Authors INTRODUCTION R. S. J. Frackowiak PART I IMAGING NEUROSCIENCE—BRAIN SYSTEMS SECTION ONE SENSORY, MOTOR AND PLASTICITY R. S. J. Frackowiak 1. R. E. Passingham, N. Ramnani, and J. B. Rowe 2. D. Corfield 3. A. Kleinschmidt 4. T. Griffiths and A. L. Giraud 5. D. McGonigle 6. N. Ward and R. S. J. Frackowiak 7. T. Good and R. S. J. Frackowiak 8. A. L. Giraud SECTION TWO VISION AND VISUAL PERCEPTION S. Zeki 9. S. Zeki 10. S. Zeki 11. S. Zeki and K. Moutoussis 12. K. Moutoussis and S. Zeki 13. A. Bartels and S. Zeki 14. R. Perry xi Front Matter.qxd 10/10/03 9:08 AM Page xii xii AUTHORS SECTION THREE HIGHER COGNITIVE FUNCTIONS C. Frith 15. C. Frith, G. Rees, E. Macaluso, and S. Blakemore 16. C. Portas, P. Maquet, G. Rees, S. Blakemore, and C. Frith 17. J. Coull and C. Thiele 18. C. Frith, H. Gallagher, and E. Maguire SECTION FOUR EMOTION AND MEMORY R. Dolan 19. J. Morris and R. Dolan 20. H. Critchley and R. Dolan 21. P. Vuilleumier, J. L. Armony, and R. J. Dolan 22. R. Elliott 23. R. N. A. Henson 24. R. N. A. Henson 25. P. Fletcher and R. N. A. Henson SECTION FIVE LANGUAGE AND SEMANTICS C. Price 26. C. Price 27. U. Noppeney 28. C. Price 29. E. McCrory 30. A. Mechelli PART TWO IMAGING NEUROSCIENCE—THEORY AND ANALYSIS 31. K. Friston Front Matter.qxd 10/10/03 9:08 AM Page xiii xiii AUTHORS SECTION ONE COMPUTATIONAL NEUROANATOMY J. Ashburner 32. J. Ashburner and K. Friston 33. J. Ashburner and K. Friston 34. J. Ashburner and K. Friston 35. J. Ashburner and K. Friston 36. J. Ashburner and K. Friston SECTION TWO MODELLING W. Penny 37. S. Kiebel and A. Holmes 38. J. B. Poline, F. Kherif, and W. Penny 39. D. Glaser and K. Friston 40. R. Henson 41. D. Glaser, K. Friston, A. Mechelli, R. Turner, and C. Price 42. W. Penny and A. Holmes 43. W. Penny and K. Friston SECTION THREE INFERENCE W. Penny 44. M. Brett, W. Penny, and S. Kiebel 45. K. Worsley 46. T. Nichols and A. Holmes 47. K. Friston and W. Penny SECTION FOUR FUNCTIONAL INTEGRATION K. Friston 48. K. Friston 49. K. Friston and C. Büchel 50. L. Harrison and K. Friston Front Matter.qxd 10/10/03 9:08 AM Page xiv xiv AUTHORS 51. K. Friston 52. K. Friston 53. K. Friston POSTSCRIPT A. Roepstorff xiv INTRODUCTION It is almost a decade since the first edition of Human Brain Functionwas conceived and planned. It was a unique book in that it tried to set out the thoughts and achievements of a school, or more accurately a laboratory working in the field of functional and structural human neuroanatomy. Historically, it marked the border between the application of tracer-based methods to brain imaging, exemplified by the PET technique of local blood flow mapping, and the blood oxygen level dependent (BOLD) technique based on magnetic resonance imaging (fMRI). The concept of a ‘group’ book was comparatively unusual in the biosciences where the tradition of the solitary scientist (often assisted by ‘juniors’) was still the dominant ethos. The evolution of human brain mapping depended on much diverse expertise from mathematics and statistics, through physics and biology to neurology, neuropsychiatry, and neuropsychology. The realisa- tion of this dependence on the expertise of many individuals from many disciplines motivated the principal investigators to create a laboratory environment that was collegiate, interactive, collaborative, and also amicable. That laboratory emerged from the Medical Research Council’s Cyclotron Unit and was incarnated in the then nascent Wellcome Trust funded Functional Imaging Laboratory, known as The FIL. The first edition marked the emigration of that group from the MRC CU to the FIL and represented the beginning of a new enterprise focussed on understanding the functional and structural architecture of the human brain and methodological developments that supported the achievement of that mission. The beginnings of statistical parametric mapping (SPM) were already described, but the advances made possible by fMRI did not make it into the first edition. This second edition, like the previous one, is written exclusively by members of the FIL, past and present and long-term collaborators. The chapters are knit together like a book rather than a series of reviews. In that sense the book remains a ‘personal’ reflection on the state of our knowledge of how human thinking, feeling, and action are instantiated in the brain. However, the methodological advances of the last 6 years that include event-related fMRI, massive improve- ments in image acquisition and pre-processing and an escape from the relative constraints of classical inference are huge. These combined with a courageous, sometimes foolhardy wish to attack interesting problems that include consciousness, free will, and feelings make it possible that the book is becoming perhaps too big for its boots. We think this may be the last time we can put together theory and application in a single volume, even though they would be focussed by a common mission. The book is now organised in sections, edited by each of The FIL’s Principal Investigators who have promoted their component of the common programme. The theory and analysis section edited by Karl Friston, abetted by John Ashburner and Will Penny, is entirely up to date and gives readers an approachable and yet professional overview of all that is possible with modern imaging of brain structure and function. The adoption of a common analytic approach internationally, though sometimes in different guises, means that this section contains contrib- utions from ‘honorary’FIL members who have worked with Karl in the context of developing and supporting SPM. Chris Frith tackles the roles of the frontal lobes including mechanisms for xv Front Matter.qxd 10/10/03 9:08 AM Page xvi xvi INTRODUCTION attention and control of action and the relevance of these mechanisms to understanding the neural correlates of consciousness. Cathy Price continues her revision of the psychological theories of the organisation of human language by marshalling more and more experimental evidence that challenges older theories based purely on behavioural observation. Ray Dolan explores the difficult areas of feeling, emotion, and their interaction with memory and cognition and provides many new insights that complement the explosion in knowledge in this area that has occurred in animal and basic biology. Semir Zeki explores the visual world that has become his unique domain and extends our understanding of why it constitutes such a paradigmatic sensation in the history of neuroscience. He approaches the problems with his usual panache and delight in intellectual exploration through controversy. I deal with action and sensation with an ever-vigilant eye on implications for recovery of function and mechanisms underlying this phenomenon. I am abetted in this work by Dick Passingham whose contributions have been massive. What of the future, for which this volume will act as our springboard? The integration of temporally resolved methods such as EEG and MEG, spontaneous or evoked, is a big challenge. It is a challenge that has received a major boost from experiments by others that have elucidated the physiological correlates of the BOLD signal and the relationship between changes in each. Mapping one type of signal onto the anatomy of the other is trivial compared to relating the biological basis of each into a common framework. A common framework should result in inferences that lead to new predictions and experiments which focus on distributions of brain activity and their correlation in time. The second area where we foresee great advances is in the application of the knowledge obtained in over a decade of experimentation in normal humans to the diseased condition. An understanding of the principles of normal functional organisation and an elaboration of experimental approaches and principled, reproducible methods of data now make this task feasible. The questions to be asked are unlikely to be primarily diagnostic but will generate ideas and facts about disease mechanisms as well as information about how these can be altered by therapy. The possibility of assessing drug effects on small well-defined groups of patients instead of by costly large population studies that last years will have to be explored. The idea that the genetic understanding of brain disease will be aided by a description of correlated changes in the working of brain systems and associated with the identification of abnormal proteins or their absence might be dismissed as a flight of fantasy except for the fact that pre- liminary data suggest otherwise. In brain diseases of ageing or neurodegenerations it is entirely possible that proteomic data will most readily be interpreted by a mixture of physiological data obtained via human imaging neuroscience and behavioural correlation. We hope that the reader will forgive the hubris that underlies the decision to produce this volume. This scientific field has excited us so much that we see no reason to hide our enthusiasm to communicate it. However, it should not be thought that we do not recognise the enormous contribution of many others worldwide to this science. Without that, much that is recorded in this book would not have been possible. The field is now so huge that one volume cannot hope to cover it completely. There is much missing in these pages that will be found elsewhere; some of it in the first edition. There is much that in the time required for printing and publication will already be out of date. However, when a field is expanding fast, a book should try to convey the process of acquisition of knowledge as much as content. In the internet-based information age, it is perhaps that function which a book can fulfill in a way that papers themselves, read individually cannot. So, we hope the reader will share our excitement when reading this volume; that the reader will be stimulated at times and at others perplexed. We will even consider it a success if the book irritates the reader or causes reflection that leads to new experiments that refute (or support) the claims and speculations based on our work recorded within. It remains for me to thank, on behalf of all my colleagues, the Wellcome Trust for its munificence. Without the decision to fund the FIL in 1994 and to renew funding in 1999 very little of the work contained in this volume would have been possible. Richard S. J. Frackowiak London September 2003 C H A P T E R 1 The Motor System MOTOR AREAS Just as there are a host of visual areas—defined loosely as areas with visual inputs (Felleman and van Essen, 1990)—so are there many motor areas — defined loosely as regions with projections to the premotor and motor areas. Thus. the term motor system will be used in this chapter to include parietal and prefrontal cortex, as well as the premotor and motor areas. All are involved in the selection, guidance or execution of action. One advantage of studying the motor system is that, as for the visual system (see Section 2, Vision and Visual Perception), the anatomy and physiology are well worked out in non-human primates. We can identify the different motor and premotor areas in the macaque monkey using cytoarchitectonics (Brodmann, 1909; von Bonin and Bailey, 1947) and receptor architectonics (Zilles et al., 1996). The anatomical connections between the different areas are well described, and an exhaustive database of these, CoCoMac, is available (Stephan et al., 2001; http://www. cocomac.org). There are also many studies in which single units have been recorded, and comparisons are available between recordings taken in different areas (e.g., Crutcher and Alexander, 1990; Muskiake et al., 1991; Matsuzaka and Tanji, 1996; Nakamura et al., 1998; Shima and Tanji, 2000). Thus, the motor system provides an excellent case study of how the findings of imaging experiments can be grounded in anatomy and physiology. ANATOMY The anatomical basis of functional localisation is the unique set of inputs and outputs for each cytoarchitectonic area. Passingham et al. (2002) have suggested the term connectional fingerprint for this set. Young (1993) has shown that each area has a unique set of inputs and outputs. The most stringent test is to compare the set of connections for regions within a single functional area. Passingham et al.(2002) have formally done this for the subregions of prefrontal cortex. The connections are plotted as fingerprints. This term was used by Zilles and colleagues (Geyer et al., 1998) to describe the particular pattern of receptor–architecture for each cortical region as demonstrated by the degree of binding for the different receptor types. They plotted the binding in a radial plot. Figure 1.1shows connectional fingerprints in the form of radial plots for the inputs and outputs of two prefrontal areas, 9 and 14. The data are taken from CoCoMac (Stephan et al., 2001). Figure 1.2uses multidimensional scaling to plot the data for all prefrontal regions in two dimensions. If any pair of areas shared the same set of connections, they would also share the identical location, but it can be seen from Fig. 1.2 that this is not true for any of these areas. We have very little direct information, other than from silver studies (Di Virgilio and Clarke, 1997), about the connections of the human brain (Crick and Jones, 1993). It is not yet clear to what extent diffusion-weighted imaging (Conturo et al., 1999; Poupon et al., 2000; Parker et al., Human Brain Function 5 Copyright 2004, Elsevier (USA). Second Edition All rights reserved. CH.01.qxd 10/10/03 9:27 AM Page 6 6 1. THE MOTOR SYSTEM W9 afferents W14 afferents W10 3 W10 W9 W11 3 W9 W11 2 2 W8B W12 W8B W12 1 1 W8A 0 W13 W8A 0 W13 W46 W14 W46 W14 W45 W24 W45 W24 W25 W25 W9 efferents W14 efferents W10 3 W10 W9 W11 3 W9 W11 2 W8B W12 2 W8B W12 1 1 W8A 0 W13 W8A 0 W13 W46 W14 W46 W14 W45 W24 W45 W24 W25 W25 FIGURE 1.1 Diagram of anatomical fingerprints for two prefrontal areas, Walker’s areas 9 and 14. The upper row shows the afferent and the lower row the efferent connections of the two areas, with 12 other prefrontal areas identified by their cytoarchitectonic numbers as designated by Walker (1940). The strength of any connection (rated as weak = 1, medium = 2, strong = 3) is shown by the radial distance. (From Passingham et al., (2002). Nature Publishing Group. With permission.) 2002) will be able to discriminate among the fine details of anatomical connections as shown by transport methods. Though it has been used to chart the course of the pyramidal tract in the macaque and human brain (Parker et al., 2002), the technique has yet to be tried for other connections in the motor system. For the moment, the information on connections comes from studies on non-human primates. The only way to access this information for human imaging studies is the mapping of activations to specific cytoarchitectonic areas. The connections between areas in the macaque brain are identified by infusing tracers into discrete cytoarchitectonic areas (Young et al., 1993; Stephan et al., 2001), and homologies are assumed between areas in the macaque and human brain that have the same cytoarchitectonic characteristics. Since the last edition of the book, advances have been made in producing probability maps for some of the areas within the motor system. This has been done for the primary somatosensory cortex (Geyer et al., 2000), area 6 (Geyer etal.,2001), area 44 (Tomaiulo et al., 1999), and the lateral prefrontal cortex (Rajkowska CH.01.qxd 10/10/03 9:27 AM Page 7 7 PHYSIOLOGY 2 STRESS=0.10 1 W45 2 - W11 W12 n o W46 i s W8A W10 W14 n e0 m W13 W25 i D W8B W9 -1 W24 -2 -2 -1 0 1 2 Dimension-1 FIGURE 1.2 Multidimensional scaling (MDS) showing two-dimensional space in which prefrontal areas are plotted according to their anatomical connections. MDS gives a high-dimensional metric representation in which distances between elements optimally reflect the overall similarity between their properties—in this case, the connections between the areas. (From Passingham et al., (2002). Nature Publishing Group. With permission.) and Goldman-Rakic, 1995). Probability maps are also available for sulci within the motor system cortex (Paus et al.,1996; Le Goualher et al., 1999; Chiavaras and Petrides, 2000). The variation in cytoarchitectural boundaries has also been demonstrated for the motor cortex (Roland and Zilles, 1994) and Broca’s area (Amunts et al., 1999). PHYSIOLOGY Functional specialisation can be studied by comparing the proportions of cells that fire in association with a particular task or component of a task. For example, Mushiake et al. (1991) trained monkeys on two sequence tasks: in one task, visual cues specified the sequence, and in the other the animals performed from memory and without visual cues. Figure 1.3 shows that cells in motor cortex fired equally on both tasks (category 4). Many cells in the supplementary motor area (SMA) and premotor cortex also fired equally on both tasks (category 4). However, in the SMA, there was an overall tendency for cells to fire only (category 7) or preferentially (categories 5 and 6) in association with movements performed from memory. By contrast, in the premotor cortex there was an overall tendency for cells to fire only (category 1) or preferentially (categories 2 and 3) in association with movements specified by external cues. Thus, these two areas differ in the activity of the populations of cells as a whole. The difference can be explained in part by the fact that there are projections to the premotor cortex from visually receiving areas of the parietal cortex such as the medial intraparietal cortex (MIP) and anterior intraparietal cortex (AIP) (Rizzolatti et al., 1998). CH.01.qxd 10/10/03 9:27 AM Page 8 8 1. THE MOTOR SYSTEM FIGURE 1.3 Distribution of cells in lateral premotor cortex (PMv), SMA, and motor cortex (MI) classified according to the degree to which they were active in association with the visually guided sequence (VS) or the sequence performed from memory (MS). 1 = exclusively related to VS, 2 = predominantly related to VS, 3 = more related to VS than MS, 4 = equally related to VS and MS, 5 = more related to MS, 6 = predominantly related to MS, and 7 = exclusively related to MS. The ordinate shows the percentage of cells and the numbers above the histograms show the number of cells. (Data from Mushiake et al., (1991). American Physiology Society. With permission.) It is the proportional differences in cell activity within the population that allow imaging studies to reveal differences between areas. For example, more cells in the SMA than in the motor cortex fire in advance of movement (Okano and Tanji, 1987; Tanji and Shima, 1994), and it has been shown using functional magnetic resonance imaging (fMRI) that there is more activity in the SMA than in motor cortex during preparation for movement (Richter et al., 1997). More cells in the pre-SMA than in the SMA increase their firing rate during the learning of new motor sequences (Nakamura et al., 1998) or change between movements (Matsuzaka and Tanji, 1996), and imaging studies have shown that there is more activity in the pre-SMA than in the SMA when subjects learn new sequences (Hikosaka et al.,1996) or vary the time at which they respond (Deiber et al., 1999). Finally, more cells in the premotor cortex than in the SMA fire when movements are externally specified, whereas more cells in the SMA fire when sequences are performed from memory (Muskiake et al., 1991; Halsband et al., 1994). Correspondingly, imaging studies have shown more activity in the premotor cortex than in the SMA early in sequence learning when the subjects are dependent on feedback cues, but more activity in the SMA than in premotor cortex when the task is automated and performed from memory (Jenkins et al., 1994; Toni et al., 1998). The high temporal resolution of single unit recording means that one can also study cell firing at different times during a task. For example, Weinrich et al.(1984) have recorded single units in the dorsal premotor cortex while monkeys perform a delayed visuomotor task. The authors report that 47% of the cells fired at the time of presentation of the cue, 34% during the delay period, and 65% at the time of response. Event-related fMRI have also made it possible to associate activations with particular event withina trial. Thus, Toni etal., (Passingham and Toni, 2001) were able to demonstrate that in the human brain there are activations in premotor cortex at the time of the instruction cue and at the time of the response, and that there is set activity during the delay (Toni et al.,1999, submitted) (Fig. 1.4). Functional MRI has also been used to distinguish prefrontal activity related to the maintenance of information in memory from the activity at the time of retrieval (D’Esposito et al., 1999; Rowe et al., 2000; Rowe and Passingham, 2001). Again, the results can be related to single-unit experiments on monkeys. For example, several studies have reported activity in prefrontal area 8 while subjects remember spatial locations (Belger et al., 1998; Courtney et al., 1998; Rowe et al., 2000; Rowe and Passingham, in press). Correspondingly, activity has been recorded in area 8A during the delay period in monkeys performing a spatial memory task (Chafee and Goldman-Rakic, 1999; Sawaguchi et al., 1999).