Comments on Online Discussion
Comment by: Craig M. Powell
Submitted 21 April 2010
Posted 21 April 2010
The conclusion of this paper, that we need better behavioral tests of higher-order cognitive function in rodent models to better mimic cognitive dysfunction in schizophrenia, does not seem very controversial. Currently, it seems most important to create a model with high constructive validity, to understand how brain function is perturbed in each model, to understand in detail how behavior is altered in each model, and to try to reverse brain function abnormalities to "treat" the behavioral abnormalities. Any efforts to create high-throughput cognitive tasks in rodent models are welcome and should be incorporated into the search for behavioral abnormalities in these disease models.
View all comments by Craig M. PowellComment by: Jared Young
Submitted 6 May 2010
Posted 6 May 2010
The review of Arguello and Gogos provides a timely reminder of the importance of assessing the cognitive performance of genetic models of schizophrenia. Moreover, they provide a very nice oversight on transgenic/mutant lines available to date to investigate cognition in schizophrenia and the work that has been done to phenotype these mice.
The majority of transgenic mice discussed, however, are not specific to mutations observed in schizophrenia patients. Although they do discuss DRD2 knockout mice, it is surprising they did not discuss α7 nAChR knockout mice, despite the obvious relation to attentional performance.
Another surprise was that despite the admission that animal tasks of working memory likely measure short-term, not working memory, a large part of the review discusses evidence of performance of transgenic animals in tasks that have no proven efficacy for the working memory domains listed by CNTRICS. In fact, the only animal task listed by Barch et al. (2009)—when reporting working memory task selection for the MATRICS—was from Haddon and Killcross (2007), and none of the mice from the review were assessed in this task. We recommend our recent review as an alternative, in which we identify cognitive tasks in rodents that provide construct validity for human testing (Young et al., 2009).
This highlights, beyond even the authors’ claim, that we do need better behavioral tests of higher-order cognitive function in rodent models to better mimic cognitive dysfunction in schizophrenia. There are now an abundance of models, as mentioned in the review, but the tasks they are being tested on rarely exhibit cross-species translational validity. Moreover, the tasks that are developed for mice are rarely fully validated as lesion/inactivation studies and are predominantly performed in rats, not mice.
The development of these tasks as high-throughput tasks remains unlikely, however. Unfortunately, it appears that for the foreseeable future, to generate the cognitive data from each mutant line, the hard work will need to be done.
I would submit that rather than investigating the answer to Jim Koenig's first question (although treatment is extremely important), the question should perhaps be rephrased into two parts:
1a. How do these phenotypes from specific transgenic lines fill in details of the watershed model of endophenotype investigation from Cannon and Keller (2006)?
1b. What subgroups of schizophrenia exhibit a pattern of cognitive deficits similar to these transgenic lines?
The authors make a valid point that the generalizability of cognitive deficits of these transgenic mice should also be examined against other neuropsychiatric disorders. This is especially true given that several of these genes that confer susceptibility to schizophrenia are also relevant to bipolar disorder (such as DISC1).
I think what must also be made clear is that assessing cognition in transgenic animals could be more productive if subgroups of schizophrenia patients with deficits in specific cognitive domains were more explicitly identified and genotyped.
Finally, I think pointing out that the recent R21/R33 RFA and following PA from NIMH to develop “Mouse Models Containing Human Alleles” may provide a lot more mice to be phenotyped very soon.
The authors do provide a timely review, and highlight the need for the development and validation of cognitive tasks in mice. We can only hope the updated review in five years time will prove to be a lot more extensive.
Barch, DM, Berman, MG, Engle, R, Hurdelbrink Jones, J, Jonides, J, MacDonald III, A, Evan Nee, D, Redick, TS, Sponheim, SR (2009) CNTRICS final task selection: Working Memory, Schiz Bull, 35(1), 136-152. Abstract
Haddon J, and Killcross, S (2007) Contextual control of choice performance: behavioral, neurobiological, and neurochemical influences. Ann N Y Acad Sci, 1104, 250-269. Abstract
Young JW, Powell SB, Risbrough V, Marston HM,
Geyer MA.(2009). Using the MATRICS to guide development of a preclinical cognitive test battery for research in schizophrenia. Pharmacol Ther. 122, 150–202. Abstract
Cannon TD, Keller MC, (2006). Endophenotypes in the genetic analyses of mental disorders. Annu Rev Clin Psychol, 2, 267–290. Abstract
View all comments by Jared YoungComment by: Ina Weiner
Submitted 19 May 2010
Posted 19 May 2010
Mouse models of schizophrenia susceptibility genes are crucial for studying the biological functions of genetic variants implicated in schizophrenia association studies. Arguello and Gogos's timely and informative review illustrates how such models are used to study the specific molecular alterations that may underlie some of the cognitive deficits of schizophrenia, or "the cognitive architecture related to genetic risk for schizophrenia."
The paper illuminates several issues worthy of consideration in this type of modeling. One is the issue of construct validity. As pointed out by the authors, when we evaluate cognition in genetic mouse models, we have two sources of construct validity: one that derives from the inducing factor/independent variable and one that derives from the behavioral manifestation/dependent variable: "…measuring relevant cognitive processes in animal models usually relies on the neuronal and psychological homology of these processes…. Using these criteria alone, however, may be problematic…. Additional support, like a solid genetic foundation, is needed for animal models to provide insight into a disease process." If both the induction and the measure have reasonably strong/accepted construct validity (e.g., amphetamine leads to disrupted PPI), induction-measure models are particularly strong. In some models, the construct of induction alone suffices: amphetamine-induced hyperactivity is a widely used and trusted animal model of positive symptoms although hyperactivity has no construct validity. However, when construct validity of both the induction and the measure are debatable, the results are difficult to interpret. The critical question becomes, What does one conclude if mutant mice do not show the expected cognitive deficits? Do we conclude that it is the specific genetic lesion, or the behavioral manifestation that is not relevant to cognition in schizophrenia? I presume the last would be the typical (although not necessarily the correct) choice. This brings me to the second issue: animal cognition (or learning or behavior).
Is there a problem with animal cognition (or learning or behavior)? The authors indicate that 1) it is still not clear whether cognitive dysfunction in schizophrenia reflects discrete independent deficits or a generalized functional impairment; 2) the genetic architecture of psychiatric disorders is still unknown, and yet the conclusion is that the problem lies in the cognitive tasks used in animals, and this is echoed in the comments of both Powell and Young: "…this highlights, beyond even the authors' claim, that we do need better behavioral tests of higher-order cognitive function in rodent models to better mimic cognitive dysfunction in schizophrenia."
I would like to suggest an opposite idea, namely, that the one solid foundation we have for animal modeling is the knowledge on animal behavior, hard-won from many decades of meticulous and extremely sophisticated empirical research that has excelled in the development of procedures for controlling the contribution of one process in order to reveal the operation of another. And, of course, this is the field that at the beginning of the previous century, stressed (in fact, too strongly) the idea that there is no dividing line between animals and humans. The starting point of Watson's "new psychology" was the "…fact that organisms, man and animal alike, adjust themselves to their environment." Over the course of the years since Watson uttered those words, thousands of experiments in hundreds of laboratories have been conducted to tease out an enormous number of variables influencing and determining animal behavior/cognition, demonstrating that rodents have impressive learning/cognitive capacities, including propositional knowledge (1) and causal reasoning (Blaisdell et al., 2006). MATRIX and CNTRICS experts have done an important job selecting specific paradigms that will foster and speed up cognitive endophenotyping and ensure a common language in animal modeling of cognition. There is a vast and rich arsenal of paradigms left to dig into, learn, and choose from for probing the cognitive domain using genetic and other models. We do need to invest efforts in developing cognitive phenotyping in mice, but extensive progress has been made in this domain (Haddon et al., 2007). The major task is to establish etiologically strong genetic models. Mouse models that express human genes or human genetic elements hold great promise for advancing the field.
Another point regarding the similarity between animal and human cognition/behavior: the authors discuss the problems of modeling WM in animals, and this is underscored by Young's comment. At the empirical level, given that there is one WM task that is accepted as measuring "genuine WM" (Haddon et al., 2007), then a comparison between performance on this task and STM tasks (delayed matching-to-sample) in several lines of mutants seems like a straightforward way to solve the question. There still remains a theoretical question of how much emphasis is given to different component processes within a cognitive domain in human and animal measures, and to what extent lack of clarity regarding full parallel is viewed as a hindrance to successful parsing of genetic contribution to cognition. WM in humans can be defined in terms of several components, in particular, mental manipulation of information (central executive functions) and maintenance of information. Is it critical that what we measure as animals' WM features all of the components? Or what we need to focus on is similarity of fundamental processes, i.e., the brain's capacity to hold information and use it online for solving a current problem, or executive control (unless executive control in WM is fundamentally different from executive control animals use in other tasks?). Of course, the specifics of these processes will differ among humans, rats, and mice, and most of all will differ as a function of specific tasks and task parameters in each of the species. But if we can be reasonably sure that the measured behavior reflects the same fundamental processes (based on similar effects of environmental and neural manipulations), this should suffice at present for the assessment of the relationship between genotype and cognitive process/endophenotype (Kellendonk et al., 2009). In fact, it is possible that fundamental processes are what we should search for. As noted by the authors, the pervasiveness of cognitive deficits in schizophrenia "…suggests a fundamental difference in some elementary and ubiquitous mechanism." It should also be kept in mind that whatever the level of resolution we want from cognitive domains must be paralleled by the same level of resolution at the genetic level. The latter seems a rather remote goal given that limited cognitive phenotyping has been done to date even in the available mutant models, and that specific mutations lead to multiple cognitive deficits that are not necessarily related.
The authors conclude with a very important statement regarding the utility of animal behavior for animal modeling: "A behavioral task may be extremely useful, regardless of its resemblance to any clinical test, if it is sensitive to some underlying disease process and predicts the clinical efficacy of therapeutic intervention."
From my perspective, what is missing in the review is an integrative summary of what can be learned from the reviewed data. After succinctly and incisively summarizing the problem of cognition and genetic liability to psychosis, the authors write, "The study of the mutant models reviewed above affords the opportunity to identify whether deficits within the cognitive domain converge merely at the behavioral level or whether there are common underlying neural correlates, and if so, how this may be related to diverse clinical phenotypes." I wish the authors had given us their insights on this (and a corresponding table).
Finally, I'd like to offer a few words regarding high throughput, since all three commentators referred to this issue. Unfortunately, complex behavior and high throughput do not coexist. In fact, I would venture a proposition that the high-throughput focus has been responsible for the fact that neurogenetics (and psychopharmacology) have neglected complex animal behavior. As pointed out by Arguello and Gogos, if complex behavior is integrated into neurogenetics (and psychopharmacology), this will advance significantly the understanding of genetic underpinnings of cognitive deficits in schizophrenia as well as novel drug discovery. However, contrary to what seems to be the popular perception—that everyone can do behavior—assessing complex animal behavior and interpreting it require training and expertise, and these should be explicitly promoted in neurosciences.
Lastly, I would like to bring to the readers' attention our recent paper on prevention of brain structural abnormalities (assessed with MRI) in an animal neurodevelopmental model, which I think is of interest given the huge surge of focus on prodrome and prevention in schizophrenia.
Piontkewitz Y, Arad M, Weiner I. Risperidone Administered During Asymptomatic Period of Adolescence Prevents the Emergence of Brain Structural Pathology and Behavioral Abnormalities in an Animal Model of Schizophrenia. Schizophr Bull . 2010 May 3. Abstract
1. Dickinson A. (1980). Contemporary animal learning theory. Cambridge: Cambridge University Press.
2. Blaisdell AP, Sawa K, Leising KJ, Waldmann MR (2006) Causal reasoning in rats. Science 311:1020-1022. Abstract
3. Haddon J, and Killcross, S (2007) Contextual control of choice performance: behavioral, neurobiological, and neurochemical influences. Ann N Y Acad Sci, 1104, 250-269. Abstract
4. Kellendonk C, Simpson EH, Kandel ER (2009) Modeling cognitive endophenotypes of schizophrenia in mice. TINS 32:347-358. Abstract
View all comments by Ina WeinerComment by: Jo Neill
Submitted 27 May 2010
Posted 27 May 2010
This excellent article is an extensive review of the relationship between mouse models of susceptibility genes for schizophrenia and their cognitive performance in tests of relevance to schizophrenia and the clinical situation. The article provides a considerable amount of relevant information concerning the schizophrenia susceptibility gene mouse models, and also highlights some of the wider issues in this field. One of the strengths of this review is the loud acknowledgement of the difficulties and inconsistencies in the field such as the uncertainty concerning which actual genes and alleles contribute most to the pathology of schizophrenia and the mechanism(s) by which they do so. One concern I think many researchers have about this work is the large number of schizophrenia susceptibility genes that have been identified, but their real significance in the clinic, and indeed usefulness, is unclear, except perhaps in the identification of certain rare alleles that account for a small number of cases as described in this paper.
This review also illuminates some broader issues surrounding animal model development, an activity clearly best suited to the most tenacious of researchers! The MATRICS, and more latterly, CNTRICS initiatives, have made great advances towards highlighting the relevant tests, target molecules, and systems for cognition in schizophrenia. The authors make some very valid points in their introduction, firstly that the clinical tests for assessing cognition in schizophrenia are problematic in terms of the specificity of the tests and the tools available. This is also the case with the animal tests. More effort is required in validating the tests for each domain of cognition affected in the illness. The authors use working memory as one example. In my experience, there is much debate about how this may be measured in animals and, indeed, whether it can be measured at all. The authors of this review raise the measurement issue, and suggest that many tests (and testers) confound short-term and working memory. Certain tests assume the use of working memory by their subjects (e.g., delayed non-matching to position and object/place/context recognition tests). They may well do so, but require more thorough validation to evaluate this. The question arises whether working memory can be measured in animals out of a spatial context (it's quite possible that it cannot). Therefore, the tests that we routinely use in the animal laboratory require careful validation. This is perhaps particularly important when faced with the uncertainty about the exact genetic alterations relevant for the disease.
The authors provide an insightful overview of the available information on cognitive deficits observed in the various tests in several mutant mouse models. One feature of this work that the authors clearly demonstrate is the need for further study. For example, only one (mismatch negativity for perception) or even none (attention) of the mouse mutant models "in the top 30 gene list" has been tested. While for some other tests, like attentional set shifting for executive function, and certain putative tests of working memory, a wider range of mutant mice have been tested, but the results are often conflicting and inconsistent. The difficulty lies in variations of the tasks used between laboratories, and their interpretation. CNTRICS has been particularly helpful in identifying the animal tests that translate best to the cognitive domains impaired in schizophrenia.
The challenges for this field are numerous: 1) to determine the most relevant and valuable mutant models, that is, which gene(s) and alleles are critical for the illness; 2) to thoroughly study these mouse models at all stages of development, i.e., pre- and post-adolescence and in middle and old age; 3) to determine sex differences; 4) to investigate neurobiological changes of relevance to schizophrenia induced by the genetic mutation; 5) to construct an ethogram so that normal behavior patterns are established before the cognitive challenge; 6) to test the mutant animals in thoroughly validated and well-understood tests for cognition of relevance to the illness; and finally 6) to assess pharmacological reversal of these deficits (for predictive validity) for which the data seem particularly sparse. This work is vital and will allow subsequent evaluation of novel targets and therapies in these animal models of the disease.
View all comments by Jo NeillComment by: Alexander Arguello
Submitted 21 June 2010
Posted 21 June 2010
We are grateful to SRF for featuring our review in this Journal Club series and thus providing this opportunity to answer questions and to discuss issues that, due to the narrow scope of the review and space limitations, we could not address. There were two major areas addressed by our review as highlighted by Dr. Weiner: 1) the creation of genetic models and 2) their behavioral characterization.
With regard to the first part, many commented on the genetic complexity of schizophrenia. I share the view that rare, highly penetrant mutations associated with disease are the best way forward for making rapid progress in understanding disease pathogenesis. While many common variants across the genome (like those listed in the SZGene database) may confer some small disease risk, it is just much easier to faithfully model and gain insight from a rare allele whose functional effect is known and whose effect size is large. That’s not to say common variants will have nothing to contribute to treatment, but in terms of genetic risk, rare variants provide the most straightforward approach for etiologically valid animal models.
With regard to the second part, which most of the comments addressed, we still need reliable, disease-relevant endpoints to anchor the characterization of animal models. After all, for any given allele there may be downstream effects that are not relevant in terms of core disease pathogenesis but contribute instead to the heterogeneity observed in syndromes like schizophrenia. Behavioral endpoints are the most disease relevant because, at least for now, psychiatric disorders are classified according to behavioral disturbances and not by genetic or biological markers. This begs the question, What behavioral endpoints should be used? Cognitive deficits are a prime choice because at least some of them are related to genetic risk and functional outcome, and they can be assessed in animal models. It would not be surprising if some models do not exhibit cognitive impairments since not every risk allele on its own will produce the totality of behavioral impairments seen in schizophrenia. But these models may nonetheless be compared at different levels of analysis with models that do show cognitive deficits.
Given the focus on cognition, MATRICS emerged to facilitate the screening of new drugs for their impact on cognitive function in the hope of improving functional outcome in patients. But the MATRICS consensus cognitive battery that was formed is, self-admittedly, a quick solution. The nominated tasks aren’t optimized for parsing out specific cognitive processes. This, however, may be beside the point if the goal is testing the efficacy of novel compounds. The review mentioned by Dr. Young does an excellent job of translating the recommendations of MATRICS into a test battery for animal models. Again, the major hope is that these tasks may be predictive, at a preclinical level, for compounds effective in the MATRICS battery and ultimately in improving the daily functioning of patients.
Nevertheless, there is still a need for identifying potential drug targets before the resulting compounds are screened. Animal models of rare alleles are poised to make a substantial impact by establishing a link between a risk mutation and specific cognitive deficits. These models, with high disease fidelity at both the genotype and, hopefully, the phenotype level, will allow us to dissect out the mechanistic link between mutation and behavior. For example, cognitive neuroscience uses rigorous, quantitative measures of behavioral performance in the context of well-designed tasks to deconstruct a psychological construct into its component cognitive processes. Knowing how these processes depend on distinct neural circuits and/or mechanisms can help narrow down the search for potential substrates underlying any behavioral impairment. Of course, it’s important to keep in mind that the rodent brain will get us only so far (Nature, 2010; Abbott, 2010). The goal should be carefully characterizing the mechanisms underlying rodent cognition and deciphering how similar or different those are to mechanisms underlying comparable cognitive processes in primates.
The utility of this approach can be seen in the case of working memory deficits in schizophrenia. As a group, patients seem to have problems in encoding and manipulating information, but not maintaining it in working memory (Cannon et al., 2005; Gold et al., 2010). They also have an overall reduced working memory capacity which itself may be influenced by elements of cognitive control (Gold et al., 2010; Vogel et al., 2005). Thus, tasks that can tease apart these processes are needed to better understand the neural changes that may contribute to working memory deficits in schizophrenia and in other disorders as well. Perhaps, therefore, it shouldn’t be a question of what’s the best working memory task, but rather of designing tasks that isolate and measure the different components of working memory.
Dr. Neill makes an important point. If schizophrenia is associated with working memory deficits across sensory modalities, rodent tasks need not and should not be restricted to the spatial domain. It is possible to design tasks that require holding visual, auditory, olfactory, and somatosensory information in working memory. Dissecting out different working memory processes across sensory domains will allow us to disentangle whether working memory deficits in different models are really the same or only superficially similar.
Dr. Koenig raises the issue of high-throughput assays. The above approach is certainly not a high-throughput one. But even in patients, early cortical processing like visual and auditory perception can influence such complex processes as social cognition. Can such a bottom-up approach (looking at how early cortical processing influences higher, late-stage processing) be exploited in animals? Can simpler perceptual measures predict performance in more cognitively demanding tasks? PPI and MMN as paradigms have been widely used, but does the lack of progress in drug discovery using these measures reflect poorly on the measures themselves or on models to which they’ve been applied?
More generally, is cognition per se the key to understanding and treating psychiatric disorders? A minority of patients with schizophrenia can have normal intellectual abilities yet still suffer from psychosis. It is likely true that, had they not developed schizophrenia, their intellectual function would have been greater. But it doesn’t seem that there’s a threshold level of intellectual function that predisposes to psychosis. A common process may be contributing to both, but this has yet to be demonstrated empirically.
Dr. Young mentions patient heterogeneity. This is evident both at the genotype and phenotype level and is thus likely true of pathogenesis itself. A single therapeutic target for all schizophrenia patients is the Holy Grail of treatment, but is this realistic? It’s an open question whether we can treat common behavioral endpoints, like cognitive changes regardless of specific neural changes, or if we must treat the underlying, perhaps patient/mutation unique pathophysiology responsible for behavioral deficits. Do the effects of rare alleles associated with schizophrenia converge at some more tractable molecular/cellular level? Do all roads lead to Rome? Or will we be lost on the way to Damascus?
Cannon TD, Glahn DC, Kim J, van Erp TG, Karlsgodt K, Cohen MS, Nuechterlein KH, Bava S, Shirinyan D. Dorsolateral prefrontal cortex activity during maintenance and manipulation of information in working memory in patients with schizophrenia. Arch Gen Psychiatry . 2005 Oct 1 ; 62(10):1071-80. Abstract
Gold JM, Hahn B, Zhang WW, Robinson BM, Kappenman ES, Beck VM, Luck SJ. Reduced capacity but spared precision and maintenance of working memory representations in schizophrenia. Arch Gen Psychiatry . 2010 Jun 1 ; 67(6):570-7. Abstract
Vogel EK, McCollough AW, Machizawa MG. Neural measures reveal individual differences in controlling access to working memory. Nature . 2005 Nov 24 ; 438(7067):500-3. Abstract
[No authors listed] Nature. 2010 May 20;465(7296):267. No abstract available.
Abbott A. Neuroscience: The rat pack. Nature . 2010 May 20 ; 465(7296):282-3. Abstract
View all comments by Alexander Arguello