- Email: [email protected]
Does schizophrenia have a cause?
Update
478
TRENDS in Cognitive Sciences
pairings may be equal overall, people tend to learn a stronger association for one of the unambiguous cues if it tends to appear later in the training corpus – that is, if it is highlighted with respect to the other unambiguous cue. The locally Bayesian model predicts all these results. At the same time, it also predicts a variety of trial-order effects that are well known to appear in learning experiments, such as that backward blocking is weaker than its forward counterpart (induced by reversing the order of trials, yielding A ! X followed by A.B ! X). The globally Bayesian model predicts the main results but fails to predict any such order effects, and often learning is slower than with the local model. Why does the local model work so well? It draws on some strengths of the Bayesian approach generally, but also benefits from the modular perspective. Retrospective revaluation occurs because when the probability of one hypothesis is increased (or decreased), the probability of a mutually exclusive hypothesis must then, according to Bayes’ theorem, decrease (or increase). This is a general property of Bayesian belief updating. However, the ability to predict trial-order effects has a more specific explanation; it emerges because the internal modules organize themselves to allocate attention towards highly predictive events, and this attentional allocation then constrains the way in which learning proceeds. Such a process is intrinsically order sensitive because if you have learned not to attend to a particular cue, other cues will take up the explanatory load on later trials, where that cue could otherwise have borne some of the load. Prospects The model developed by Kruschke leaves many of the known features of retrospective revaluation unexplained. For instance, the role of associations between
Vol.10 No.11
the cues themselves (such as associations between A and B) is well known to be crucial in generating retrospective effects [2], but this is not predicted by the particular version Kruschke describes (it could be incorporated into a more complex version). For many researchers, the sheer computational and mathematical intensity of Bayesian models, together with the difficulty of applying them to any but the simplest situations, leaves them cold. Nevertheless, to have shown that a set of key results that have previously resisted a unified explanation might all emerge from a single set of principles, and that they might be ‘rational’ in a Bayesian sense, is a startling achievement. Future research in this area, relating and comparing this approach to the important work of others such as Dayan and Kakade [3] and Courville, Daw and Touretzky [4], promises much. A strong prior belief is that Kruschke’s framework will have a significant influence on how we think about simple learning processes. We will be able to revise this belief – in a Bayesian manner, of course – as more evidence accumulates. References 1 Kruschke, J.K. (2006) Locally Bayesian learning with applications to retrospective revaluation and highlighting. Psychol. Rev. 113, 677–699 2 Dickinson, A. and Burke, J. (1996) Within-compound associations mediate the retrospective revaluation of causality judgements. Q. J. Exp. Psychol. 49B, 60–80 3 Dayan, P. and Kakade, S. (2001) Explaining away in weight space. In Advances in Neural Information Processing Systems 13 (Leen, T.K. et al., eds), pp. 451–457, MIT Press 4 Courville, A.C. et al. (2005) Similarity and discrimination in classical conditioning: a latent variable account. In Advances in Neural Information Processing Systems 17 (Saul, L.K. et al., eds), pp. 313– 320, MIT Press 1364-6613/$ – see front matter ß 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.09.004
Book Review
Does schizophrenia have a cause? Mind, Brain and Schizophrenia by Peter Williamson. Oxford University Press, 2005. £38.99 (hbk) (296 pp.) ISBN 0-19-517637-5
Timothy J. Crow SANE POWIC, University Department of Psychiatry, Warneford Hospital, Oxford, OX3 7JX, UK
This is a bold book, but I will argue not bold enough. Williamson sets out to give an account of contemporary research and concepts of schizophrenia. The author is versatile and knowledgeable, crossing disciplinary boundaries nimbly and summarizing whole tracts of the recent literature. His particular strength is in neuroimaging and neuroanatomy. However, Williamson’s knowledge of neuropathology seems weaker, which imposes constraints on some of the theorizing. He writes, ‘The hippocampus is implicated by Corresponding author: Crow, T.J. ([email protected]) Available online 2 October 2006. www.sciencedirect.com
almost any line of investigation that has been attempted. . .There is ample evidence from structural and functional imaging as well (p. 127)’. However, in a study by my research group, hippocampal structure was found to be remarkably normal in the post-mortem schizophrenic brain in terms of areas of fields, cell densities and even cell size [1]. Similarly, the amygdala and thalamus, much loved in the radiological literature as foci of the disease process, seem surprisingly to be intact after death. Thus, Williamson writes, ‘The thalamus is essential to attention, sensory processing, and conscious experience, all functions that could be involved in schizophrenia (p. 131)’. This might well be true, but the problem is to constrain the search. Our only significant finding in the thalamus was that the asymmetry to the right of the
Update
TRENDS in Cognitive Sciences
pulvinar in the normal brain was reversed in patients with schizophrenia [2]. The primary change must be elsewhere. The cortex is the obvious structure of sufficient size and late development to account for the ventricular enlargement that is now the most reliable and uniform biological correlate. Williamson considers these and other problems in relation to existing theories – disconnection, aberrant lateralization, an ‘imbalanced’ brain and cognitive dysmetria, as well as in terms of neurodegeneration, excitotoxicity and transmitter deficit or excess as pathophysiological mechanisms. The account of the literature he provides is open-minded; with respect to my own preferred hypothesis, his judgement is that ‘The evidence for language laterality abnormalities in schizophrenic patients is not overwhelming (p. 97)’. However, Cullen et al. [3] have recently shown that pyramidal cells in area 9 of dorsolateral prefrontal cortex are more dense on the left than on the right in normal subjects and that this asymmetry is reversed in patients with schizophrenia, and that asymmetries of cell size and shape (larger and more triangular on the left in normal subjects) are lost. Asymmetry (the torque) is a powerful variable that we perceive only dimly. The general weakness I discern in this well-documented account is a failure to exclude theories and reach a definite conclusion, even though this might prove to be wrong. The marketing blurb says that the book ‘. . .argues that it is time to set aside the search for a single cause. . .and focus on the disorder’s final common pathway’ and this is what Williamson does. Thus, obstetric complications are still included, even though, in my view, they are ruled out by the cohort studies and case-control meta-analyses. There is little reference to viruses (my own ‘road to Damascus’ experience), but there is an account of pre-natal influenza (p. 36) without mention of the decisive findings from the National Child Development Study cohort that rule this out [4]. The contemporary crux is the chaos in psychosis genetics. Williamson points out ‘. . .that there have been
Vol.10 No.11
many failures to replicate these linkages, which suggests that the genes related. . .are associated with risk in only a subgroup of patients (p. 38)’. No! Unreplicated findings are probably just noise; they should be viewed with critical scepticism. Again, he states, ‘It is increasingly clear that no single genetic abnormality is likely to account for the large degree of the variance. . .as time goes on it is likely that a number of genes will be weakly linked to schizophrenia.’ Well, they might be, and the literature will probably grind to a halt under the weight of them. The alternative is referred to late in the last chapter under the heading ‘Why is schizophrenia a uniquely human disorder?’. To my mind this question gives us a way forward. First it suggests an interpretation of nuclear symptoms – that these are primary disorders of language, and second it suggests an entirely different approach to genetics – that the barrier to be overcome is the crucial chain of transitions from the chimpanzee and hominin precursor to Australopithecus to Homo erectus and modern Homo sapiens [5]. Towards the end, I thought Williamson was heading in the right direction, although still in serious danger of being held back by meaningless polygenes and other distractions and, above all, by the lack of a cogent hypothesis. References 1 Walker, M.A. et al. (2002) Estimated neuronal populations and volumes of the hippocampus and its subfields in schizophrenia. Am. J. Psychiatry 159, 821–828 2 Highley, J.R. et al. (2003) Low medial and lateral right pulvinar volumes in schizophrenia: a postmortem study. Am. J. Psychiatry 160, 1177– 1179 3 Cullen, T.J. et al. (2006) Anomalies of asymmetry of pyramidal cell density and structure in dorsolateral prefrontal cortex in schizophrenia. Brit. J. Psychiatry 188, 26–31 4 Crow, T.J. et al. (1991) Schizophrenia and influenza. Lancet 338, 116–117 5 Williams, N.A. et al. (2006) Accelerated evolution of Protocadhern11X/Y: a candidate gene-pair for cerebral asymmetry and language. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 141, 623–633
1364-6613/$ – see front matter ß 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.09.008
Free journals for developing countries In 2002, the WHO and six medical journal publishers launched the Health InterNetwork Access to Research Initiative, which enabled nearly 70 of the world’s poorest countries to gain free or reduced-cost access to biomedical literature through the internet. Currently more than 70 publishers are participating in the program, providing access to over 2000 journals. Gro Harlem Brundtland, former director-general for the WHO, said that this initiative was ‘‘perhaps the biggest step ever taken towards reducing the health information gap between rich and poor countries’’.
For more information, visit www.who.int/hinari www.sciencedirect.com
479
478
TRENDS in Cognitive Sciences
pairings may be equal overall, people tend to learn a stronger association for one of the unambiguous cues if it tends to appear later in the training corpus – that is, if it is highlighted with respect to the other unambiguous cue. The locally Bayesian model predicts all these results. At the same time, it also predicts a variety of trial-order effects that are well known to appear in learning experiments, such as that backward blocking is weaker than its forward counterpart (induced by reversing the order of trials, yielding A ! X followed by A.B ! X). The globally Bayesian model predicts the main results but fails to predict any such order effects, and often learning is slower than with the local model. Why does the local model work so well? It draws on some strengths of the Bayesian approach generally, but also benefits from the modular perspective. Retrospective revaluation occurs because when the probability of one hypothesis is increased (or decreased), the probability of a mutually exclusive hypothesis must then, according to Bayes’ theorem, decrease (or increase). This is a general property of Bayesian belief updating. However, the ability to predict trial-order effects has a more specific explanation; it emerges because the internal modules organize themselves to allocate attention towards highly predictive events, and this attentional allocation then constrains the way in which learning proceeds. Such a process is intrinsically order sensitive because if you have learned not to attend to a particular cue, other cues will take up the explanatory load on later trials, where that cue could otherwise have borne some of the load. Prospects The model developed by Kruschke leaves many of the known features of retrospective revaluation unexplained. For instance, the role of associations between
Vol.10 No.11
the cues themselves (such as associations between A and B) is well known to be crucial in generating retrospective effects [2], but this is not predicted by the particular version Kruschke describes (it could be incorporated into a more complex version). For many researchers, the sheer computational and mathematical intensity of Bayesian models, together with the difficulty of applying them to any but the simplest situations, leaves them cold. Nevertheless, to have shown that a set of key results that have previously resisted a unified explanation might all emerge from a single set of principles, and that they might be ‘rational’ in a Bayesian sense, is a startling achievement. Future research in this area, relating and comparing this approach to the important work of others such as Dayan and Kakade [3] and Courville, Daw and Touretzky [4], promises much. A strong prior belief is that Kruschke’s framework will have a significant influence on how we think about simple learning processes. We will be able to revise this belief – in a Bayesian manner, of course – as more evidence accumulates. References 1 Kruschke, J.K. (2006) Locally Bayesian learning with applications to retrospective revaluation and highlighting. Psychol. Rev. 113, 677–699 2 Dickinson, A. and Burke, J. (1996) Within-compound associations mediate the retrospective revaluation of causality judgements. Q. J. Exp. Psychol. 49B, 60–80 3 Dayan, P. and Kakade, S. (2001) Explaining away in weight space. In Advances in Neural Information Processing Systems 13 (Leen, T.K. et al., eds), pp. 451–457, MIT Press 4 Courville, A.C. et al. (2005) Similarity and discrimination in classical conditioning: a latent variable account. In Advances in Neural Information Processing Systems 17 (Saul, L.K. et al., eds), pp. 313– 320, MIT Press 1364-6613/$ – see front matter ß 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.09.004
Book Review
Does schizophrenia have a cause? Mind, Brain and Schizophrenia by Peter Williamson. Oxford University Press, 2005. £38.99 (hbk) (296 pp.) ISBN 0-19-517637-5
Timothy J. Crow SANE POWIC, University Department of Psychiatry, Warneford Hospital, Oxford, OX3 7JX, UK
This is a bold book, but I will argue not bold enough. Williamson sets out to give an account of contemporary research and concepts of schizophrenia. The author is versatile and knowledgeable, crossing disciplinary boundaries nimbly and summarizing whole tracts of the recent literature. His particular strength is in neuroimaging and neuroanatomy. However, Williamson’s knowledge of neuropathology seems weaker, which imposes constraints on some of the theorizing. He writes, ‘The hippocampus is implicated by Corresponding author: Crow, T.J. ([email protected]) Available online 2 October 2006. www.sciencedirect.com
almost any line of investigation that has been attempted. . .There is ample evidence from structural and functional imaging as well (p. 127)’. However, in a study by my research group, hippocampal structure was found to be remarkably normal in the post-mortem schizophrenic brain in terms of areas of fields, cell densities and even cell size [1]. Similarly, the amygdala and thalamus, much loved in the radiological literature as foci of the disease process, seem surprisingly to be intact after death. Thus, Williamson writes, ‘The thalamus is essential to attention, sensory processing, and conscious experience, all functions that could be involved in schizophrenia (p. 131)’. This might well be true, but the problem is to constrain the search. Our only significant finding in the thalamus was that the asymmetry to the right of the
Update
TRENDS in Cognitive Sciences
pulvinar in the normal brain was reversed in patients with schizophrenia [2]. The primary change must be elsewhere. The cortex is the obvious structure of sufficient size and late development to account for the ventricular enlargement that is now the most reliable and uniform biological correlate. Williamson considers these and other problems in relation to existing theories – disconnection, aberrant lateralization, an ‘imbalanced’ brain and cognitive dysmetria, as well as in terms of neurodegeneration, excitotoxicity and transmitter deficit or excess as pathophysiological mechanisms. The account of the literature he provides is open-minded; with respect to my own preferred hypothesis, his judgement is that ‘The evidence for language laterality abnormalities in schizophrenic patients is not overwhelming (p. 97)’. However, Cullen et al. [3] have recently shown that pyramidal cells in area 9 of dorsolateral prefrontal cortex are more dense on the left than on the right in normal subjects and that this asymmetry is reversed in patients with schizophrenia, and that asymmetries of cell size and shape (larger and more triangular on the left in normal subjects) are lost. Asymmetry (the torque) is a powerful variable that we perceive only dimly. The general weakness I discern in this well-documented account is a failure to exclude theories and reach a definite conclusion, even though this might prove to be wrong. The marketing blurb says that the book ‘. . .argues that it is time to set aside the search for a single cause. . .and focus on the disorder’s final common pathway’ and this is what Williamson does. Thus, obstetric complications are still included, even though, in my view, they are ruled out by the cohort studies and case-control meta-analyses. There is little reference to viruses (my own ‘road to Damascus’ experience), but there is an account of pre-natal influenza (p. 36) without mention of the decisive findings from the National Child Development Study cohort that rule this out [4]. The contemporary crux is the chaos in psychosis genetics. Williamson points out ‘. . .that there have been
Vol.10 No.11
many failures to replicate these linkages, which suggests that the genes related. . .are associated with risk in only a subgroup of patients (p. 38)’. No! Unreplicated findings are probably just noise; they should be viewed with critical scepticism. Again, he states, ‘It is increasingly clear that no single genetic abnormality is likely to account for the large degree of the variance. . .as time goes on it is likely that a number of genes will be weakly linked to schizophrenia.’ Well, they might be, and the literature will probably grind to a halt under the weight of them. The alternative is referred to late in the last chapter under the heading ‘Why is schizophrenia a uniquely human disorder?’. To my mind this question gives us a way forward. First it suggests an interpretation of nuclear symptoms – that these are primary disorders of language, and second it suggests an entirely different approach to genetics – that the barrier to be overcome is the crucial chain of transitions from the chimpanzee and hominin precursor to Australopithecus to Homo erectus and modern Homo sapiens [5]. Towards the end, I thought Williamson was heading in the right direction, although still in serious danger of being held back by meaningless polygenes and other distractions and, above all, by the lack of a cogent hypothesis. References 1 Walker, M.A. et al. (2002) Estimated neuronal populations and volumes of the hippocampus and its subfields in schizophrenia. Am. J. Psychiatry 159, 821–828 2 Highley, J.R. et al. (2003) Low medial and lateral right pulvinar volumes in schizophrenia: a postmortem study. Am. J. Psychiatry 160, 1177– 1179 3 Cullen, T.J. et al. (2006) Anomalies of asymmetry of pyramidal cell density and structure in dorsolateral prefrontal cortex in schizophrenia. Brit. J. Psychiatry 188, 26–31 4 Crow, T.J. et al. (1991) Schizophrenia and influenza. Lancet 338, 116–117 5 Williams, N.A. et al. (2006) Accelerated evolution of Protocadhern11X/Y: a candidate gene-pair for cerebral asymmetry and language. Am. J. Med. Genet. B. Neuropsychiatr. Genet. 141, 623–633
1364-6613/$ – see front matter ß 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.09.008
Free journals for developing countries In 2002, the WHO and six medical journal publishers launched the Health InterNetwork Access to Research Initiative, which enabled nearly 70 of the world’s poorest countries to gain free or reduced-cost access to biomedical literature through the internet. Currently more than 70 publishers are participating in the program, providing access to over 2000 journals. Gro Harlem Brundtland, former director-general for the WHO, said that this initiative was ‘‘perhaps the biggest step ever taken towards reducing the health information gap between rich and poor countries’’.
For more information, visit www.who.int/hinari www.sciencedirect.com
479