The parallel visual motion inputs into areas VI and V5 of human cerebral cortex

Transcription

1 Brain (1995), 118, The parallel visual motion inputs into areas VI and V5 of human cerebral cortex D. H. ffytche 1, C. N. Guy 2 and S. Zeki 1 department of Anatomy, University College London and 2 Blackett Laboratory, Physics Department, Imperial College, London, UK Correspondence to: Dr D. H. ffytche and Professor S. Zeki, Department of Anatomy and Developmental Biology, University College London, Gower Street, London WC1E 6BT, UK Summary Published clinical evidence has led us to hypothesize that there are parallel pathways which lead to the striate (VI) and prestriate cortex in the human brain. We have used the technique of visually evoked EEC coupled to magnetoencephalography (MEG) to test our hypothesis, by detecting the timing of arrival of signals into these visual areas, using published PET evidence to guide us in the location of the evoked response sources. We found that, if the moving stimulus has a speed of 22 s~', signals arrive in V5 before VI; with speeds of <6 s ', signals arrive in VI first. We conclude that, in addition to the classical picture of a sequential input to prestriate cortex through VI, there is also a fast parallel input which by-passes VI. The parallelism manifests itself only as a function of the characteristics of the visual stimulus, a phenomenon we describe as dynamic parallelism. The results obtained help us explain the residual motion vision of patients with lesions in VI or V5. Keywords: EEG; magnetoencephalogram; motion evoked response; VI; V5; parallel input; dynamic parallelism Abbreviations: EMEG = electromagnetoencephalography; MEG = magnetoencephalogram; TMS = transcranial magnetic stimulation Introduction Area V5 of the primate visual cortex has given us profound insights into the organization of the brain, and chief amongst these is the functional specialization that is a feature of the different visual areas within it (Zeki, 1974a, 1978). The specialization of V5 itself, in both monkey and man, is for visual motion, one of the most primordial visual signals. Because of the importance of motion vision for quick detection and early evasion, both of which might involve conscious decisions, one would naturally assume that signals reach V5 rapidly, perhaps more rapidly than other visual areas which are concerned with other attributes of the visual scene, such as form or colour. The input to V5 is dominated by the magnocellular or M system which is well suited to register motion and which also provides an important input to other prestriate areas such as V3 and the thick stripes of V2 (for a review, see Zeki and Shipp, 1988). One might, therefore, expect that a motion stimulus would activate these areas, in addition to V5, which is precisely what human imaging studies have shown (Watson et al, 1993). Here the electrophysiological literature presents us with a puzzle. Oxford University Press 1995 Studies of evoked visual activity in human cerebral cortex that have employed motion have concluded that the prestriate cortex, including area V5, is not activated by motion (Maier et al., 1987; van Dijk et al., 1987); other studies have concluded that any activity in the prestriate cortex follows that in the striate cortex (VI) in a sequential manner (Drasdo et al, 1993; Probst et al., 1993), with peak times in prestriate cortex variously given at the surprisingly long latencies of between 163 and 220 ms after onset of visual motion. But evoked potentials recorded from the scalp give a somewhat imperfect guide both to the timing of arrival of signals in the cortex and to the origin of the evoked activity (Regan, 1989; Celesia et al, 1993). A far better measure of the arrival of signals is obtained by intracortical electrodes and direct extracellular recordings from human brains have shown that the earliest signals reach the striate cortex with latencies of -30 ms (Wilson el al, 1983; Ducati et al, 1988), while recordings from the monkey striate and prestriate cortex have detected neuronal activity at -30 ms for flash, pattern reversal and moving stimuli (Kraut et al, 1985; Petersen et al,

2 1376 D. H. Jfytche et al 1988; Raiguel et al, 1989; Schroeder et al, 1991; Kawano et al, 1994). Here enters the second puzzle in the picture provided by evoked potential studies. It has been supposed that if signals reach the prestriate cortex they do so through VI, never directly (Kaufman and Williamson, 1990; Drasdo et al, 1993; Probst et al, 1993). Such conclusions sit oddly with the evidence derived from anatomical studies in the monkey which show that not only is there an input to V5 from VI (Cragg, 1969; Zeki, 1969, 1971) but that there is also a direct input to the prestriate cortex which by-passes VI, either from the lateral geniculate nucleus (Fries, 1981; Yukie and Iwai, 1981) or from the pulvinar (Standage and Benevento, 1983). What the function of these direct pathways may be has not been clarified, but our studies with patient G.Y., blinded by a lesion in VI, show that he can discriminate consciously the direction of motion of crude visual stimuli under conditions in which signals reach the prestriate cortex, including areas V5 and V3, directly, without passing through VI (Barbur et al., 1993). Such results suggest that the direct input to the prestriate cortex, even though less prominent than the one through VI, may nevertheless be a significant input, capable of mediating an independent, conscious visual perception of motion without the active participation of VI. Thus the two connected problems of the timing of arrival of visual signals linked to motion in the visual cortex and of the sequence in which the striate and prestriate cortex are activated seemed to be worth investigating. Thirty years of evoked potential literature has taught us that such a study is fraught with difficulties, for there is no obvious and straightforward relationship between the evoked activity registered from the scalp and the underlying firing of cells in the cortex (Regan, 1989; Celesia et al, 1993). In spite of these difficulties, the results of clinical studies suggested to us that we could use this apparently imperfect technique to advantage to solve our problem. The clinical evidence relates to the perception of motion in the absence of VI or of V5. We found a very interesting difference in the effects of VI and V5 lesions on visual motion perception. In both types of lesion some degree of crude visual motion perception is possible and, in both, the affected patient is consciously aware of having seen the visual stimulus. But there the similarity ends. Patients with lesions in VI can only detect high contrast stimuli that are rapidly displaced, e.g. at >15 s" 1, and our studies with PET have shown that this residual vision is mediated by prestriate cortex (Barbur et al., 1993). By contrast, the studies that we and others have done on patient L.M., with a lesion in V5, have shown that such residual visual motion that she has is limited to very slow speeds, <6 s~', presumably mediated through the intact areas of the cortex, including areas VI and a functionally impaired V3 (Hess et al., 1989; Shipp et al, 1994). We hoped that these clinical observations contained within them the answer to our dilemma, for if fast motion were able to activate prestriate cortex without VI and slow motion VI without prestriate cortex, then this may be the key to finding parallel activation. Our experiment revolves therefore around the question of whether signals from fast moving stimuli have direct access to prestriate cortex. Our working hypothesis was very simple: we reasoned that signals from slow moving stimuli reach VI first and are then relayed to the prestriate cortex in the conventional sequential manner, consistent with the previously published evoked response literature; by contrast, signals from fast moving stimuli are relayed directly to V5 and possibly V3, without passing first through VI, a result that would be inexplicable by an exclusively sequential model of the visual cortex. Put more simply, we reasoned that subcortical nuclei send parallel inputs into striate and prestriate cortex, but that whether the two inputs become active simultaneously depends on the speed of the stimulus. If true, this would imply that the direct input to the prestriate cortex depends upon the stimulus and is not an immutable feature; there is, in other words, a dynamic parallelism. At any rate, it seemed worth testing this hypothesis. In the event, it provided us with surprising results, not hinted at by the evoked potential literature. A brief, preliminary report of these results has been published elsewhere (ffytche et al., 1994). Methods General background To undertake such a study, we needed a technique that had enough temporal resolution to show the arrival of fast signals in the visual cortex, and enough spatial resolution to be able to distinguish between striate and prestriate activation. The only technique with adequate temporal resolution is the visual evoked response, even though, with the exception of those who have used intracortical electrodes (Ducati et al, 1988), specific electrical filters (Whittaker and Siegfried, 1983) or large numbers of averages (Yoneda et al., 1995) the method itself has not hinted that signals might arrive in the visual cortex at delays of <35 ms. The principle behind this technique is that a visual stimulus sets in train a series of neuronal responses that produce electrical currents within the brain. These evoked currents are a result of voltage gradients set up within the cortex along the long axis of pyramidal cells, by post-synaptic depolarization and hyperpolarization (see Lopes da Silva and Van Rotterdam, 1987). The currents are much smaller than the spontaneous background cerebral activity and are only detectable after signal averaging, but it is these same voltage gradients that are recorded as the visual evoked response of the EEC The magnetic fields produced by the currents are detectable as the visual evoked response of the MEG. The electrical field detected by the EEG and the magnetic field detected by the MEG are thus intimately connected, both being manifestations of the same cortical current. No matter where the evoked current is located within the brain, and there may be several currents each arising from a different cortical area, it will always have an electrical gradient and a magnetic field associated with it. What is less

3 Parallel inputs to VI and V certain is whether these fields can always be detected outside the head. There are three issues to consider here: (i) the distance of the neuronal source current from the magnetic and electrical sensors; (ii) the cancellation of equal and opposite currents; (iii) the orientation of the neuronal source current with respect to the surface of the head. With respect to issue (i), direct comparisons of depth and surface electrodes (Cooper et al., 1965; Alarcon et al., 1994) have shown that physiological activity deep inside the head cannot be directly detected by electrodes or magnetic sensors outside the head; rather the EEG and the MEG are dominated by activity in the cortex immediately below it. Since our areas of interest, VI and the prestriate cortex (more specifically areas V5 and V3), are known to be on the surface of the brain, at least in part, we felt confident that the external detectors would be close enough to the evoked currents in these areas to detect their electrical and magnetic fields. VI is situated on the occipital pole, extending onto the medial surface of each hemisphere, while V5 and upper V3 are on the lateral surface of each occipital lobe, the latter extending onto the medial surface (see Watson et al., 1993; Shipp et al., 1995). Issue (ii) is of the cancellation of currents. If several currents are induced around a spherical dendritic tree, e.g. that of a stellate cell, then the currents on one side of the neuron will cancel the currents on the other side producing a 'closed field' and, as a result, no magnetic or electrical field will extend outside the head (Lorente de No, 1947). A 'closed field' can also occur on a larger scale following the activation of an approximately spherical area of cortex, but the overall configuration of areas VI, V5 and V3, none of which is spherical, suggests that, while some cancellation of current would take place, these areas produce an electrical or magnetic field detectable outside the head. Issue (iii) is of the orientation of the current. Radial currents do not produce an external magnetic field (see Romani, 1989). This is because the magnetic field produced by a current is orthogonal to the axis of the electrical gradient (Faraday's right hand rule) and, more importantly, because a radial current within a sphere produces secondary currents that cancel the component of the magnetic field which extends outside the head. Figure 1 demonstrates how a current flowing from the depth to the surface of the cortex on the crest of a gyrus will produce a magnetic field that does not extend outside the head and is thus not detectable, however close it is to an external MEG sensor. By contrast, the area of electrical positivity is directly under the skull, very close to the EEG electrode, and is thus easily recorded. Conversely, currents in the wall of a gyrus, again flowing from the depth to the surface of the cortex, will produce a magnetic field that, in this case, does extend outside the head, and an area of electrical positivity that will make less of a contribution to the EEG. As the EEG and MEG methods are complementary we decided to use simultaneous recording of the evoked electrical and magnetic fields to test our hypothesis [electromagnetoencephalography (EMEG)]. While we were satisfied that we could fulfil our prerequisite temporal resolution we were left with the problem of spatial resolution. The MEG has a better spatial resolution than the EEG because it is less affected by the blurring effects of the skull and cerebrospinal fluid (Romani, 1989) but we remained concerned that, in order to localize a source such as V5 or V3, we would need to rely on a further mathematical interpretation of the recordings, a procedure riddled with uncertainties. This results from the fact that the external EEG and MEG fields cannot give a unique answer to the location of a source within the head. A straightforward analogy would be that knowing a sealed opaque jar contains 10 does not enable one to say which coin combinations make up the sum. This problem is known as the inverse problem and a number of mathematical strategies have been developed to provide a solution, none of which can give a unique source location and each of which is open to criticism (Balish and Muratore, 1990). In order to avoid the ambiguity inherent in an inverse solution we therefore preferred to localize our current sources with a different technique. We turned to the results of our previous studies, in which we had used the technique of PET to detect changes in regional cerebral blood flow when subjects were presented with a moving as opposed to a stationary checkerboard (Watson et al., 1993). Changes in regional cerebral blood flow were found in VI and the prestriate cortex, including areas V2, V3 and V5. The PET technique, however, lacks the temporal resolution to show the sequence of activation between these areas. It seemed reasonable to combine the temporal merits of the EMEG with the spatial merits of PET. Using the locations derived from our published PET studies (Watson et at., 1993) and the laws that govern the electrical and magnetic fields produced by the current sources, we used a computer model to predict what the activity on the scalp would look like when these different areas in the visual cortex are active. We then looked at the real evoked EMEG data to see if and when these predicted fields occurred, and hence the sequence of activation between cortical areas. In summary, we compensated for the poor spatial resolution of the evoked response by localizing our sources with published PET evidence and we compensated for the poor temporal resolution of PET studies by using the EMEG approach. Modelling The details of the modelling procedure are described in the Appendix, but we will present the results here and show that the electromagnetic field distributions allowed us to distinguish striate and prestriate activity. Figure 2 shows the EMEG field produced when current sources are placed in both hemispheres on the modelled cortical surface in area VI, along the length of the calcarine fissure, the lingual gyrus and the cuneus (Stensaas et al., 1974). On the right is the electrical field (high potentials light, low potential dark) with a maximum on the mid-

4 1378 D. H. jfytche et al. line. The EEG amplitude declines steeply laterally so that at 10 cm from the midline the voltage is less than one-tenth of its maximum value. The associated magnetic field is shown on the left and demonstrates a quadrapolar pattern with a pair of magnetic fields, one going into the head and one coming out, on each side of the midline. Hence, with appropriately placed MEG sensors, one would expect to see a phase reversal between the midline and lateral occiput or above and below the equator. An important conclusion was that the EMEG field is dominated by the region of striate cortex that extends onto the occipital pole, the cortical representation of the fovea. The consequence is that, as long as the occpital pole is activated, the distribution of EMEG remains the same, even if a moving stimulus is better detected in the periphery of the visual field and hence the anterior extent of the calcarine fissure. The exact position of human area V5 with respect to the gyral anatomy, whether it lies predominantly on the crest or in the wall of the gyrus, is important as it influences the distribution of the predicted magnetic field. Figure 3A shows a section of human inferior temporal gyrus, stained to show the patterns of myelination. In this example, the myelinated area which corresponds to Flechsig's Feld 16 and to area V5 as well (Watson et al., 1993) appears on the crest of the gyrus and extends onto both walls (S. Zeki, unpublished data). If the gyrus is orientated radially in the skull, then it is predominantly activity in the walls that will be detected by MEG sensors. The result of placing current sources in V5 bilaterally, after incorporating the anatomical and PET evidence into the model, is shown in Fig. 3B. The electrical field is the opposite of that produced by VI, with an electrical maximum lateral to the midline and a voltage gradient of 5:1 from lateral to midline. The maximum lateral voltage is in a region that is not covered by the conventional or the Queen Square electrode system [see Regan (1989) for a description of the electrode positions], which necessitated the development of our own electrode system {see below). In contrast, the magnetic field, while shifted laterally, is qualitatively the same as the result of placing sources in VI. Finally, Fig. 4 shows the distribution of magnetic and electrical fields when sources are placed in VI and V5 simultaneously. The important difference between this pattern and that produced by sources in VI alone is that, for simultaneous striate and prestriate sources, there is more activity lateral to the midline, resulting in a smaller voltage gradient (3:1 as opposed to 10:1). This pattern is unaltered when we include prestriate sources in a region of the cortex corresponding to area V3 (Watson et al., 1993, Shipp et al., 1995). In our model we assume that the density of evoked current is the same in different regions of the cortex, so that the amplitude of the surface EEG or MEG produced by a given area results simply from a combination of its size, its orientation and its distance from the scalp. Of course the polarity of the electrical field and the direction of the magnetic Fig. 1 A sagittal MRI slice through the brain with a small region is magnified in A and B. A current source marked in white on the crest of the gyms is shown in A. The + and - mark the electrical potential and the black rings the magnetic field. For a source on the crest of the gyrus, the magnetic field does not extend outside the head. A current source in the wall of the same gyrus is shown in B. In this case the magnetic field marked with the arrow does extend outside the head.

5 Parallel inputs to VI and V5 EEG MEG MEDIAL Fig. 2 The result of modelled sources in VI in both hemispheres. The spherical MEG and EEG contour maps viewed from the back of the head are presented flattened in the top row. The EEG map is coloured in shades of grey to show the gradient of electrical potential on the surface of the head the lighter the shade, the higher the electrical potential. Magnetic field going into the head and coming out of the head are represented by different shading in the MEG map. The position of the nine electrodes to be used in the evoked response experiments are marked as black or white circles within the zone of EEG map outlined by the square. Note that, because of the flattened spherical projection, the electrode grid appears distorted. A posterior and medial view of the left hemisphere is shown below to indicate the geometry of the extended sources. Area VI is shown in white passing along the calcarine fissure. Area V2 is represented by the black strip surrounding VI and area V3, divided into upper (V3u) and lower (V31) regions, is shown in white. Area V5 is marked in black on the lateral surface of the occipital lobe. field that the modelling program calculates is arbitrary, depending on whether we model the current source flowing from the surface to the depth of the cortex or vice versa. What does not change, even by inverting the currents, is the surface map or topography of the fields. It should also be noted that, since the model itself describes the EMEG topography at an arbitrary point in time, it does not in itself give any temporal information. While the modelling program can give a mathematically correct estimate of the distribution of electrical and magnetic fields, it is in a sense unnecessary as, on a very simple level, it is obvious that activity on the lateral surface of the brain will produce an EEG field that extends further from the midline than activity originating in the occipital pole alone. Considering the marked differences in EEG field distributions on the scalp we felt confident that we should be able to differentiate striate from prestriate activity and thus be able to determine the sequence of activation of these areas. If a moving stimulus of appropriate velocity activates prestriate as well as striate cortex, then this should be reflected in the surface distribution of electrical activity recorded over the occiput. In contrast, the predicted magnetic activity was similar regardless of the cortical areas modelled; we therefore decided to concentrate on the pattern of evoked electrical activity and to use the simultaneous MEG recordings to aid our interpretation of the results. Stimuli Because we had used the published PET results to localize our current sources and to model the predicted EMEG fields, it was important to match the characteristics of the checkerboard used in our PET study as closely as possible. We therefore used the identical checkerboard in our evoked response experiments and kept the luminance, contrast and the position in the visual field the same. However, unlike the PET experiments in which each subject viewed the moving checkerboard presented for 90 s six times, many more presentations are required to obtain an evoked response. We had to shorten the presentation time, but we were also aware POSTERIOR 1379

6 1380 D. H. ffytche et al. MEG EEG LATERAL Fig. 3 (A) A section of human V5 stained to show the heavy myelination within the cortex. The darker staining in layer 4 of the cortex extends from the anterior wall, over the crest and on to the posterior wall. (B) The result of the modelling sources in V5. The conventions are as in Fig. 2. A lateral view of the right hemisphere is presented below. V5 is shown in black. that the time we chose would have a substantial effect on the waveform of the evoked response. The percentage time a checkerboard moves with respect to the amount of time it remains stationary (duty cycle) and the time between successive movements (interstimulus interval) alters the waveform of the evoked activity (Schlykowa et al., 1993; Bach and Ullrich, 1994). Since our main concern was to detect the first signals arriving in the cortex and determine whether they arrived sequentially or in parallel, it seemed reasonable to choose a duty cycle and interstimulus interval that emphasized the earliest components of the evoked response. The only other checkerboard characteristic that needed to be changed was the check size. Like presentation time, the check size also has an important influence on the

7 Parallel inputs to VI and V5 MEG 1381 EEG waveform of the evoked response. We chose smaller checks (11' and 22') that emphasized the earliest components of the response (Onofrj et ai, 1991; Bodis-Wollner et ai, 1992). The choice of a control stimulus with which to compare our movement evoked responses, ideally one that did not activate the prestriate cortex, proved more awkward. One problem resulted from the fact that it was difficult to eliminate the impression of motion in reversing checkerboard stimuli. In the pattern reversal condition, all checks changed from black to white and visa versa at the same time, producing a strong sensation of a jumping movement. Similarly, a stimulus designed to match the moving checkerboard for the number of contrast changes occurring in a given time (asynchronous pattern reversal) resulted in a strong impression of incoherent motion. Another problem was that, in the macaque, the appearance of a stationary checkerboard (pattern onset) without an obvious directional or motion component activates the neurons of V5 and the related area V5A (MST) within 30 ms (Kawano et ai, 1994). One might predict, therefore, that all three stimuli would activate the prestriate cortex in exactly the same way as a moving checkerboard and thus prove poor controls. One solution was to use the opposite of pattern onset pattern offset where the checkerboard changes to an isoluminant grey. This stimulus did not produce an obvious impression of motion. In the end we decided to use pattern offset and the two reversal conditions to compare with fast motion. Subjects and materials Six male volunteers with normal corrected vision were tested with ethical approval. All gave their informed consent. Subjects 1 and 2 had previously been investigated with PET and the locations of VI/V2, V3 and V5 established. All six subjects had electrical evoked responses. Since we were using the MEG as a secondary measure, however, only half of these had simultaneous electrical and magnetic evoked responses. Table 1 shows the number of evoked response averages from each subject for each of the five conditions. The asynchronous reversal condition was tested in one subject and not used in the statistical comparisons. Subject 2 was unable to complete all four conditions, and we therefore have no record of his response to slow motion. The repeated averages from the same condition were summed to derive a single, individual mean response. Amplitude and latency values from the six individual mean responses were used in the statistical tests and the individual mean responses were combined to make grand average responses for each condition, thus ensuring that each subject contributed equally to the results. The stimuli were presented on a high resolution Amiga monitor (Commodore Business Machines Inc., West Chester, Pa., USA) with a screen refresh rate of 50 Hz (20 ms frame rate) and consisted of a stationary random checkerboard, identical in all conditions, covering the central 30 X20 of the visual field, each check subtending 11' or 22', presented for 900 ms (45 frames) at a contrast of 91% and a luminance of 109 cd/m2 with a central fixation point (as the checkerboard is not symmetrical, this value is the luminance of a matched, isoluminant grey field). To elicit a movement response, the stationary phase was followed by coherent horizontal motion in one direction at between 5 s "' and 22 s ~', for 200 ms (10 frames), using a hardware scroll to redraw the checkerboard displaced by a fixed number of pixels each frame. In the pattern offset condition, the stationary phase was followed by the checkerboard changing to an isoluminant grey, while in the pattern reversal condition the stationary phase was followed by a reversal of the checks from black to white and white to black and a second reversal in the opposite direction at 200 ms. In the asynchronous pattern reversal condition, the stationary phase was followed by a 200-ms period in which each check reversed independently and at random, in order to match the contrast changes at each point on the screen in the fast motion condition. Thus the moving or asynchronously reversing stimuli appeared to be identical when the field of view was restricted to a small region. The frequency of reversals in the asynchronous condition could be altered to match both the slow and fast moving stimuli (see Fig. 5). In order to minimize magnetic noise and thus allow Fig. 4 The result of modelling sources in VI and V5 simultaneously. The conventions are as in previous figures.

8 1382 D. H. ffytche et al. Table 1 The number of evoked response averages recorded Subject PET V5 EM EG Fast motion 4 4 Total Slow motion Pattern offset 2 Pattern reversal Asynchronous reversal 3 2 I a 3 i t6 13 MOTION stationary OFFSET stationary stationary grey grey stationary REVERSAL \ stationary stationary ASYNCHRONOUS REVERSAL Reversals stationary 900ms Multiple Random Reversals stationary 200ms I Fig. 6 Electromagnetoencephalography recording arrangements. Binoculars had X10 magnification and effective viewing distance of the screen was 0.45 m. The gantry supporting the dewar is not displayed TIME Fig. 5 A diagrammatic representation of the four stimulus types used. The cycle is repeated for 100 times to obtain each evoked response. Effective reversal frequency is 0.9 Hz. simultaneous recording of EEG and MEG, we placed the monitor as far from the MEG sensors as possible. To increase the visual angle subtended by the checkerboard to match the PET experiments, an optical system was designed to magnify the distant screen. Subjects lay prone on a pneumatic bed with their heads supported by a head rest. They viewed the monitor using a fixed pair of binoculars and mirror (see Fig. 6). MEG Recording arrangements The EEG was recorded using a grid of nine Ag/AgCl electrodes with 5 cm spacing over the hemisphere (one right and one left) that showed the greater V5 regional cerebral blood flow response in our previous PET study (Watson et al., 1993). In the four subjects who had not undergone a PET study, the grid was applied to the right hemisphere. The electrode locations (Fig. 7) were arranged so that the activity Fig. 7 The recording arrangements. The view is of the back of the head with Oz marked in the centre. The black dots show the electrode positions and are numbered to correspond to the Results section. The white disc represents the dewar, and has been displaced to reveal the underlying EEG electrodes. The real position during recording is shown by the circle on the occiput. The grey numbered circles in the dewar represent the gradiometer coils, numbered to correspond to the Results section. moving

9 in striate and prestriate cortex predicted by our model would be detected; thus for a right hemisphere application the top left electrode was placed at Oz (10% of the nasion-inion distance above the inion). To bring the lowest row of electrodes as close to the inferior surface of the skull as possible, subjects lay with their necks maximally flexed. In subject 5 an additional six electrodes were placed in symmetrically equivalent locations over the left hemisphere and a further electrode 5 cm above Oz. All electrode impedances were maintained below 3 kq and were referenced to Fz (midline electrode 30% of distance from nasion to inion). The EEG was band pass filtered between 0.3 s (30% reduction at 0.5 Hz) and 700 Hz on a Mingograph EEG Universal (Siemens Elma). The MEG was recorded using a hexagonal array of six second-order gradiometer pick-up coils coupled to radiofrequency superconducting quantum interference devices. The gradiometers operate at liquid helium temperatures and are housed in a non-magnetic glass-fibre dewar. The pickup coils were balanced in situ after each thermal cycle from room temperature. The dewar was supported from the ceiling by a wooden gantry that allowed it to be moved and tilted to optimize positioning with respect to the occiput. The recordings took place in an unshielded environment with noise levels of ft/vhz above 2 Hz. To help minimize ambient noise the recording sessions were performed at night. The MEG was band pass filtered between 0.75 and 100 Hz with additional notch filters at 50 and 150 Hz to remove mains interference. The MEG sensor positions with respect to the occiput were recorded in two of the subjects by measuring the distance of three markers on the dewar rim from appropriate EEG electrodes. Triggered by a pulse from the Amiga, an epoch of at least 312 ms of the filtered EEG and MEG was digitized at 1.6 KHz on a Microlink 16 channel 12 bit A/D converter, using a sample and hold procedure. A 60 ms pre-stimulus interval -was included in each epoch by triggering the A/D converter before the start of the moving, reversing or offset phase of the visual stimulus. The A/D converter was connected via an IEEE 488 standard bus to an IBMcompatible 486DX2 PC. Data acquisition and averaging were controlled by in-house software. A threshold voltage that discriminated eye blink or electromyographic artifact and artifact free EEG was set for each subject. Epochs exceeding the threshold were rejected on-line. One hundred sweeps of artifact-free EEG or EMEG were summed and averaged for each response. After DC linear trend and baseline correction, the averages were digitally filtered between 3 and 45 Hz using a discrete Fast Fourier transform method. All voltage measurements were made with respect to the pre-stimulus interval. For each electrode the voltage was displayed numerically in steps of 600 (is and the timing of the most negative number in the ms time window was taken to be the latency of Nl. In some electrodes the response in the same time window was dominated by PI, with the Nl component appearing as a deflection on the Parallel inputs to VI and V positive wave. In these instances the latency of Nl was taken as the most negative number in the deflection. Contour maps of the voltage distribution over the occiput were produced by linear interpolation method. Statistical comparisons were made using a two-way, repeated measures ANOVA (Electrode location X Condition) after normalization of the N1 component (McCarthy and Wood, 1985). Correlations in the time window of / = 0 ms to t = 270 ms between the evoked activty at Oz and each of the remaining eight electrodes (elecx) were produced for a range of time differences between the two records (7) using the formula: correlation (x) = Oz(()*elecX(; + T) Results The results demonstrate three salient features: that the EMEG topography in the movement evoked response cannot be explained by the activation of striate cortex alone; that signals arrive in the striate and prestriate cortex in parallel; that the parallelism is dependant on the characteristics of the stimulus. The grand average EMEG distributions in each of the five conditions are presented in Figs 8-10 and show several features. Figure 8A shows the grand average of fast motion responses. The evoked response is widely distributed over the occiput and, comparing these results with those in Fig. 8B, where the EEG electrodes are placed on both sides of the head, it can be seen that the response is symmetrical across the midline; it follows that we are not losing information by recording from one side of the head only. The evoked EEG activity shows an early positivity (P ), with an onset of 47 ms at the midline and 35 ms laterally. There follows a sequence of negative and positive waves (Nl, PI, N2). Since, for our purposes, the arrival of signals in the different cortical areas were of greatest interest, we have restricted our analysis to the earliest responses the P and Nl waves. The distribution of EEG voltage over the nine electrodes at the peak of the Nl response (78 ms) is shown in Fig. 8E where the maximum EEG amplitude is at the midline and there is a moderate downward gradient of voltage from midline to lateral electrodes (1.5:1). Simultaneously with the Nl response, the MEG shows a phase reversal (magnetic field coming out of the head in one set of sensors and going into the head in another) between the medial (6 and 5) and the lateral (1 and 3) MEG sensors (Fig. 8F). The distribution of the Nl and the associated MEG phase reversal closely matched our predictions of simultaneous striate and prestriate activity, whether the latter is due to V5 alone, or V5 and V3 (cf. Fig. 4). In fact, this distribution is specific to fast motion as the contour maps in Fig. 9C and F demonstrate. In both

10 1384 D. H. ffytche et al. B 1 j /"^ i^ f \ 1 1 D H 120 ms 63 ms 78ms EMEG 800 ft Fig. 8 (A) The grand average of six individual mean EEG evoked responses to fast motion (22 s '). Each numbered box shows the response from a single electrode whose position on the occiput is given in Fig. 7. The length of each trace is 312 ms, each small dash marking 12 ms. A 60 ms prestimulus interval is shown before the vertical line. The responses at electrodes Oz and 6 have been annotated to show the different waves, negativity is upwards. (B) The grand average of EEG evoked responses from Subject 5 with 16 EEG electrodes (four responses). The traces bounded by the dark square correspond to the electrodes in part A. Six electrodes are placed over the left hemisphere in equivalent positions and one electrode is placed 5 cm above Oz. The top left electrode has been omitted because of noise. The conventions are as in part A. (C) Inter-subject variability the six fast motion individual mean responses superimposed. The conventions are as in part A. (D) Intra-subject variability four repeat responses from Subject 1. The traces are averaged together to produce the individual mean. (E) The contour map derived from part A showing the distribution of voltage over the nine EEG electrodes at 63 and 78 ms. The lighter the shade of grey the higher the voltage. (F) The grand average EMEG record for fast motion. The MEG is shown in the shaded boxes. Sensor positions over the occiput are given in Fig. 7. The simultaneously recorded EEG is shown in the boxes immediately above. The conventions are as in part A. The vertical grey bands show the relationship between the Nl and the MEG phase reversal between sensors 5 and 6, with an up-going deflection, and sensors 3 and 1, with a down-going deflection. EEG

11 Parallel inputs to VI and V OFFSET B SIHIft 120ms ASYNCHRONOUS REVERSAL 120ms 3.5uV Fig. 9 (A) The grand average of six individual mean EEG evoked responses to pattern offset. (B) The grand average EMEG record for pattern offset. (C) The contour map derived from part A at Nl. (D) The grand average of six individual mean EEG evoked responses to pattern reversal. (E) The grand average EMEG record for pattern reversal. (F) The contour map derived from part D at Nl. (G) The grand average of the EEG evoked responses to asynchronous pattern reversal from Subject 5 (two responses). The conventions are as in previous figures.

12 1386 D. H. ffytche et al. the pattern offset and pattern reversal control conditions, the EEG at Nl shows a much steeper voltage gradient from midline to lateral (5:1 reversal and 6:1 offset). Figure 9B and E shows the simultaneous MEG phase reversal. The EMEG distribution in both control conditions matches our modelled predictions for striate activity (see Fig. 2). The difference in EEG distribution between fast motion and control conditions at Nl can be tested using a two-way ANOVA for repeated measures, after standardizing the Nl amplitude (see McCarthy and Wood, 1985). The Electrode locationxcondition interaction is significant in the comparisons of fast motion versus pattern offset [F(8,40) = 5.91, P< 0.001], and fast motion versus pattern reversal [^(8,40) = 4.86, P < 0.001] but not of fast motion [F(8,32) = 1.68, n.s.], suggesting that the cortical areas generating the fast motion response are different to those generating the pattern offset or pattern reversal responses, but are not significantly different to those areas generating the slow motion response. Asynchronous pattern reversal was not mapped or included in the statistical comparison because the evoked Nl, if present at all, was no larger than the noise levels in the pre-stimulus interval. Having established that the recorded EMEG field in the fast motion condition matches the modelled prediction for simultaneous striate and prestriate activity, we can turn to a consideration of the sequence of activation of these areas. An important observation is the fact that the lateral prestriate EEG activity starts at the same time or indeed slightly before the midline activity. This is evidenced by the change in EEG distribution between 63 and 78 ms shown in Fig. 8E. At 63 ms the EEG map matches the modelled distribution for V5 activity with a lateral peak and a midline minimum (cf. Fig. 3B); 15 ms later the map has changed to the simultaneous striate and prestriate pattern. The change in the contour map results from the fact that the Nl peak recorded laterally precedes the midline Nl peak by 2 ms (77±8 ms lateral electrode 6 versus 79±8 ms Oz). This latency shift is small compared with the differences in latency between subjects at any one electrode, although it is statistically significant in a subgroup of three individuals with responses of the same latency and prominent lateral Nl activity (P < 0.05 twotailed t test for related samples). A more elegant way of demonstrating the latency shift is to compare the entire Nl wave at the midline and laterally, rather than use the single value for latency. In order to enhance Nl before making the comparison, we first normalize the latency of Nl in each subject by shifting the individual mean responses in time until Nl in each subject is aligned at the Oz electrode. The resultant 'latency normalized' grand average response is filtered between 18 and 24 Hz to further emphasize the Nl component, although the result remains the same without this selective filtering. We then assess the relative timing of Nl at each electrode with respect to Oz by using the correlation method described above. Nl in the lateral electrodes 6 and 9 best matches Oz at approximately -6 ms (correlation electrode 6 = 0.75; electrode 9 = 0.85), con- B ttx& ^«FAST MOTION H^ ms E Fast motion lateral electrode 6 Midline electrode Oz x 0.9 Time difference between Oz and elelectrode. D c 3 H H...,.M,.,N2., SLOW MOTION _ Io. 0.25(lV Slow motion lateral electrode 6 Fig. 10 (A) The grand average of two individual mean (Subjects I and 5) EEG evoked responses to slow motion (<6 s"') are shown in the thin black trace. Superimposed stippled is the grand average fast motion response from the same two subjects. (B) Inter-subject variability the five slow motion individual mean responses superimposed. (C) Intra-subject variability three repeat responses from Subject I. The traces are averaged together to produce the individual mean. The conventions are as in previous figures. (D) The correlation between Oz and electrodes 3, 6 and 9 for fast and slow motion. Each numbered box shows the results from a single electrode. The vertical axis is the correlation between Oz and each electrode (Ot. 1: each dash = 0.1). The horizontal axis is the time difference (T) introduced between Oz and each electrode (-18 to +18 ms). The latency of Nl has been normalized at Oz between subjects. (E) The grand average of five individual mean EEG responses to slow motion and the corresponding five fast motion responses. The recordings from the midline (Oz) and laterally (electrode 6) have been superimposed and the time scale expanded. The responses have been filtered between 18 and 24 Hz to accentuate the Nl component. The latency of Nl has been normalized at Oz between subjects and conditions. firming that lateral Nl activity (the activity we are attributing to V5) precedes the midline or striate activity (Fig. 10D). If we make the assumption that signals cannot be relayed through an area without causing an EMEG deflection, then we are forced to one sole conclusion, namely that signals

13 EOG 120 ms j3(iv Fig. 11 A single fast motion response with simultaneous electrooculogram from Subject 1 (electro-oculogram filtering: high pass 0.3 s, low pass 15 Hz ) The conventions are as in Fig. 8A. are arriving in the prestriate and striate cortex independently and in parallel. Our argument that the parallel activation of the prestriate cortex is velocity-dependant would be greatly strengthened if the early V5 activity were not found with slow moving stimuli. This is exactly what we found. For slow moving stimuli, the lateral activity follows the medial activity by 5 ms (100±8 ms Oz, 105±10 ms lateral: P < 0.05 twotailed / test for related samples of five individual means). Correlating the entire Nl wave medially and laterally, as for fast motion, the lateral Nl in electrodes 6 and 7 follows the midline by -12 ms (Fig. 10D correlation electrode 6 = 0.7; correlation electrode 3 = 0.75). Figure 10A shows that the slow motion response (shown in black) is delayed with respect to the fast motion response (shown in grey); in order to illustrate the difference between slow and fast motion, we have normalized the responses at Oz for each individual in both conditions to a single value. The results are presented in Fig. 10E where the temporal relationships between the two conditions and electrode location for the Nl response are summarized. In order to ensure that eye movement artifact did not account for the more lateral distribution of the fast motion activity, we recorded the electro-oculogram of one subject. Eye movements are associated with MEG activity detectable at virtually any scalp area (Oakley et al., 1989) and could be a source of EEG artifact by influencing the reference electrode. The record, shown in Fig. 11, demonstrates that the evoked response is not contaminated by eye movement. Discussion From the results given above, we draw the following conclusions, (i) Signals relating to fast visual motion reach the prestriate cortex before they reach the striate cortex. It follows that there are parallel inputs to striate and prestriate cortex, at least as far as motion is concerned, (ii) Signals relating to slow visual motion do not reach the motionspecialized prestriate cortex in parallel, but rather are relayed through VI. It follows that the parallelism we speak of is not rigid and ever present; rather it is dynamically tuned to the stimulus. Parallel inputs to VI and V EEG, MEG and modelling In discussing the evoked activity we will highlight several points. (i) Our evoked responses to pattern offset and pattern reversal EEG are identical to those obtained in other studies (Onofrj et al., 1991; Halliday, 1993). In the fast motion condition the PI response corresponds to the activity described by Clarke (1973a, b), Spekreijse et al. (1985) and Van Dijk et al. (1987). (ii) The evoked response does not end with the P and Nl waves, and there are obvious differences between the motion and pattern offset conditions when considering PI and N2 (cf. Figs 8A and 9A), but we have restricted ourselves to a consideration of the early components as these later activities must reflect not only the signals arriving in an area for the first time, but also processing within an area, as well as the return of signals along the descending projections from higher areas that characterize the circuitry of the visual cortex (Zeki and Shipp, 1988). (iii) Our analysis of the different conditions is not of the amplitude of the early evoked activity, which varies between conditions, but rather its distribution as revealed in the Nl contour maps. The amplitude of Nl is greatest for pattern offset or pattern reversal, less for fast and slow motion and absent for asynchronous reversal. We believe that this does not imply that more signals reach the cortex in the former as compared with the latter conditions but rather that some stimuli are able to evoke a clear early component, while others are not. (iv) The evoked responses vary in the prominence of P in the grand average figures; being a distinct feature of fast and slow motion, but not of pattern reversal and pattern offset. (v) The MEG has a distinct phase reversal from midline to lateral occiput in all stimulus conditions, although the amplitude of the reversal varies with the size of Nl. As our modelling predicted, the addition of prestriate activity in the fast motion condition, while changing the EEG topography, did not alter the MEG topography. The interpretation of our evoked responses, attributing the differing patterns of activation to different cortical areas, depends on the validity of our model. We have used the average Talairach brain in the model (Talairach and Tournoux, 1988) and are thus obscuring individual variations in the gyrification of the occipital lobe and in the location of the functionally specialized areas (Brindley, 1972; Ono et al., 1990; Watson et al., 1993). However, this possible limitation is an advantage when using this model to interpret an average evoked response from several individuals. We also make the assumption that the summed postsynaptic activity that causes a change in regional cerebral blood flow will, if synchronized, be a current source that contributes to the evoked potential. There is some evidence that this is the case (Celesia et al., 1982, 1984) but, to our knowledge, no direct measurements comparing regions of increased cerebral blood flow and the location of cortical currents have been made. Finally, our

14 1388 D. H. ffytche et al. latency and amplitude comparisons assume that the Nl wave recorded by a given electrode reflects physiological activity in the cortex immediately beneath it. An alternative interpretation might be that the Nl wave is the sum of activity in several disparate cortical regions, remote from the electrode and that the differences in the shape and latency of Nl result from changes in the contribution each region makes to the final wave. Whichever interpretation is chosen, both lead to the same conclusion, that different visual areas or combinations of areas process slow motion, fast motion and control stimuli. The absolute latencies of signal arrival in the visual cortex All the physiological evidence suggests that signals can reach the visual cortex at very short latencies. The optic nerve contains a range of fibre diameters and should, therefore, have a range of conduction velocities (Chako, 1948). The fastest conducting fibres have been described in cat optic nerve and found to conduct at a velocity of 70 m s~' (Hsiang- Tung Chang, 1956). The fastest fibres in the monkey, which we prefer to use for our comparisons, have a speed of 20 m s~' for fibres outside the eye and 1.2 m s~' for the intraretinal portion (Ogden and Miller, 1966). Assuming that some fibres in the human optic nerve have similar conduction velocities, we estimate that the fastest signals can reach the cortex of V5 in -30 ms. Our figure is based on the following calculations: (i) a 22 ms latency between onset of the visual stimulus and a response in the retinal ganglion cell (Kuffler, 1953); (ii) 1 mm of intraretinal nerve fibre (1 ms conducted at 1.2 m s"'); (iii) 12 cm between the ganglion cell layer of the retina and the cortex (6 ms if conducted at 20 m s~'); (iv) delays of 0.5 ms at the synapse in the lateral geniculate nucleus or, if the route via the superior colliculus and pulvinar is used, a delay of 1 ms. This should give a minimum latency of -30 ms from onset of the visual stimulus to arrival of signals in V5 or VI, not taking into account synaptic delays within the cortex itself. Our estimate of minimal latency cannot be exact as we know that the ganglion cell response depends on the contrast and luminance of the stimulus (Kuffler, 1953; Shapley and Victor, 1978) and we do not know how the human ganglion cell would respond under our experimental conditions. On the other hand, the very fast predicted latency compares well with the figures presented in Table 2. Given that the peak of activity (P ) that we have recorded from what we presume to be the prestriate cortex is 49 ms, it follows that there is a lag of -20 ms between the first arrival of signals in the cortex and the ability of a technique such as visually evoked EEG or MEG to detect them. There are several explanations to account for this difference. The first is that, because of the range of conduction velocities in the optic nerve, signals will arrive in an area spread over time, thus accounting for the range of response latencies within any given region of cortex (Wilson et al, 1983; Raiguel et al, 1989). The peak of an evoked response component like P may thus reflect the average latency of signals arriving in an area while the onset of the scalp recorded evoked response may be better related to the minimum latency. The onset of P in the fast motion condition is at 35 ms in the lateral electrodes and 47 ms medially, a result that is in closer agreement with the intracortical recordings. Since a certain minimum number of cortical neurons must be active and synchronized to be detectable through the attenuating effects of CSF, bone and scalp, it may be that the fastest conducting fibres are too few in number to be detected (Cooper et al, 1965). Another explanation follows from the demonstration that, for pattern reversal, visual stimuli in monkey visual cortex, the first signals arriving in the cortex from the lateral geniculate nucleus form a closed field (Schroeder et al, 1991). In other words, while activity is demonstrable in the layer 4C of the striate cortex with intracortical electrodes, this activity is not detectable by surface electrodes. Assuming that these findings apply equally to the human brain, it is thus quite possible that, in the movement response as in the reversal, the arrival of signals in layer 4 of V5 or layer 4C of VI will not be externally recordable and therefore that the first activity we detect is caused by later processing within the same areas. Whatever the cause of the discrepancy between the onset of the surface and intracortical recordings, if we assume that the explanation is the same for VI and V5, then we can use the peak of activity in the earliest waves (P and Nl) as an index, admittedly an imperfect one, of signals arriving in the two areas. The fact that the lateral prestriate activity is detectable on the scalp before the medial striate activity implies parallelism even if we are estimating the arrival of signals a few milliseconds too late. Comparison to transcranial magnetic stimulation results The experiments undertaken in this laboratory using transcranial magnetic stimulation (TMS) (Beckers and Zeki, 1995) to interfere with the ability to perceive motion suggest that signals from moving stimuli reach V5 before 30 ms, a result entirely compatible with the neurophysiology of the visual pathway in man and in close agreement with the results presented here. The estimate of arrival of signals derived from this study is slightly later (35 ms) but the difference could easily be accounted for by any of the explanations given above. The TMS experiment also suggested that the significant processing of motion signals in VI occurs ms after V5. The present study did not detect such a time difference between the activation of V5 and VI, nor have direct extracellular recordings from the monkey cortex. The simplest explanation to account for the difference between the TMS and the electrophysiological results would be that, for fast moving stimuli, the perceptually

15 Parallel inputs to VI and V Table 2 The absolute latencies of signals in the visual cortex Study Technique Area Minimum latency (ms) Wilson et al. (1983) Whittaker and Siegfried (1983) Krauts al. (1985) Ducati et al. (1988) Petersen et al. (1988) Raiguel et al. (1989) Schroeder et al. (1991) Maunsell and Gibson (1992) Givre et al. (1994) Kawanoe/a/. (1994) Yonedat et al. (1995) Beckers and Zeki (1995) Man Man Macaque Man Owl monkey Macaque Macaque Macaque Macaque Macaque Man Man important processing is mediated by early signals arriving in V5; disrupting the early parallel input to VI is of no consequence. Early signals may reach V1 and be detectable using electrophysiological techniques, but may not be necessary for the perception of motion. It might be that these fast signals arriving in V1 are not coding motion information, but may be conveying different aspects of the stimulus, pattern components for example. The later motion signals that do reach VI may pass along slower conducting fibres and therefore, by stimulating VI at 60 ms, motion perception is impaired. Whatever the explanation, the two techniques seem to provide complementary insights into the visual cortex, for the TMS method can show the timing of perceptually salient signals in a way that scalp or even cortically recorded electrophysiology cannot. Despite a century of endeavour, we are far from understanding the relationship between activity in neuronal populations and conscious experience. Parallel inputs to striate and prestriate cortex Our evoked responses have led us to conclude that there are two motion pathways to the visual cortex. One that is dominant when the motion is slow and is perhaps derived mainly from the parvocellular layers of the lateral geniculate nucleus; it arrives first in VI and conveys information to the prestriate cortex in the conventional sequential manner reported by others. We speculate that, in this case, the primary cortical processing occurs in VI. Translated to the pathological state, this allows a patient with a lesion in V5 to discriminate slow motion through the intact visual areas, including VI, as has been already shown (Hess et al., 1989; Shipp etai, 1994). The second pathway becomes increasingly dominant as stimuli move at speeds in excess of 6 s" 1. It is probably derived mainly from the magno-cellular layers of the lateral geniculate nucleus and passes to VI, V5 and possibly V3 in parallel, without a hierarchical relay. We speculate that it is this pathway that allows a patient with a Single unit Scalp VEP, flash Multiunit, flash Intra cerebral Single unit, flashed bars Single unit, moving bars Multiunit and VEP, reversal and flash Single unit, stationary grating Multiunit, flash Single unit, motion and pattern onset Scalp VEP/MEG flash Magnetic stimulation, motion Occipital cortex Striate cortex Medial occipital cortex V5 Striate cortex, V2, V5 Striate cortex Striate cortex Striate cortex, V4 V5, MST Striate cortex V5, striate cortex V5, 45-V1.71-V2 30 (from diagram, peak at 40) , VI and V4 37 (velocity dependant) 39^4 <3O-V5, 65-V1 lesion in VI, and therefore an apparent hemianopia, to discriminate rapid motion presented to his 'blind field' consciously (see Barbur et al., 1993). The experiments we report in this paper are therefore, in a sense, no more than a formal demonstration of conclusions that can be directly derived from the published clinical literature and it is surprising that no one has yet done so. Here we emphasize that we are not stating that all visual inputs to cortex have parallel components that go to VI and to the prestriate cortex; in fact, our results show that the parallelism is a dynamic process, even within a single sub-modality such as motion, and may depend upon the characteristics of the visual stimulus. The notion of a parallel input to striate and prestriate cortex should not seem outrageous. There is much in both the anatomical and physiological literature that favours such a conclusion, although the evidence is based largely on the monkey. Anatomically, it has been shown that there is a direct input to the prestriate cortex from both the lateral geniculate nucleus (Fries, 1981; Yukie and Iwai, 1981) and the pulvinar (Standage and Benevento, 1983). Although this alternative pathway has been little studied compared with the massive pathway that reaches V5 through VI, there is no reason to suppose that it has no function or that it does not contribute in an important way to the physiology of V5. One can get an idea of its importance by studying that physiology in animals in which VI has been inactivated, either through lesions or by cooling (Rodman et al., 1989; Girard et al., 1992). The characteristic physiology of V5, the presence of directional selectivity (Zeki, 19747?), is not lost; rather it is maintained, with the cells losing only their narrow sensitivity profiles for direction. The conclusion derived from these animal studies by Bullier et al. (1993) is that the accessory visual motion pathway which arrives in V5 without passing through VI carries signals 'which may not reach consciousness'. In fact, the importance of this alternative pathway in man is shown in our studies on a patient blinded by lesions of VI, who is nevertheless conscious of both the

16 1390 D. H. ffytche et al. presence and direction of motion in his blind field (Barbur et al., 1993). In this patient, areas V5 and V3 are active. Clearly such residual capacity can only be accounted for by the activity of this alternative pathway which reaches V5 without passing through VI and can therefore mediate a simple and crude conscious vision without pre- or postprocessing in VI. Moreover, the notion that signals may arrive in V5 before VI is evident in the only direct electrophysiological study to compare the timing of arrival of signals in VI and V5 (Raiguel et al., 1989), although the authors did not draw the obvious conclusion. For if, as their results show, the shortest latency cells are found in V5, the conclusion would be the same as the one that we have drawn from our present study, namely that there is a fast parallel pathway to V5 which bypasses VI. Finally, intracerebral recordings in monkey V5A (MST), an area closely related to area V5, have demonstrated an 8 ms shift in latency (from 52 to 60 ms) with an associated decrease in spike count when comparing fast (20 s~') with slow (10 s" 1 ) motion (Kawano et al., 1994). While the human scalp activity occurs 16 ms later, this shift is equivalent to the change in amplitude and latency of the lateral EEG activity between the slow and fast motion conditions described here. We believe that the shift in latency reflects the change from direct to indirect inputs to the prestriate cortex. Comparison of results with previous evoked potential studies The evidence for parallel inputs to striate and prestriate cortex is therefore not new; it is simply that its significance was not appreciated and the conclusions were not drawn. We discuss below the question why previous studies using the evoked potential technique have failed to identify this important principle of the organization of the visual pathways, bearing in mind that the aims of studies may differ, that the technology available today is better than that available only a few years ago and that much of the information which allows us to interpret our results in a new way has only been recently described. There are probably several reasons which have allowed us to identify a fast, parallel, early component which evoked response studies had not identified before. Among these is the problem of how to interpret the evoked EMEG waveform. Some authors have not attempted to interpret the evoked activity in physiological terms (Clarke, 1973a, b; Schlykowa et al., 1993; Bach and Ullrich, 1994). Other studies have chosen specific stimulus characteristics for the purpose of studying the late EEG activity evoked by motion and have therefore lost the opportunity to investigate the onset of the early waves (Schlykowa et al, 1993). The shift in the waveform from dominant early to dominant late components results from changes in the interstimulus interval and the duty cycle {see Bach and Ullrich, 1994). These authors have argued that because of adaptation, the evoked component specific to motion will get smaller if, during each moving/ stationary cycle, the moving phase is longer than the stationary phase. Using this argument, they identified the later negativity as being motion specific. In fact, our evidence suggests that the distribution and latency of the earliest detectable activity has motion specific components as well. Another reason why earlier studies may not have identified parallel activation of VI and V5 is because of the velocity dependence. Table 3, below, shows that, with the exception of Clarke's study, all have used speeds of 6 s" 1 or less, and most have been below 5 s~', i.e. speeds at which the early, parallel, activity in V5 is not detectable in our records. Although Clarke did not demonstrate parallel activation with fast stimuli, he was only using two electrodes and did not, therefore, map the surface distribution of EEG activity. Thus, even though he noted a shift in the latency of PI with high velocities (>10 s~'), he made the reasonable, though incorrect, assumption that the evoked motion response at these speeds was the same as the pattern offset response. The difference between our findings and the previously published literature may also lie in the techniques we have used. Previous studies (e.g. Maier et al., 1987; Probst et al., 1993) have relied on a mathematical solution of the inverse problem to localize current sources; we did not do so, but used instead the published PET data to infer the position of the source. Mathematical dipole solutions can obscure the distributed nature of a current source by constraining the position of the source to one or more points. The dipole solution is a poor mathematical approximation that accounts for the evoked EEG or MEG field, but does not necessarily need to reflect the true location of the physiological activity {see Balish and Muratore, 1990). Finally, in a more general but no less important sense, one gets the distinct (and understandable) impression that the time-old picture of a visual input to the brain for which VI is the exclusive gateway, together with an absence of information about the visual areas of the prestriate cortex in the human brain, constrained authors to interpret their results in a more classical tradition. For example, some authors considered that one cannot obtain prestriate activity using motion stimuli. Maier et al. {1987), using principle component analysis of the scalp recorded EEG, state that the motion 'component shows a clear area 17 origin', while Van Dijk et al. (1987), recording subdurally from the macaque using strips of electrodes state that the 'motion specific' responses 'originate in the primary visual cortex'. But Van Dijk's subdural strips did not extend into the superior temporal sulcus, where V5 is buried in the monkey. In the human brain V5 is superficial and in preliminary direct recordings from the human occipital cortex that we have carried out with David Sandeman and Stuart Butler and their colleagues at Bristol, we have found that the cortical evoked response to motion does arise in V5 as well as VI. Probably for similar historical reasons, the evidence for parallelism was not entertained even when the evoked potential evidence could be interpreted to suggest just such a possibility. For

17 Parallel inputs to VI and V Table 3 Visual evoked responses to motion Study and technique Montage s~ ] Waveform VI latency V5 latency Mackay and Rietveld (1968) EEG Clarke (1972, I973«, b) EEG: lower field Spekreijse et al. (1985) EEG: full-field and hemifield Maier et al. (1987) EEG: hemifield Van Dijk et al. (1987) EEG Kaufman and Williamson (1990) MEG: full-field (change in velocity) Schlykowa et al. (1993) EEG: full-field and lower field Probst et al. (1993) EEG: hemifield Drasdo et al. (1993) EEG Bach and Ullrich (1994) EEG: full-field *One subject only. Oz Reference: ear Inion cm lateral Reference: right mastoid 0z+3 cm. 4 cm lateral Reference: mid-frontal 24 electrode grid over occiput Reference: mid-frontal Monkey subdural and human inion Reference: vertex Seven sensors 5cm lateral to Oz Reference: ear 29 electrode full head montage Mapping over occiput. Reference: Fz Oz+5 cm lateral Oz Reference: variable Not stated (slow and fast) 3 5 Not stated example, as late as 1991, Ossenblok and Spekreijse found in their studies with pattern onset that, of the three components, the earliest could be traced to 'area 18', the middle component to VI and the last component to 'area 19'. In spite of this, they do not discuss the possibility of a parallel input, even if only to dismiss it. The evidence that we give here confirms therefore the classical picture of a sequential input to cortical areas, but goes beyond, to show that there is an alternative, parallel and stimulus-dependent, input to V5. It follows that the visual input to area V5 is a good deal more complex than we had previously supposed and that parallel retino-cortical inputs to it and to VI constitute one of its important features. It shows, moreover, that these parallel inputs are in daily use and are therefore not, as Gross (1991) has argued on the basis of cortical and sub-cortical lesions in monkeys, redundant in the intact brain, expressing themselves only in the pathological state. The fact that the parallelism that we describe, depends upon the characteristics of the stimulus and is not a mandatory pathway, active whenever a visual stimulus is in motion, should perhaps be seen in a somewhat broader context, as an extension of the concept of operational connections (Zeki, 1990, 1993). That concept supposes that 4.9 N60 P96 (NIO0)* P140 NI90 N80 PI 12 Not stated PI 25 Not shown P96 N220 Contralateral Nl 12 PI88 N265 PI06 N163 P110 N180 No comment No comment No comment Mathematical principle component analysis: motion response from VI at 120 ms All response derived from VI Occipital source identified but latency not stated Mathematical source analysis: V1 Source at 100 ms PI06 from VI No conmienl No comment No comment No comment No contribution of area 18 or 19 to motion response No contribution of V5 to motion VEP Mathematical source analysis: movement area in rolandic fissure at 200 ms Area V5 responsible for N220 component Mathematical source analysis: V5 Source at 189 ms NI63 from V5 No comment N180 is motion specific when two areas, A and B, are in anatomical connection, not all the operations carried out by A are relayed to B but only those that are of importance to B; different sets of operational connections may come into use and transmit signals to B, depending upon the operation. In the same way, the presence of a parallel anatomical pathway from the retina to V5 does not mean that it is always in use but only when the stimulus in the field of view necessitates it. Conclusion A straightforward and simple clinical observation of the consequences of lesions in human VI and V5 on motion perception, together with the results of studies on single cell responses in the human and monkey brain, have led us to a study that, without recourse to anatomical tracing methods, has given us new and unexpected insights into the circuitry underlying visual motion perception. We have used the relative merits of several different techniques in combination, in order to overcome the shortcomings of each individually. The MEG and EEG equipment that we have used is relatively unsophisticated, with only nine electrodes and six gradiometers. In addition, both the EEG and the MEG methods

18 1392 D. H. ffytche et al. have their limitations, partly because of the insoluble inverse problem and partly because of our poor understanding of how to interpret a surface electrical or magnetic wave in terms of the underlying physiology. The PET method that we have relied on has no temporal resolution and stimulation methods such as TMS, and can only disrupt with confidence cortical areas that are at or very near the brain surface, thus limiting their use. There are, of course, other non-invasive techniques that we could turn to and the most promising among these is that of functional MRI. We plan to use this method in the future but it was perhaps fortunate that we did not do so for this study, for it can only indicate changes in neurophysiological activity indirectly by measuring changes in blood flow. It follows that the temporal resolution of functional MRI is limited by the latency of the change in blood flow in response to a stimulus, which is 500 ms (Cohen and Bookheimer, 1994). This is much too slow to detect the changes that we report here, which show that many of the important events leading to conscious visual perception are well over before the first change in blood flow is detected by functional MRI. This is not to imply that there are not other significant cortical processes related to visual motion and to other visual processes which occur much later and would therefore be amenable to better study with functional MRI, but only that the study of the arrival of signals in the areas of the cerebral cortex is beyond its current temporal resolution. Thus, in the absence of the perfect investigative tool, we have opted for the interim solution of combining methods, none of them ideal but which, jointly, have allowed us to investigate problems of great interest with existing technology. At any rate, it is a source of satisfaction to us that, using relatively simple methods and results already published, we were able to gain a new and unexpected, if small, insight into the complexities of the visual brain. Acknowledgements We wish to thank all our subjects for their patience and cooperation during the tests. We are also grateful to John Romaya for adapting the PET stimulus, as well as Gareth Davies, Simon Walker and Declan McKeefry for their help with the EEG and MEG recordings. This work was supported by the Wellcome Trust. D.H.ff. is a Wellcome Research Training fellow. References Alarcon G, Guy CN, Binnie CD, Walker SR. Elwes RD, Polkey CE. Intracerebral propagation of interictal activity in partial epilepsy: implications for source localisation. J Neurol Neurosurg Psychiatry 1994;57:435^t9. Bach M, Ullrich D. Motion adaptation governs the shape of motionevoked cortical potentials. Vision Res 1994; 34: Balish MS, Muratore R. The inverse problem in electroencephalography and magnetoencephalography. Adv Neurol 1990; 54: Barbur JL, Watson JD, Frackowiak RS, Zeki S. Conscious visual perception without VI. Brain 1993; 116: Beckers G, Zeki S. The consequences of inactivating areas VI and V5 on visual motion perception. Brain 1995; 118: Bodis-Wollner I, Brannan JR, Nicoll J, Frkovic S, Mylin LH. A short latency cortical component of the foveal VEP is revealed by hemifield stimulation. Electroencephalogr Clinical Neurophysiol 1992; 84: Brindley GS. The variability of the human striate cortex. J Physiol (Lond) 1972; 225: 1-3P. Bullier J, Girard P, Salin P-A. The role of area 17 in the transfer of information to extrastriate visual cortex. [Review]. In: Peters A, Rockland KS, editors. Cerebral cortex, Vol. 10. New York: Plenum Press, 1994: Celesia GG, Polcyn RD, Holden JE, Nickles RJ, Galley JS, Koeppe RA. Visual evoked potentials and positron emission tomographic mapping of regional cerebral blood flow and cerebral metabolism: can the neuronal potential generators be visualized? Electroencephalogr Clin Neurophysiol 1982; 54: Celesia GG, Polcyn RD, Holden JE, Nickles RJ, Gatley JS, Koeppe RA. Visual evoked potentials and regional cerebral blood flow: a method to visualize neuronal potential generators. In: Nodar RH, Barber C, editors. Evoked potentials II. Boston: Butterworth, 1984: Celesia GG, Bodis-Wollner I, Chatrian GE, Harding GF, Sokol S. Spekreijse H. Recommended standards for electroretinograms and visual evoked potentials. Report of an IFCN committee [see comments]. Electroencephalogr Clin Neurophysiol 1993; 87: Comment in: Electroencephalogr Clin Neurophysiol 1994; 91: Chacko LW. An analysis of fibre-size in the human optic nerve. Brit JOphthal 1948:32: Clarke PG. Visual evoked potentials to sudden reversal of the motion of a pattern. Brain Res 1972; 36: Clarke PG. Visual evoked potentials to changes in the motion of a patterned field. Exp Brain Res 1973a; 18: Clarke P. Comparison of visual evoked potentials to stationary and to moving patterns. Exp Brain Res 1973b; 18: Clarke S. Modular organization of human extrastriate visual cortex: evidence from cytochrome oxidase pattern in normal and macular degeneration cases. Eur J Neurosci 1994; 6: Cooper R, Winter AL, Crow HJ. Walter WG. Comparison of subcortical, cortical and scalp activity using chronically indwelling electrodes in man. Electroencephalogr Clin Neurophysiol 1965: 18: Cohen MS, Bookheimer SY. Localization of brain function using magnetic resonance imaging. [Review). Trends Neurosci 1994; 17: Cragg BG. The topography of the afferent projections in the circumstriate visual cortex of the monkey studied by the Nauta method. Vision Res 1969: 9: 733^17. Drasdo N, Edwards L, Thompson DA. Models of the visual cortex based on visual evoked potentials. In: Gulyas B, Ottoson D, Roland PE, editors. Functional organisation of the human visual cortex. Oxford: Pergamon Press, 1993:

19 Ducati A, Fava E, Motli ED. Neuronal generators of the visual evoked potentials: intracerebral recording in awake humans. Electroencephalogram Neurophysiol 1988: 71: ffytche DH, Walker S, Guy C, Zeki S. A comparison of the latency of signals in human VI and V5, using the technique of visual evoked response to motion (VERM) [abstract]. J Physiol (Lond) 1994; 477: 57P. Fries W. The projection from the lateral geniculate nucleus to the prestriate cortex of the macaque monkey. Proc R Soc Lond B Biol Sci 1981:213: Girard P, Salin PA, Bullier J. Response selectivity of neurons in area MT of the macaque monkey during reversible inactivation of area V1. J Neurophysiol 1992; 67: Givre SJ, Schroeder CE, Arrezo JC. Contribution of extrastriate area V4 to the surface-recorded flash VEP in the awake macaque. Vision Res 1994; 34: Gross CG. Contribution of striate cortex and the superior colliculus to visual function in area MT, the superior temporal polysensory area and the inferior temporal cortex. Neuropsychologia 1991; 29: Guy CN. Magnetic vector potential calculations without integration. Phys Med Biol 1990; 35: Halliday AM. Evoked potentials in clinical testing. 2nd editon. London: Churchill Livingstone, Hess RH, Baker CL Jr, Zihl J. The 'motion-blind' patient: low-level spatial and temporal filters. J Neurosci 1989; 9: Horton JC, Hoyt WF. The representation of the visual field in human striate cortex. Arch Ophthalmol 1991; 109: Hsiang-Tung Chang. Fiber groups in primary optic pathway of cat. J Neurophysiol 1956; 19: Kaufman L, Williamson SJ. Neuromagnetic localization of neuronal activity in visual and extravisual Cortex. In Cohen B and Bodis- Wollner I, editors. Vision and the brain. New York: Raven Press 1990: Kavanagh RN, Darcey TM, Lehmann D, Fender DH. Evaluation of methods for three-dimensional localization of electrical sources in the human brain. IEEE Trans Biomed Eng 1978; 25: Kawano K, Shidara M, Watanabe Y, Yamane S. Neural activity in cortical area MST of alert monkey during ocular following responses. J Neurophysiol 1994; 71: Kraut MA, Arrezo JC, Vaughan HG Jr. Intracortical generators of the flash VEP in monkeys. Electroencephalogr Clin Neurophysiol 1985; 62: Kuffler SW. Discharge patterns and functional organization of mammalian retina. J Neurophysiol 1953; 16: Lopes da Silva F, Van Rotterdam A. Biophysical aspects of EEG and magnetoencephalogram generation. In: Niedermeyer E, Lopes da Silva F, editors. Electroencephalography. 2nd ed. Baltimore: Urban and Schwarzenberg, 1987: Lorente de No R. Action potential of the motoneurons of the hypoglossus nucleus. J Cell Comp Physiol 1947; 29: Parallel inputs to VI and V Mac Kay DM, Rietveld WJ. Electroencephalogram potentials evoked by accelerated visual motion. Nature 1968; 217: Maier J, Dagnelie G, Spekreijse H, van Dijk BW. Principal components analysis for source localization of VEPs in man. Vision Res 1987; 27: Maunsell JH. Gibson JR. Visual response latencies in striate cortex of the macaque monkey. J Neurophysiol 1992; 68: McCarthy G, Wood CC. Scalp distributions of event-related potentials: an ambiguity associated with analysis of variance models. Electroencephalogr Clin Neurophysiol 1985; 62: Oakley M, Sullivan G, Roeder S, Flynn E. Scalp distribution of magnetic fields generated by horizontal eye movements and eye blinks. In: Williamson S, Hoke M, Stroink G, Kotani M, editors. Advances in biomagnetism. New York: Plenum Press, 1989: Ogden TE, Miller RF. Studies of the optic nerve of the rhesus monkey: nerve fiber spectrum and physiological properties. Vision Res 1966; 6: Ono M, Kubik S, Abernathey CD. Atlas of the cerebral sulci. Stuttgart: Thieme, Onofrj M, Bazzano S, Malatesta G, Fulgente T. Mapped distribution of pattern reversal VEPs to central field and lateral half-field stimuli of different spatial frequencies. Electroencephalogr Clin Neurophysiol 1991; 80: Ossenblok P, Spekreijse H. The extrastriate generators of the EP to checkerboard onset. A source localization approach. Electroencephalogr Clin Neurophysiol 1991; 80: Petersen SE, Miezin FM, Allman JM. Transient and sustained responses in four extrastriate visual areas of the owl monkey. Exp Brain Res 1988; 70: Probst T, Plendl H, Paulus W, Wist ER, Scherg M. Identification of the visual motion area (area V5) in the human brain by dipole source analysis. Exp Brain Res 1993; 93: Raiguel SE, Lagae L, Gulyas B, Orban GA. Response latencies of visual cells in macaque areas VI, V2 and V5. Brain Res 1989; 493: Regan D. Human brain electrophysiology: evoked potentials and evoked magnetic fields in science and medicine. New York: Elsevier, Rodman HR, Gross CG, Albright TD. Afferent basis of visual response properties in area MT of the macaque: 1 Effects of striate cortex removal. J Neurosci 1989; 9: Romani G. Fundamentals on neuromagnetism. In: Williamson S, Hoke M, Stroink G, Kotani M, editors. Advances in biomagnetism, New York: Plenum Press, 1989: 33^16. Schlykowa L, van Dijk BW, Ehrenstein WH. Motion-onset visualevoked potentials as a function of retinal eccentricity in man. Brain Res Cog Brain Res 1993; I: Schroeder CE. Tenke CE. Givre SJ. Arrezo JC. Vaughan HG Jr. Striate cortical contribution to the surface-recorded pattern-reversal VEP in the alert monkey [published erratum appears in Vision Res 1991; 31: I]. Vision Res 1991; 31: Shapley RM, Victor JD. The effect of contrast on the transfer

20 1394 D.H. jfytche et al. properties of cat retinal ganglion cells. J Physiol (Lond) 1978; 285: Shipp S, de Jong B, Zihl J, Frackowiak RS, Zeki S. The brain activity related to residual motion vision in a patient with bilateral lesions of V5. Brain 1994; 117: Shipp S,Watson J, Frackowiak R, Zeki S. Retinotopic maps in human prestriate visual cortex: the demarcation of areas V 2 and V 3. Neuroimage 1995; 2: Spekreijse H, Dagnelie G, Maier J, Regan D. Flicker and movement constituents of the pattern reversal response. Vision Res 1985; 25: Standage GP, Benevento LA. The organization of connections between the pulvinar and visual area MT in the macaque monkey. Brain Res 1983; 262: Stensaas SS, Eddington DK, Dobelle WH. The topography and variability of the primary visual cortex in man. J Neurosurg 1974; 40: Talairach J, Tournoux P. Co-planar stereotaxic atlas of the human brain. Stuttgart: Thieme, van Dijk BW, Dagnelie G, Spekreijse H. Motion onset-offset visual evoked potentials from rhesus visual cortex. In: Barber C, Blum T, editors. Evoked potentials III: the Third International Evoked Potentials Symposium. Boston: Butterworth, 1987: Watson JD, Myers R, Frackowiak RS, Hajnal JV, Woods RP, Mazziotta JC, et al. Area V5 of the human brain: evidence from a combined study using positron emission tomography and magnetic resonance imaging. Cereb Cortex 1993; 3: Whittaker S, Siegfried J. Origin of wavelets in the visual evoked potential. Electroencephalogr Clin Neurophysiol 1983; 55: Wilson CL, Babb TL, Halgren E, Crandall PH. Visual receptive fields Appendix Modelling procedure A three-dimensional, smoothed surface outline of the occipital lobe was constructed, based on the atlas of Talairach and Tournoux (1988). The left and right hemispheres were composed of a series of 44 transverse slices, separated by 2 mm and orientated parallel to the calcarine fissure. Each slice was an ellipse whose axis was twisted with respect to the Talairach y-axis (sagittal) and whose focal point varied in the z-axis (transverse). The calcarine fissure and the inferior temporal gyrus were introduced by local spatial modulations of the ellipses. The surface was constructed by joining corresponding points on adjacent slices with a series of straight lines, orientated coronally, to give a distorted rectangular mesh. The regions corresponding to the striate cortex, upper and lower V3 and V5 were identified and the co-ordinates labelled, based on anatomical and PET evidence (Stensaas et al., 1974; Horton and Hoyle, 1991; Watson et al., 1993; Clarke, 1994; Shipp et al., 1995): striate cortex, 23 cmvhemisphere: upper and lower V3. 13 cnr/hemisphere: V5. 4 cm 2 /hemisphere. In each area, a series of unit current dipoles were placed normal to the local surface, calculated from the two adjacent sides of the mesh, at each transverse/coronal intersection. Four hundred and and response properties of neurons in human temporal lobe and visual pathways. Brain 1983; 106: Yoneda K, Sotaro S, Masato Y, Morihiro S. The early component of the visual evoked magnetic field. Neuroreport 1995; 6: Yukie M, Iwai E. Direct projection from the dorsal lateral geniculate nucleus to the prestriate cortex in macaque monkeys. J Comp Neurol 1981; 201: Zeki SM. Representation of central visual fields in prestriate cortex of monkey. Brain Res 1969; 14: Zeki SM. Convergent input from the striate cortex (area 17) to the cortex of the superior temporal sulcus in the rhesus monkey. Brain Res 1971; 28: Zeki SM. The mosaic organization of the visual cortex in the monkey. In Bellairs R, Gray EG, editors. Essays on the nervous system, a festschrift for Professor J. Z. Young. Oxford: Clarendon Press, 1974a: 327^3. Zeki SM. Functional organization of a visual area in the posterior bank of the superior temporal sulcus of the rhesus monkey. J Physiol (Lond) 1974b; 236: Zeki SM. Functional specialisation in the visual cortex of the rhesus monkey. Nature 1978; 274: Zeki S. The motion pathways of the visual cortex. Blakemore C, editor. Vision: coding and efficiency. Cambridge: Cambridge University Press, 1990: Zeki S. A vision of the brain Oxford: Blackwell Scientific, Zeki S, Shipp S. The functional logic of cortical connections. [Review]. Nature 1988; 335: Received January 26, Revised April 4, Second revision August 16, Accepted September 18, 1995 forty-eight dipoles were distributed between the areas in each hemisphere. The electric potential arising from each dipole was calculated using an adaptation of the three sphere model described by Kavanagh et al. (1978). A best fitting sphere of radius 7 cm, with centre at Talairach co-ordinate y = -3.6 cm, x = 0 cm, z = 0 cm represented the brain surface. The skull and scalp were modelled by concentric spheres with radii 7.7 and 8.1 cm, respectively. The conductivities of brain tissue, skull and scalp were taken in the ratios I, and 1, respectively. The calculated potential of each current dipole was multiplied by the area of the mesh at each point in order to eliminate the effects of undersampling. The electrical potential at 12X12 grid points referenced to Fz were calculated using 25 terms in the series expansion (we estimate that this provides an accuracy of >5% with respect to the homogeneous sphere in which there is a closed analytic expression for the potential). The magnetic flux was calculated using the approximate closed form expression for the magnetic vector potential (Guy, 1990). The resulting fields from dipoles in different regions, or parts of the regions, were summed to produce the modelled data presented in Figs 2-4. Evenly spaced contours were produced by linear interpolation between grid points.