelier cells are precisely such a case. So the changes in parvalbumin and the GABA transporter both suggest that inhibitory control increases. Then on the postsynaptic side, you could interpret it the opposite way.

SRF: The GABA(A) receptors just plummet.

DL: Yes, at least the detectability of GABA(A) receptors that contain an α2 subunit. With my colleague Guillermo Gonzalez-Burgos, we are now using living in-vitro slices of monkey prefrontal cortex to sort out what actually happens physiologically at that synapse as a function of age. The basic question is: If you stimulate a chandelier cell, and you record from the pyramidal cell that receives that input, how does that differ between a 15-month-old and a 45-month-old monkey? (This is the ideal experiment; we are using a less direct one that is more practical.) We have ways of manipulating the transporter and the postsynaptic receptor to try to ferret out what contribution the changes in those actually make. At this point we can say that something really interesting in inhibitory control is happening at that synapse during normal adolescence, but what really is the consequence of that change remains to be determined. Our hypothesis is that the net effect of those changes is to enhance the ability of these chandelier cells to entrain populations of pyramidal cells to fire together.

SRF: And how do you think this is different in adolescents who are developing schizophrenia? Taken together, do your findings support the view that circuitry changes give rise to schizophrenia, or that they unmask a prior disturbance?

DL: These are both very difficult questions to directly address experimentally, and represent one of the major challenges in postmortem human brain research. How do you determine when you examine the brain of a 20-, 30-, or 40-year-old what happened when that individual was a teenager? My best guess is that both disturbances are operative in many individuals with schizophrenia: This circuitry is abnormal due to some combination of risk gene variants and as a consequence the developmental trajectory of these circuitry components is altered during adolescence, perhaps augmented by certain age-related life experiences such as cannabis exposure. We hope to be able to understand the latter type of processes through studies in monkeys.

SRF: We have touched on the chandelier neurons and how they synapse onto the pyramidal cells. What else did you find in your molecular circuitry studies?

DL: Our postmortem studies have operated in three domains. The first is characterizing the nature of the GABAergic disturbance in studies with my colleague Takanori Hashimoto. The second is characterizing the nature of the disturbance in the excitatory pyramidal cells. The third has been a non-hypothesis-driven, or unbiased, approach to figuring out what's going on that Karoly Mirnics, Pat Levitt, and I have been pursuing using microarray technology. We are seeking an eventual convergence across these three domains, but for now each of those investigations operates on a parallel stream and leads to different ideas and hypotheses. We are not pursuing them because we think that they are three different types of schizophrenia, but that these are different ways to get a handle on how the system is disturbed.

SRF: Can you tell me the most important finding in each of these three areas—the GABAergic, the excitatory input, and the microarray-based studies?

DL: Regarding GABAergic disturbance, a subset of GABA neurons are affected and a large proportion of GABA cells appear to be relatively normal. One type of GABA neuron that's affected is the chandelier cell. We have tried to make the case that the disturbance in the networks that these neurons form could contribute to the impairment in synchronized activities in neurons that is measurable by EEG, and which is shown clinically to be altered in people with schizophrenia when they do working memory tasks.

SRF: Fascinating. Expand on that, please?

DL: Sure. Different subclasses of GABA neurons tend to be interconnected with members of their own subclass (though there are some exceptions to that). There are at least five examples of different classes of GABA neurons, each of which is connected both through chemical and electrical synapses with each other, and in other cases just electrically. Those networks of neurons seem to be imbued with the capacity to oscillate at different frequencies. And those different frequencies of oscillation appear to be associated with different network properties that are critical for certain types of brain processes. In the case of parvalbumin-containing neurons that are electrophysiologically fast-spiking (this includes chandelier cells), this oscillatory behavior is in the range of γ band frequency, that is, 40 to 80 Hz. Other populations of GABA neurons oscillate at different frequencies.

From the standpoint of relating the disturbance in these parvalbumin neurons to functional disturbances in schizophrenia, two observations are critical. First, studies by Marc Howard and colleagues (Howard et al., 2003) showed that the power of γ band activity in the prefrontal cortex increases in proportion to working memory load. This activity stays high while the information is being maintained in working memory and then immediately declines once that information in used. If you think of working memory as the ability to acquire a limited amount of information, keep it in mind, then use it and lose it so that you can acquire the next set of information, this γ band activity directly tracks with that process. It goes up when you're acquiring information, stays high while you're maintaining it, and then goes down once you have used it. That shows a potential relationship between what these fast-spiking parvalbumin-containing neurons are specialized to do and an emergent physiological property associated with working memory.

Second, studies show that individuals with schizophrenia have reduced γ band power in the prefrontal cortex when they're doing working memory-type tasks. In combination, these two observations strongly suggest that the disturbances in the fast-spiking neurons might be fundamental to the disturbance in brain wave activity that underlies the disturbance in working memory.

SRF: Let's move to the pathogenesis piece of your approach. We are getting down to the molecules and why GABAergic signaling by the chandelier cells is impaired in various ways. What did you find there?

DL: The idea here is that any given abnormality in the brain could fit into one of four categories for the illness. It could be a cause, it could be a downstream consequence, it could be a compensation to detrimental abnormalities, or it could be a confound that is related to the treatment and not to the disease process. So we asked: Where do these GABA abnormalities fit in? Which of these four "C's" is it?

One thing that would strengthen the idea that the GABA abnormalities really underlie this pathophysiology and contribute to the clinical syndrome would be if we had evidence that there are more proximal processes present in the illness—that is, pathogenetic mechanisms—that give rise to the GABA abnormalities. We often see things that occur together in the illness but usually can't extract cause and effect from those. We therefore searched for changes that correlate with the GABA disturbances and looked for a tractable model system to determine whether what we observe as a correlation in the illness actually is cause and effect. In this case, there was evidence that signaling by the neurotrophin molecule BDNF through its receptor TrkB is important for the development and maintenance of markers of GABA function. This is particularly so for subclasses of GABA neurons that seemed to be altered in schizophrenia, but not for a subclass that we thought was unaffected in the illness. Specfically, the changes in expression of both GAD67 (glutamic acid decarboxylase, responsible for the synthesis of GABA) and parvalbumin were very strongly correlated with the reduced expression of TrkB in people with schizophrenia, but no such relationship was present for the calretinin GABA neurons which are not altered in schizophrenia. So it provided a way to contrast the affected with the unaffected, in this case the parvalbumin with the calretinin neurons.

We took this to a mouse genetic model where expression of TrkB is decreased and asked whether that model shows similar GABAergic disturbance as seen in humans with schizophrenia (Hashimoto et al., 2005). It did, and so, the mouse model provided proof-of-concept evidence that reduced signaling through TrkB could be an upstream "cause" of the cell type-specific alterations in GAD67 expression present in schizophrenia.

SRF: Distinguishing between the four "C's" is notoriously difficult in research across many brain diseases, especially when working with postmortem tissue. Please make that point clear for us with the example of BDNF signaling, down-regulation of the GABA-synthesizing enzyme GAD67, and presynaptic and postsynaptic changes. Assign "Cs" to those steps to illustrate how you've tried to tease them apart.

DL: You have to put "cause" very much in quotation marks because we have to push further upstream to understand what causes the loss of TrkB. That said, I'd venture that a reduction in signaling through the TrkB receptor is a cause for a decrease in GAD67 synthesis, which is a consequence. The changes in presynaptic parvalbumin and GAT1 and postsynaptic GABA(A) α2 subunit are compensations. Our studies in humans and antipsychotic treatment of nonhuman primates indicate that none of those findings are confounds due to the treatment of the illness.

SRF: That's a great example. More generally, how do modern research methods now enable scientists to tackle the problem of the four "Cs"?

DL: Two thoughts here. For one, it helps if you recognize the need to triangulate. On any given abnormality, the more independent lines of evidence you can bring to bear on the same interpretation, the stronger the argument.

Take the confound issue, for example. We would ideally like to study only people who were in their first episode of schizophrenia and had never taken antipsychotic medications. That's impossible. So how can we assess what the drug effects might be? One thing we do is look at people who were on or off treatment at the time of death. If both groups show the abnormality at hand, then that's consistent with it being disease process, not treatment. A limitation to that is we don't know how long the drug effects last. Another way to triangulate is to look at individuals with other illnesses who were treated with the same medications. If they don't have the abnormality, then that is a second line of evidence that it's not due to the drugs but the disease process. The third line of evidence involves exposing monkeys to these medications in a fashion that exactly mimics their clinical use. If those monkeys don't have the abnormality, then that again is consistent. You've got to recognize these are monkeys, not humans, and they are normal. So each independent experiment has its own set of limitations. But if they all converge upon the same finding, that gives you greater confidence in the conclusion.

For another, you have to recognize that you can only ask certain questions in each type of experimental mode. Postmortem human brain is critical to ask certain questions, and uniquely suited for them. But it has limitations for other types of questions, so you need to be able to move to a model system like the monkey, which is the closest we can get to the circuitry of the human cortex. The prefrontal cortex differs a lot between primates and rodents, so there are clear limitations to rodents. Then again, there are questions one can only ask in rodents. We try to have a certain degree of agility, and use the system best suited to the particular question we want to address.

SRF: To hark back to a question above, what did you find with regard to excitatory disturbances in schizophrenia and with your microarray research?

DL: We think there is clear evidence of disturbances in glutamatergic systems and excitatory function in the prefrontal cortex. The question we've been trying to get at is whether there is a subset of pyramidal cells that participate in particular circuits that are preferentially affected. Our data to date suggest that pyramidal cells located in deep layer 3 of the cortex may be more severely affected than pyramidal neurons in other layers. Our ongoing research efforts are in two domains. One is to find out whether we can identify molecular markers of the seemingly more vulnerable neurons in order to facilitate their study. The other has been to ask if there is something about the connectivity of these neurons, either in terms of the inputs they receive or where their axons go, that would help us understand functional disturbances we should expect to see as a result of morphological abnormalities.

SRF: Have you gotten answers to those questions yet? For example, you just had a paper about dendritic spines of layer 3 pyramidal neurons (Kolluri et al., 2005).

DL: This paper followed up a previous study where we showed that the dendritic spines were markedly reduced in density on pyramidal cells in deep layer 3 but less so in superficial layer 3 (Glantz and Lewis, 2000). In this recent study, we didn't see the same abnormalities in layers 5 and 6 of the same subjects, so it was a negative finding that supported the idea of specificity in either cell type or circuitry disturbances. So we recently looked at the expression in schizophrenia of genes that encode protein products important for spine formation and maintenance (Hill et al., 2006; see SRF related news story). One of our other ideas was this: Given that these deep layer 3 pyramidal cells are situated in the input zone of projections from the mediodorsal thalamus, perhaps these changes are downstream of abnormalities in the thalamus? Right now, the literature is mixed on that. There are studies showing functional disturbances in that portion of the thalamus, as well as imaging studies reporting that this portion of the thalamus is smaller. However, the postmortem studies are now evenly divided between ones reporting fewer neurons in the mediodorsal thalamus and others reporting no change in neuron number, and our study (Dorph-Petersen et al., 2004) falls in the latter category. That's in terms of the number of neurons; it doesn't mean that the inputs can't be disturbed but makes it a bit harder to sort that out. So we're looking for alternative ways to ask about the functional integrity of thalamic inputs to the prefrontal cortex.

SRF: How do thalamic inputs change during development and in adolescence? How does synaptic pruning tie in, especially in children who are on their way to develop schizophrenia? Postmortem studies may miss this if they look at tissue 30 years later; if by that time neuron number is down a tad, it's hard to interpret what that means.

DL: I agree. One of the strengths of postmortem human research is that it's the only way to get at the molecular, cellular, and circuitry level in schizophrenia. But if you imagine the life of a person with this illness as a motion picture of epic length, we're really trying to figure out that whole story with but one frame of the movie, and a crumpled one at that. That's where agility in terms of the preparations used becomes important.

SRF: What is known about whether the pruning of synapses, presumably on pyramidal neurons in prefrontal cortex, is different in schizophrenia? Whether children with schizophrenia lose more synapses in that area or not?

DL: The findings from convergent data are consistent with there being fewer excitatory inputs to these pyramidal cells. The open question is whether that represents an overexuberant pruning process or a normal pruning process on top of a deficient formation of those synapses, where you didn't form enough to start with and then prune the same amount.

SRF: Presumably, chandelier neuron synapses with a GABA deficit can't compete for survival as well as other neurons with stronger, more coordinated activity. Would all that you found about the deficits in the chandelier neurons not be consistent with the idea that they are pruned at a greater rate because they don't fire as well?

DL: Let me mention two ideas related to this that we're interested in. The first is whether the GABAergic disturbances indeed are proximal to the excitatory disturbances. Several findings from basic research support this hypothesis. For example, the work of Takao Hensch at the RIKEN Institute, who looked at the development of ocular dominance in the visual cortex, showed that GABA is essential for the onset of the critical period, which shapes the excitatory inputs that allow one eye to predominate over another in a particular column. He also showed that perturbing those GABAergic functions changes spine densities. It's quite an extrapolation to move from that system to the prefrontal cortex. But at least it provides a sort of proof of concept that if, say, this deficiency in GAD67 that has been repeatedly observed in schizophrenia is something that happens early in life, then one could at least propose a mechanism whereby one of the consequences of that would be a disturbance in excitatory circuitry. Interestingly, at least one group has associated allelic variance in the GAD67 gene with schizophrenia (DeLuca et al., 2004).

SRF: Because the activity-dependent pruning would be different?

DL: Yes. An alternative—not necessarily mutually exclusive—view comes out of our gene array studies. One of the initial observations we made in Karoly Mirnics's first paper on the topic (Mirnics et al., 2000) is that there seems to be a conserved deficiency in the levels of mRNA transcripts for genes that encode proteins that are involved in the machinery of presynaptic neurotransmitter release. If these gene expression abnormalities reflect an early process, a person who is going to develop schizophrenia later in life could have a disturbance in the strength of synaptic function. The excess number of synapses present before adolescence may be sufficient to compensate for this decrease in function, but when the pruning process hits in adolescence, then two things would occur. First, you would lose this compensatory ability, and second, dysfunctional synapses might be excessively pruned. We published this as a hypothesis that we thought was intriguing but was an extrapolation from existing data (Mirnics et al., 2001). We think it could provide a mechanistic explanation for how you get deficient pruning and a clinical syndrome after adolescence, but also a brain that doesn't quite work right prior to adolescence. In terms of cognitive function, this appears to be a feature of schizophrenia. These kids seem to have cognitive disturbances very early in life, many years before they become clinically ill.

SRF: Karoly Mirnics and you have pioneered microarray studies in schizophrenia. After much initial excitement about these techniques, some disillusionment set in around issues such as sample quality, validation, interpretation. What's your view today? Have the fishing expeditions been useful exercises, for your lab, anyway?

DL: Absolutely, but we have to recognize two caveats. The first is that the level of analysis has to be continually refined. In our first studies, we ground up the entire cortical gray matter and tried to extract signal, and we did so on platforms that had limited sensitivity. Now, more focused dissections of tissue are used, whether that be individual cortical layers or more homogeneous cell populations in combination with platforms that have greater sensitivity to try to improve the signal. This advance is clearly improving the quality of microarray findings; we can now better interpret and verify them.

The second issue was known from the outset but tended to get lost in the buzz. It is that these findings give you just levels of transcripts in the tissue. They tell you nothing about how those abnormalities arise or what their consequences are for the circuitry. Any microarray finding has to be placed in this context.

In our case, we found underexpression of the gene regulator of G-protein signaling 4 (RGS4). It popped up on the arrays without prior hypothesis. Curiously, the gene was on the locus that Linda Brzustowicz had shown to be highly associated with schizophrenia (Brzustowicz et al., 2000), and the protein had a very interesting function with regard to transmitter pathways thought to be involved in schizophrenia. So the microarray finding in concert with these other data suggested that this might be a susceptibility gene for the illness. Indeed, studies from Vish Nimgaonkar and our collaborators here at Pittsburgh as well as other groups have shown that RGS4 is a potential risk gene for schizophrenia. However, as with all the other putative risk genes, there are problems in interpreting different haplotypes, finding pathogenic SNPs, etc. (Talkowski et al., 2006). But at least it serves as an illustration of how this expedition approach can potentially yield valuable fish of different species, one of which might be risk genes.

SRF: Let's finish summarizing your translational research program. The last section on your Web site is about treatment interventions. You describe a small trial at your site. What is it you're testing?

DL: The idea is that these GABAergic disturbances in the PFC give rise to a decrease in γ band oscillation, and that leads to the working memory impairment. So the trial is designed to ask: Can we increase GABA signaling at the axon initial segment of pyramidal cells in a manner that preserves the timing of the input? The timing is critical. We had to think about a way to approach this that would not just cause chandelier cells to fire more often, and that wouldn't increase GABAergic signaling activity broadly. Either could do more harm than good. We went after a compound that would be selective for GABA(A) receptors that contain an α2 subunit. We also wanted a positive allosteric modulator, that is, a drug that would increase chloride ion flow through those receptors only when GABA was normally present. The idea is to boost the signal of chandelier cells only when chandelier cells are firing. We identified an existing compound through a pharmaceutical company and are conducting a double-blind, placebo-controlled, randomized trial. The output measure is a global assessment of cognitive function, as well as a particular assessment of γ band activity during a task that taps prefrontal function. The silver bullet would be that people who take this medication show a global improvement in cognitive function that makes a difference in their daily lives. I think a more plausible outcome would be evidence that this compound improves prefrontal physiology in a way that would be consistent with improvement in cognitive function. Whether this compound itself will turn out to be clinically useful is unknown at present.

SRF: …so it's not a drug that is currently used.

DL: Correct.

SRF: It's an experimental compound. Is it a form of a benzodiazepine? I thought that GABA(A) receptors are already the target of benzodiazepines.

DL: Correct. The limitation of currently available benzodiazepines is that they have activity at GABA(A) receptors that contain α1, α2, α3, or α5 subunits. For example, it is clear that activity at α1 produces the sedative component of benzodiazepines, and being sedated is not particularly good for cognition. Likewise, activity at α5 impairs hippocampal function, which is also bad for cognition. So we think that even though there is some benefit to using currently available benzodiazepines in the treatment of schizophrenia, the cognitive costs outweigh that benefit.

SRF: Numerous GABA(A) agonists are under development in several companies, but more often for epilepsy than for schizophrenia. Targeting the α2 subunit is quite a new thing. Is that right?

DL: Yes.

SRF: What is upstream of TrkB in schizophrenia?

DL: We don't know yet. One of the things we do know about TrkB is that its expression is activity-dependent. So one scenario is that NMDA receptor hypofunction—a popular hypothesis in the disorder—could be upstream of that. This is, right now, the most viable hypothesis, and we are trying to test it.

SRF: Why not try to treat upstream and devise TrkB agonists?

DL: There are two reasons why we haven't gotten excited about pursuing that. We think the deficiency in the expression of the TrkB receptor is what's really driving the abnormality. If the receptor is deficient, can you actually have enough ligand to appropriately enhance its signal transduction? If the critical abnormality were production of BDNF and the TrkB receptor were normal, then that would be a more viable approach. The other reason not to do it is that we want to apply the treatment as close to the physiological disturbance as we can get in order to attack the critical functional piece and to minimize the friendly fire, if you will.

SRF: On that—loss of signal versus loss of receptors—I'd like to talk about mouse models. If I got it right, you studied the TrkB underexpressors from Louis Reichardt's lab as well as inducible BDNF knockouts made in Lisa Monteggia and Eric Nestler's lab, and you found differing effects in them that were very interesting. Exactly what did you measure and what did you find?

DL: We used the same measures that we'd used in the postmortem human studies. That is, we looked at transcript levels of GAD67 and parvalbumin, and we have extended that to other markers now in unpublished studies. In both the embryonic and the adult BDNF knockouts, we failed to see a change in either of those transcripts in the prefrontal cortex. But in the TrkB hypomorph, both transcripts were down in a gene dose-related fashion, so the homozygotes that underexpress TrkB to a greater extent had a greater loss of GAD67 and parvalbumin than did the heterozygotes. And, at the cellular level, the deficit in GAD67 transcripts exactly recapitulated what we observed in schizophrenia, that is, a subset of GABA neurons so underexpress this transcript that we can't detect it, but the remaining GABA neurons express GAD67 normally.

SRF: Does that mean Reichardt inadvertently made a partial schizophrenia mouse model?

DL: I prefer to be circumspect when talking about schizophrenia models. This animal mimics a component of a cause (TrkB) and a consequence (GAD67/parvalbumin) in the illness. I can't say at this point that it bears a relationship to schizophrenia beyond that. We are breeding those animals here and want to ask if they show altered γ oscillatory behavior.

SRF: I was going to ask you if the mice have a deficit in working memory and if you can even measure that in mice.

DL: The general feeling is that mice are not the sharpest rodents in the cutlery drawer. Some [researchers] have questioned the extent to which you can actually tap into working memory in mice. But there are other ways to look at γ oscillations besides in the context of a behavioral task. For us, the question to start with is just: Do these animals lack the machinery necessary to generate γ oscillations under conditions where wild-type animals can?

SRF: More generally, are there any useful mouse models for at least aspects of schizophrenia? Alzheimer research flourished tremendously once the mouse models came along, even though they were only partial models, too. Now there seems to be a new model every month.

DL: The key thing for Alzheimer disease was that there were specific molecules and neuropathological abnormalities one could study; it wasn't dependent solely on clinical outcomes. I think we're moving in that direction in schizophrenia, and mouse models are going to turn out to be critically important tools in understanding the disease process. But we are still at an early stage. My own bias is that mice are going to prove to be useful in the relationships between molecular abnormalities and physiological measures, perhaps less so in the realm of behavior.

SRF: True, schizophrenia is much more complex when it comes to behavior than Alzheimer disease. But even there, in the beginning there were skeptical discussions on how one can know a mouse is demented when it's not too bright to begin with. You have worked on many fronts and I'd like to leave our readers with a summary idea. I'll take a stab at recapping the chain of events we've touched on in this conversation. Different lines of investigation that have moved forward in parallel have come up with data to suggest that reduced TrkB neurotrophin signaling in chandelier neurons leads to reduced GAD67 expression, that is, less GABA in parvalbumin chandelier neurons in prefrontal cortex. Presynaptic GAT1 expression in chandelier cell cartridges goes down; GABA(A) α2 receptor expression of pyramidal cell axon initial segments go up in an effort to make the most of the limited GABA available. This compensation is insufficient, and you end up with impaired inhibitory transmission that controls the timing of the output of pyramidal neurons in layer 3 that project to other areas of association cortex. This throws off the synchronous firing of these pyramidal cell networks. In turn, you weaken the γ band oscillation during working memory tasks, and that is one of the problems that's been observed in people with schizophrenia. You also think that synaptic pruning and the marked changes in these GABA connections during adolescence unmask disturbances in these systems that have been developing earlier in life. This is a gross oversimplification, and you would hasten to add that it is but one of many things that are happening in this illness. But is it basically the gist of your results to date?

DL: Yes. You've done a great job of capturing what took many more words for me to express. I would add only that the ultimate test of the value of this strategy for trying to dissect an aspect of the disease process of schizophrenia is whether it tells us the best molecular targets for new drug development and how such drugs should be designed to manipulate those targets.

SRF: I thank you for your time.

DL: It's been my pleasure.